Compare commits

...
Author SHA1 Message Date
shadowdaoandClaude Opus 5.5 05ed1c37a5 fix(aws): phase 7 parked minors
Build App (Preview) / compute-version (pull_request) Successful in 9s
Secret Scan / scan (push) Successful in 21s
Secret Scan / scan (pull_request) Successful in 9s
Build App (Preview) / create-release (pull_request) Successful in 2s
Build Container / build-container (pull_request) Successful in 2m59s
Build App (Preview) / build-macos (pull_request) Successful in 6m17s
Build App (Preview) / build-windows (pull_request) Successful in 12m0s
Build App (Preview) / build-linux (pull_request) Successful in 21m5s
Build App (Preview) / prune-previews (pull_request) Successful in 4s
Build App (Preview) / test (pull_request) Successful in 24m1s
Interrupted-start notice gets its own toast topic and headline; the launch-type switch note only shows after a successful save and is dropped by later saves; README states the CreateLogStream scoping limit.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-17 05:24:56 -08:00
shadowdaoandClaude Opus 5.5 70cfa6b757 test(aws): Fargate logs under the narrowed scope
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 19:01:28 -08:00
shadowdaoandClaude Opus 5.5 6fa7ef5605 docs(aws): record phase 7 live run 7 (log scoping)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:44:47 -08:00
shadowdaoandClaude Opus 5.5 d85292ea00 fix(aws): add the EC2 root volume to the cost estimate; name public IPv4
The EC2 estimate adds the 50 GiB gp3 root volume at the region's gp3
GB-month price (Pricing API, cached per region like the instance price;
fixture recorded 2026-11-16). The note says whether it is included and
that a public-mode environment's IPv4 address adds $0.005/hour.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:40:12 -08:00
shadowdaoandClaude Opus 5.5 1d1e356fcd fix(aws): EC2 host settings polish
HostSection says what a launch-type switch changed besides the type,
checks the saved reservations against the instance type once on load,
and is re-indented. The EC2 capacity cache keeps none/unusable for one
minute only. RuntimeSection's VPN hint on EC2 speaks of an ECS task on
the user's instance. Joining over an incomplete onboarding with no
record warns that None removes half-built EC2 capacity, and starts from
the capacity last chosen in this run when there is one.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:37:07 -08:00
shadowdaoandClaude Opus 5.5 77db6028d0 fix(aws): write logs only into the user's own streams; no log rights on the instance role
The instance role (reachable from a Docker-access project's host-network
container) could PutLogEvents into any user's streams. It now has no log
rights: awslogs writes with the execution role
(ECS_ENABLE_AWSLOGS_EXECUTIONROLE_OVERRIDE=true). The task and execution
roles create streams in the environment's group but write only into
log-stream:<owner_key>/* (every awslogs stream's prefix); the Fargate
fingerprint moves deliberately. The live test asserts task logs are
written. README: per-user logs, orphan clock skew, verdict wording;
boundary version comments no longer mention Session Manager.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:33:57 -08:00
shadowdaoandClaude Opus 5.5 8fe0810901 fix(aws): never let the user record vouch for EC2 names tagged as another owner's
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:30:41 -08:00
shadowdaoandClaude Opus 5.5 8c9ffd4c4c fix(aws): never attach to an exec-off EC2 task after a restart; widen the orphan margin
Reconcile stops a placed or orphaned exec-off task (REPLACED_REASON,
checked; the ARN is cleared only once the stop succeeded) and marks a
recent one still waiting Error with its ARN kept and an 'interrupted'
warning. Reattach refuses an exec-off task. ORPHAN_MARGIN is 10 minutes
(clock skew). A failed EC2 record read no longer blocks attaching to a
live task. One helper builds the task cluster list.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:30:05 -08:00
shadowdaoandClaude Sonnet 5.5 719a75b46d docs(aws): phase 7
Spec: phase 7 done, rulings (E1-E11 as built, R1-R4, cold-start rerun),
data model, feature table. CLAUDE.md: EC2 rules and the docker-socket
step; sandbox is off on all AWS. aws/README.md: triple-c-user admin
setup and caveats, measured sizes.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
2026-11-16 18:19:39 -08:00
shadowdaoandClaude Opus 5.5 6e24977cf0 fix(aws): replace an orphaned exec-off EC2 task instead of refusing for ever
A found task with ECS Exec off that still waits for an instance is
another start's only while one could still be waiting for it (created
within the EC2 placement wait plus 2 minutes); an older one is a crashed
start's orphan and is replaced (stopped with its own reason, checked,
waited for). INSTANCE_RACE now mentions Stop. The rerun's race records
the winner's task, as the first launch's race does; a first task another
start replaced says so. TaskInfo gains created_unix.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:16:22 -08:00
shadowdaoandClaude Opus 5.5 374fb5a35d test(aws): harden the phase 7 live test; record the update probe
The guard paginates Auto Scaling groups and treats a failed read as not
yet gone; the test asserts exactly one live task after a cold start,
the sandbox off in the task's settings and bwrap failing (E8). README:
updating a group to an unsupported template version is denied by the
launch check (live probe); run 6 after the review fixes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 18:12:51 -08:00
shadowdaoandClaude Opus 5.5 10f473f3da fix(aws): show the sandbox as off wherever an AWS project can't run it
PermissionModeControl read the raw sandbox_mode_enabled, so an AWS
project showed "Sandbox ON — bubblewrap isolation" and no Bypass
caution although the host forces it off. A shared sandboxEffective()
(false for any AWS project) now feeds both it and the Runtime toggle.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 17:59:44 -08:00
shadowdaoandClaude Opus 5.5 c25761789e fix(aws): never run a second task for a stopped or contested cold EC2 start
The cold-start rerun now runs again only once an instance has taken the
first task and it is still meant to run; a first task that stops by
itself (a user's Stop, a failed placement) ends the start with its
reason. The second launch goes through find_or_run, so a concurrent
start's task is reported as START_RACE instead of doubled. A found task
with ECS Exec off that is still waiting for an instance belongs to a
start in progress (INSTANCE_RACE); a placed one is stopped, the stop is
checked and waited for (its own 3-minute budget) before a new run.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 17:59:09 -08:00
shadowdaoandClaude Opus 5.5 fc3456c44e test(aws): live phase 7 EC2 checks
a_phase7_ec2_workspace_round_trips: onboarding with per-user EC2
capacity through prepare_onboarding (tagged workspace role, all four
admin files), its collision check and refusal, admin-tier scaling
probes, a cold and a warm EC2 start (Docker access, VPN, ECS Exec into
the bridge task), in-container checks (docker ps, wg/tun,
src_valid_mark, bwrap, IMDS, introspection), RunTask on another user's
cluster, stop and scale-in, removal, re-apply without EC2 with an
instance up, offboard and destroy. The guard (ec2_cleanup) force-deletes
any per-user EC2 capacity. Results in aws/README.md and the spec.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 17:52:36 -08:00
shadowdaoandClaude Opus 5.5 6c7c2ea4ef fix(aws): keep Claude Code's sandbox off on EC2 too (E8)
Measured live on the ECS-optimized AL2023 AMI (kernel 6.1,
max_user_namespaces 30937): bwrap --unshare-user in the ECS task fails
with "No permissions to create new namespace", as on Fargate. Docker's
default seccomp profile refuses it without CAP_SYS_ADMIN. sandbox_allowed
is false for every AWS project, and the Runtime toggle is locked off on
EC2 as on Fargate.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 16:35:30 -08:00
shadowdaoandClaude Opus 5.5 43fcc520ab fix(aws): run a cold EC2 task again once its instance is up
ECS turns enableExecuteCommand off on a RunTask that has to wait for an
instance (PROVISIONING behind a capacity provider), and the task keeps it
off once placed: verified live (CloudTrail shows true in the request,
false in the answer; a RunTask with an instance up keeps it). Such a
task can never be reached, so every cold EC2 start timed out.

A cold start now lets that first task bring the instance up, stops it as
soon as it is placed, waits for it to stop and runs the task again onto
the instance; a second task without ECS Exec is stopped and reported. A
live task found with ECS Exec off is stopped and replaced.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 16:34:38 -08:00
shadowdaoandClaude Opus 5.5 bf5a810e0d fix(aws): allow template-only Auto Scaling group creates
A CreateAutoScalingGroup that names only a launch template sends no
autoscaling:InstanceTypes (verified live), so the scaling file's
Null:false on it denied every per-user group. Drop it; the type is held
by the caller's own ec2:RunInstances launch check on the template, which
also refuses MixedInstancesPolicy overrides of unsupported types.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 15:24:55 -08:00
shadowdaoandClaude Opus 5.5 9e1a63cde5 feat(aws): EC2 projects in the UI, Docker access and VPN
Host settings get the launch-type control (Fargate | EC2 with the user's
instance type, disabled with the reason when their onboarding has no EC2
capacity), switching through launchTypePatch, which now also clears the
gateway sidecar and Spot for EC2 and snaps sizes per side. EC2 projects size
CPU/memory reservations checked by Rust's fit rule (ec2_fit_problem, new
command) against the user's record (aws_ec2_capacity, new command); no
ephemeral size, Spot or architecture; notes for scale-to-zero, Docker access
as root on the instance and no sidecar. The sandbox follows the project on EC2.

The cost estimate prices an EC2 project as its instance type's AmazonEC2
on-demand Linux shared-tenancy price (recorded fixture), cached per region and
type with the Fargate cache's rules. The entrypoint's Docker-socket group step
now runs in aws mode when the socket exists; the VPN skill follows
VPN_SUPPORT_ENABLED there as on Docker.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 15:03:57 -08:00
shadowdaoandClaude Opus 5.5 d6ef6a5fd2 feat(aws): run projects on per-user EC2 capacity
EC2 projects register an EC2/bridge task definition (no runtime platform or
ephemeral storage; task-level reservations checked against the user's
instance type with a 10% memory margin), with the instance's Docker socket
and the Docker VPN pieces (NET_ADMIN, /dev/net/tun, src_valid_mark) when
enabled, and run on the user's cluster through their capacity provider.
The start reads the user record, waits up to 10 minutes with "Starting an
instance" progress, and maps RunTask access denied to the triple-c-user
mismatch. Stop, reconcile, removal and the watch follow the task's own
cluster, so a running EC2 task outlives its record. The sandbox stays on
for EC2; the gateway sidecar is refused there for now.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 14:46:06 -08:00
shadowdaoandClaude Opus 5.5 98e330060e fix(aws): EC2 name check by owner tag; keep unusable records; bundle warning
The pre-apply check now decides ownership by the resources' triple-c:owner
tag (record as fallback), so a failed apply of one's own capacity can be
re-applied; any cluster status but INACTIVE holds the name. An onboarding
seam (OnboardApi) makes the record read and the name lookups testable, and
prepare_onboarding settles everything before launch. With no choice given,
an unreadable or invalid record is refused instead of removing capacity.
Bundles warn when their cluster name looks taken; the README check prints
owner tags. The UI keeps the recorded type when the instance list fails and
says when an apply removes EC2 capacity.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 14:31:30 -08:00
shadowdaoandClaude Opus 5.5 1a2c530be8 feat(aws): choose EC2 capacity when onboarding
Self-onboarding and the onboarding bundle take an EC2 capacity choice:
none, or an instance type from the app's table (grouped x86/ARM, default
m7i.large), 1-4 instances and the user's triple-c-user value. A re-apply
keeps the recorded capacity unless changed (the record gains
max_instances). Before any self-onboarding apply with EC2 the app refuses
per-user names (ACTIVE cluster or Auto Scaling group) that this owner's
record doesn't name (R3). OnboardingState carries the user's compute from
their record; the panel shows it. The bundle README's collision check now
comes before the apply step.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 14:22:31 -08:00
shadowdaoandClaude Opus 5.5 e9c20287b7 fix(aws): EC2 name-collision step in onboarding bundles; bridge agent config
Task 3 fix round 1 (rulings R2, R3):
- Onboarding admin bundles' README gains an "EC2 capacity" step: before
  applying with ec2_instance_type, check with describe-clusters /
  describe-auto-scaling-groups that triple-c-<env>--u-<tag> is unused
  unless this owner's /triple-c/users/<env>/<owner key> record names it
  (exact names; <tag> when the bundle has none). Tested.
- User data for bridge tasks (R2): ECS_ENABLE_TASK_IAM_ROLE=true instead
  of ECS_AWSVPC_BLOCK_IMDS; IMDSv2 hop limit 1 kept, comment corrected;
  the test asserts no ECS_ENABLE_TASK_ENI / awsvpc settings.
- Reverted unrelated rustfmt churn in tofu_modules/tests.rs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 14:07:28 -08:00
shadowdaoandClaude Opus 5.5 9410c098ed feat(aws): optional per-user EC2 capacity in the onboarding module
Phase 7 Task 3 (E1-E4, E11; onboarding module version 2):
- ec2.tf: with ec2_instance_type set, the user's own cluster
  triple-c-<env>--u-<ec2_user_tag> with exactly one capacity provider
  (managed termination protection, managed scaling 100% one instance at
  a time, 60 s warm-up), an Auto Scaling group of 0..ec2_max_instances
  (desired left to ECS, protect_from_scale_in, force_delete off so the
  provider drains before deleting, fixed launch-template version, default
  tags + AmazonECSManaged), a launch template on the ECS-optimized AL2023
  AMI by architecture (IMDSv2 hop 1, 50 GiB gp3 encrypted root, the
  environment's tasks group, public IP as the environment's tasks get
  one, ECS_CLUSTER + ECS_AWSVPC_BLOCK_IMDS), the instance role
  triple-c-<env>-<owner key>-instance (bounded, tagged triple-c:cluster,
  ECS agent on its cluster only, ECR pull, logs; no SSM core) and its
  instance profile, and the user record /triple-c/users/<env>/<owner key>.
  Empty type (the default): count 0 everywhere, Fargate plan unchanged.
- Variables ec2_instance_type (the app's 30 types), ec2_max_instances
  (1-4), ec2_user_tag (as validate_ec2_user_tag, required with a type);
  onboarding_vars takes Option<&Ec2Compute> and validates the same.
- The environment record's assign_public_ip is now format-checked.
- tofu_modules.rs: ONBOARD_MODULE_VERSION 2, user_record_parameter_name,
  USER_RECORD_FIELDS, Ec2Compute, EC2_MAX_INSTANCES, ec2.tf embedded.
- Tests: EC2 calls merged into MODULE_CALLS under real types
  (all_module_calls removed); gating + version-1 resource fingerprint;
  launch template, group, capacity provider, instance role, AMI by arch,
  record fields, names, variable rules; the instance role inside the
  boundary with its cluster principal tag.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 14:00:23 -08:00
shadowdaoandClaude Opus 5.5 9829ff0c43 feat(aws): per-user EC2 clusters keyed by the triple-c-user principal tag
Fix round 2 for the phase 7 policies (amends plan E1):
- Per-user EC2 names: triple-c-<env>--u-<user tag> for the cluster,
  capacity provider, Auto Scaling group and launch template
  (user_ec2_stem, validate_ec2_user_tag; `--` can't occur in environment
  names or user tags, so the patterns are unambiguous); the instance
  role/profile stay owner-key named (user_instance_role_name, <= 64).
- workspace-user-tasks.json: RunTask/StartTask on a triple-c-*--u-*
  cluster is denied unless it is the caller's own
  (${aws:PrincipalTag/triple-c-user}), and denied without the tag.
- Admin: capacity providers and groups narrowed to triple-c-*--u-*; the
  boundary's agent cluster tag must match triple-c-*--u-*.
- Instance-type fail-open closed: ec2:InstanceType required (StringEquals,
  volumes in their own statement); group create requires
  autoscaling:InstanceTypes present.
- The compute file is split: environment-admin-scaling.json (groups,
  instance profiles, service-linked roles) is the admin tier's fourth file.
- Evaluator learns ArnNotLike. README: triple-c-user setup, sizes,
  still-open probes.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 13:44:55 -08:00
shadowdaoandClaude Opus 5.5 e6fbf4e133 fix(aws): pin, type-limit and fence per-user EC2 capacity permissions
Review fix round 1 for the phase 7 admin compute file and boundary:
- Auto Scaling groups: create needs autoscaling:LaunchTemplateVersionSpecified
  (update BoolIfExists), no target groups / load balancers, supported
  instance types; CreateOrUpdateTags only on already-tagged groups.
  ModifyLaunchTemplate dropped (the default version never moves).
- RunInstances: instance and volume need the component request tag,
  the instance a supported type (the app's ec2_instance_spec table,
  pinned by a test); the security group must be the template's own and
  component-tagged, the image Amazon's and the template's.
- Boundary: the ECS agent only on the cluster its role's
  triple-c:cluster principal tag names (must be triple-c-*); Session
  Manager / ec2messages dropped (plan E4 amended).
- Evaluator learns Bool/BoolIfExists; the simulator sends the Boolean
  keys typed Boolean.
- README: instance types, pinning, -instance suffix is not an admin
  fence, RunInstances wording, still-open probes, IMDS/Docker note.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 13:17:54 -08:00
shadowdaoandClaude Opus 5.5 7393143762 feat(aws): admin and boundary permissions for per-user EC2 capacity
Phase 7 Task 2 (E4, E5, E11):
- New third admin file environment-admin-compute.json (3,705 chars):
  capacity providers (triple-c-*), component-tagged launch templates,
  triple-c-* Auto Scaling groups (request tag to create, resource tag to
  change/protect/delete), RunInstances only from a tagged launch template
  with Amazon images, tags only at create, instance profiles (triple-c-*),
  PassRole of triple-c-*-instance roles to ec2.amazonaws.com only,
  service-linked roles for Auto Scaling and ECS only, the public
  ECS-optimized AMI parameter, autoscaling/capacity-provider reads.
- Boundary v2 (2,625 chars): the ECS agent's calls on triple-c-* clusters
  and container instances, TagResource only at RegisterContainerInstance,
  DiscoverPollEndpoint and Session Manager core. BOUNDARY_MODULE_VERSION=2
  with a template_version local and default tag, pinned by a test.
- Checks for every new action; tests for tag/name scoping, launch-only-
  through-templates, PassRole to EC2 only for instance roles, the
  workspace tier changing no compute, the boundary's agent fence.
- EC2_COMPUTE_CALLS in tofu_modules/tests.rs: the onboarding module's
  expected calls, evaluated against the admin tier and in the write drift
  test until Task 3 moves them into MODULE_CALLS.
- README: third file, sizes, boundary re-apply, resource table rows,
  scoping notes and "still open" points for the live run.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 13:07:37 -08:00
shadowdaoandClaude Opus 5.5 c6b46a7347 fix(aws): refuse Docker/VPN on Fargate without grandfathering; launchTypePatch
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 12:56:18 -08:00
shadowdaoandClaude Opus 5.5 86f253b7da feat(aws): EC2 launch type setting
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 12:54:24 -08:00
shadowdaoandClaude Opus 5.5 edef77feed docs(aws): phase 7 plan — ECS on EC2
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 12:46:40 -08:00
jknapp 1659788b70 Merge pull request 'AWS host phase 6: EFS, copy-at-connect, setup scripts, lifetime cap, ports, gateway sidecar, cost estimate' (#83) from feat/aws-phase6 into main
Secret Scan / scan (push) Successful in 22s
Build Container / build-container (push) Successful in 4m51s
2026-11-16 20:36:36 +00:00
78 changed files with 11231 additions and 546 deletions

No files matched your search

+64 -3
View File
@@ -292,7 +292,7 @@ docker exec stdout → tokio task → emit("terminal-output-{sessionId}") → li
when it drops, so an `AgentEndpoint` stays valid. The `StartSession` token goes to the plugin
through its environment, never argv, and the plugin's environment is stripped of every
inherited `AWS_*` variable before the session's own credentials are added.
`aws_host.rs` is `ProjectHost::Aws` (an ECS Fargate task per project; a running task is
`aws_host.rs` is `ProjectHost::Aws` (an ECS task per project, Fargate or, on the user's own instance, EC2; a running task is
attached to with the token read back from its secret, never rotated; a task a failed start ran
is stopped again) and `tunnels.rs` the registry of open tunnels by project — an AWS project's
`ContainerTarget::for_project` is its tunnel's endpoint, its stored `container_id` is that
@@ -367,8 +367,11 @@ docker exec stdout → tokio task → emit("terminal-output-{sessionId}") → li
push's snapshot), may push first (under an in-container `timeout`, since the agent does not kill an
exec the client gave up on), then writes `pushed` or `discard` to `/run/triple-c/stop-reason`
(root-owned; the entrypoint's TERM hook reads it, and the idle watcher writes `idle` or `lifetime`) so the hook
doesn't push twice; the marker is cleared if StopTask fails. Claude Code's sandbox is forced off in
the generated settings (bubblewrap can't create namespaces on Fargate, verified live).
doesn't push twice; the marker is cleared if StopTask fails. **Claude Code's sandbox is forced off
on every AWS project, Fargate and EC2** in the generated settings (`sandbox_allowed` is false; the
Runtime toggle is locked off, `sandboxEffective` in `awsWorkspace.ts`): bubblewrap fails on Fargate
and under Docker's default seccomp profile on EC2 too ("No permissions to create new namespace",
verified live), so a `sandbox_allowed` that said yes would only break every command.
**LLM gateway sidecar** (`gateway_sidecar`, Q7): a second, `essential: false` container `gateway`
(`docker::gateway::registry_image_ref`, digest-pinned, amd64+arm64; `memoryReservation` 512; no
`environment`, no `dependsOn`) whose `secrets` are exactly `TRIPLEC_GATEWAY_{CONFIG,API_KEY,MASTER_KEY}`
@@ -384,6 +387,60 @@ docker exec stdout → tokio task → emit("terminal-output-{sessionId}") → li
`FILTERS`) → per-region x86/ARM rates cached at `<data_dir>/triple-c/aws-pricing/<region>.json` for
24 h; `aws_cost_estimate(connection_id, region, config)` never errors — denied/offline/slow (8 s) is
`None` ("Estimate unavailable"), a failed region isn't re-asked for 5 min, a stale cache still answers.
EC2 projects (E10) pass `instance_type` (the user's, from `aws_ec2_capacity`): the `AmazonEC2`
on-demand Linux/shared-tenancy hourly price (`EC2_FILTERS`), cached as `<region>--ec2-<type>.json`
with the same rules — the instance's price, not the reservations — plus the 50 GiB gp3 root volume
(`GP3_FILTERS`, `<region>--ebs-gp3.json`, best effort: without it the note says it's missing;
`EC2_ROOT_VOLUME_GIB` is pinned to `ec2.tf`). A public IPv4 address ($0.005/h) is only in the note.
**ECS on EC2 (phase 7; `aws/terraform/user-onboard/ec2.tf`, `workspace.rs` `ec2_plan`, `aws_host.rs`).**
Opt-in per onboarding (`ec2_instance_type`); a project's `launch_type` is `Fargate | Ec2`. Why the
rules are what they are:
- **Per-user cluster `triple-c-<env>--u-<tag>`**, named by the principal tag `triple-c-user`
(`user_ec2_stem` in `tofu_modules.rs`, the only place). One capacity provider per cluster,
because `PutClusterCapacityProviders` replaces the list and two modules would undo each other on
a shared cluster. Environment names and tags never contain `--`, so `triple-c-*--u-*` matches
exactly the per-user clusters. **Two Denies in `workspace-user-tasks.json`** (`RunTask`/`StartTask`
on a `--u-*` cluster other than the caller's own, and on any when the caller has **no**
`triple-c-user` principal tag) are the whole isolation between EC2 users: don't loosen them, and
keep the missing-tag Deny (a variable that resolves to nothing must not fall through). A denial
maps to `EC2_TAG_MISMATCH` ("different triple-c-user value").
- **The instance role is scoped by its own cluster tag.** `triple-c-<env>-<owner_key>-instance`
(owner key, not user tag: role names are account-global) has only the ECS agent's calls, and the
**boundary** allows them on `cluster/${aws:PrincipalTag/triple-c:cluster}` (must match `--u-*`).
A Docker-access project is root on its instance and holds that role, so the tag is what stops it
registering the instance into someone else's cluster. No SSM core and **no log rights** on the
instance (awslogs writes with the execution role: `ECS_ENABLE_AWSLOGS_EXECUTIONROLE_OVERRIDE`).
The task and execution roles `PutLogEvents` only into `log-stream:<owner_key>/*` of the shared
log group, so `log_stream_prefix` must keep starting with the owner key. The boundary
is versioned (2): an account admin re-applies it once before anyone onboards with EC2.
- **Launch templates and groups are pinned to a numeric template version, and the admin policy
denies `$Latest`/`$Default`**: the admin's `iam:PassRole` and `ec2:RunInstances` checks look at the
template when it is created, and a version edited afterwards would carry any instance type or
role (a PassRole bypass). The type list (`ec2_instance_spec` in `models/project.rs`) lives in the
model table, the HCL variable and the policy (`ec2:InstanceType` in the compute file,
`autoscaling:InstanceTypes` twice in the scaling file); change all together. A template-only group
create sends no `autoscaling:InstanceTypes`, so don't add a `Null` condition on it (live-verified).
- **Bridge networking, not awsvpc (R2):** the instance is the user's alone, awsvpc tasks on EC2 get
no public IP, and bridge keeps IMDS out of reach with hop limit 1 (the agent on the host network
still reaches it). The ECS agent's introspection port (172.17.0.1:51678) stays reachable;
accepted. The gateway sidecar is refused on EC2 (no shared loopback).
- **Cold-start rerun + orphan rule.** ECS turns ECS Exec **off** for a task that has to wait for an
instance (the agent tunnel needs it), so the first RunTask only brings the instance up; once
placed, the app stops it, waits for STOPPED and runs again via `find_or_run` (so a concurrent
start never doubles a task: `START_RACE`), requiring exec on (`EXEC_OFF`). A found exec-off task
that is still waiting for an instance is another start's (`INSTANCE_RACE`) while younger than
`ec2_wait + ORPHAN_MARGIN`; older is an orphan, stopped (`REPLACED_REASON`) and replaced. Keep
both launches going through `find_or_run`.
- **Docker access = the instance's own socket**, mounted at `/var/run/docker.sock` (root on the
user's own instance, acceptable only because nobody else's task can land there); VPN =
`NET_ADMIN` + `/dev/net/tun` + `src_valid_mark`. A Fargate project can't keep either: switching
to Fargate clears both flags (`launchTypePatch`, R1).
- **Collision check before an apply** (`ec2_collision_refusal`, `prepare_onboarding`): the module
would adopt another user's cluster of the same name, so the app refuses a name owned by someone
else (resources' `triple-c:owner` tag, the user record `/triple-c/users/<env>/<owner_key>` as
fallback). Admin bundles carry the same check in `README.txt`.
- Scale to zero (min 0): the first start after ~17 min idle is cold (~3.5-4 min), warm ~20-30 s.
Re-applying onboarding without EC2 drains and deletes the group even with an instance running.
**Environments are OpenTofu modules** (`aws/terraform/{environment,user-onboard,permissions-boundary}`;
spec D2–D10). `tofu_modules.rs` embeds them and is the **only** place the shared names are built
(environment, `/triple-c/environments/<env>` record, owner key — `sha256("<region>:<aws:userid>")`,
@@ -505,6 +562,10 @@ docker exec stdout → tokio task → emit("terminal-output-{sessionId}") → li
The re-clone closes the apply lock's fd (a later connect never waits for it); `--prepare` takes it.
The functions both scripts use live in `container/entrypoint-lib.sh`
(`/usr/local/lib/triple-c/`), sourced by the entrypoint first; it only defines things.
**The Docker-socket group step** (`# >>> docker_socket`) runs in aws mode as well as Docker mode: an
EC2 project with Docker access mounts the instance's `/var/run/docker.sock`, and the step adds
`claude` to a group with the socket's gid (`docker-host` if none exists). Fargate never has the
socket, so it is inert there.
**The `TRIPLEC_` env prefix is reserved** on both hosts (`RESERVED_ENV_PREFIXES`; `workspace.rs`
drops custom variables in it): the entrypoint's settings live there.
- **Browser runtime libraries are baked in; browser *binaries* are not.** A layer runs
+27
View File
@@ -367,6 +367,32 @@ dependencies = [
"uuid",
]
[[package]]
name = "aws-sdk-autoscaling"
version = "1.144.0"
source = "registry+https://github.com/rust-lang/crates.io-index"
checksum = "a08cb197f98dad7b1a3d4e7fcc94fbc13a66c2ea5465ba57fdb5dd00f865d6e8"
dependencies = [
"arc-swap",
"aws-credential-types",
"aws-runtime",
"aws-smithy-async",
"aws-smithy-http",
"aws-smithy-json",
"aws-smithy-observability",
"aws-smithy-query",
"aws-smithy-runtime",
"aws-smithy-runtime-api",
"aws-smithy-schema",
"aws-smithy-types",
"aws-smithy-xml",
"aws-types",
"fastrand",
"http",
"regex-lite",
"tracing",
]
[[package]]
name = "aws-sdk-cloudwatchlogs"
version = "1.165.0"
@@ -7466,6 +7492,7 @@ dependencies = [
"aws-config",
"aws-credential-types",
"aws-runtime",
"aws-sdk-autoscaling",
"aws-sdk-cloudwatchlogs",
"aws-sdk-ecs",
"aws-sdk-efs",
+3
View File
@@ -80,6 +80,9 @@ aws-runtime = { version = "1.10.4", default-features = false }
# The default sleep and time source for the hand-built SdkConfig (no env/profile
# loading); the version and feature already in the tree.
aws-smithy-async = { version = "1.3.1", default-features = false, features = ["rt-tokio"] }
# commands/environment_commands.rs: the pre-apply check that a user's EC2 names
# (Auto Scaling group) are not another onboarding's (phase 7, R3).
aws-sdk-autoscaling = { version = "1.144.0", default-features = false, features = ["default-https-client", "rt-tokio", "behavior-version-latest"] }
aws-sdk-s3 = { version = "1.159.0", default-features = false, features = ["default-https-client", "rt-tokio", "behavior-version-latest"] }
# aws/tofu.rs: `setsid` puts the detached OpenTofu runner in its own session,
+3
View File
@@ -22,6 +22,8 @@
"allow-aws-task-log",
"allow-aws-workspace-status",
"allow-aws-environment-max-hours",
"allow-aws-ec2-capacity",
"allow-ec2-fit-problem",
"allow-remove-aws-resources",
"allow-rebuild-project-container",
"allow-reconcile-project-statuses",
@@ -82,6 +84,7 @@
"allow-cancel-aws-deployment",
"allow-push-aws-saved-state",
"allow-onboard-aws-user",
"allow-list-ec2-instance-types",
"allow-deploy-aws-permissions-boundary",
"allow-save-aws-admin-bundle",
"allow-detect-host-timezone",
+1 -1
View File
@@ -1 +1 @@
{"__app-acl__":{"default_permission":null,"permissions":{"allow-acquire-claude-token":{"identifier":"allow-acquire-claude-token","description":"Enables the acquire_claude_token command without any pre-configured scope.","commands":{"allow":["acquire_claude_token"],"deny":[]}},"allow-add-aws-keys-connection":{"identifier":"allow-add-aws-keys-connection","description":"Enables the add_aws_keys_connection command without any pre-configured scope.","commands":{"allow":["add_aws_keys_connection"],"deny":[]}},"allow-add-aws-profile-connection":{"identifier":"allow-add-aws-profile-connection","description":"Enables the add_aws_profile_connection command without any pre-configured scope.","commands":{"allow":["add_aws_profile_connection"],"deny":[]}},"allow-add-marketplace":{"identifier":"allow-add-marketplace","description":"Enables the add_marketplace command without any pre-configured scope.","commands":{"allow":["add_marketplace"],"deny":[]}},"allow-add-marketplace-gh-host-account":{"identifier":"allow-add-marketplace-gh-host-account","description":"Enables the add_marketplace_gh_host_account command without any pre-configured scope.","commands":{"allow":["add_marketplace_gh_host_account"],"deny":[]}},"allow-add-marketplace-token-account":{"identifier":"allow-add-marketplace-token-account","description":"Enables the add_marketplace_token_account command without any pre-configured scope.","commands":{"allow":["add_marketplace_token_account"],"deny":[]}},"allow-add-project":{"identifier":"allow-add-project","description":"Enables the add_project command without any pre-configured scope.","commands":{"allow":["add_project"],"deny":[]}},"allow-add-scheduled-task":{"identifier":"allow-add-scheduled-task","description":"Enables the add_scheduled_task command without any pre-configured scope.","commands":{"allow":["add_scheduled_task"],"deny":[]}},"allow-apply-marketplace-now":{"identifier":"allow-apply-marketplace-now","description":"Enables the apply_marketplace_now command without any pre-configured scope.","commands":{"allow":["apply_marketplace_now"],"deny":[]}},"allow-apply-settings-import":{"identifier":"allow-apply-settings-import","description":"Enables the apply_settings_import command without any pre-configured scope.","commands":{"allow":["apply_settings_import"],"deny":[]}},"allow-aws-connection-identity":{"identifier":"allow-aws-connection-identity","description":"Enables the aws_connection_identity command without any pre-configured scope.","commands":{"allow":["aws_connection_identity"],"deny":[]}},"allow-aws-connection-sso-login":{"identifier":"allow-aws-connection-sso-login","description":"Enables the aws_connection_sso_login command without any pre-configured scope.","commands":{"allow":["aws_connection_sso_login"],"deny":[]}},"allow-aws-cost-estimate":{"identifier":"allow-aws-cost-estimate","description":"Enables the aws_cost_estimate command without any pre-configured scope.","commands":{"allow":["aws_cost_estimate"],"deny":[]}},"allow-aws-environment-max-hours":{"identifier":"allow-aws-environment-max-hours","description":"Enables the aws_environment_max_hours command without any pre-configured scope.","commands":{"allow":["aws_environment_max_hours"],"deny":[]}},"allow-aws-policy-document":{"identifier":"allow-aws-policy-document","description":"Enables the aws_policy_document command without any pre-configured scope.","commands":{"allow":["aws_policy_document"],"deny":[]}},"allow-aws-sso-refresh":{"identifier":"allow-aws-sso-refresh","description":"Enables the aws_sso_refresh command without any pre-configured scope.","commands":{"allow":["aws_sso_refresh"],"deny":[]}},"allow-aws-task-log":{"identifier":"allow-aws-task-log","description":"Enables the aws_task_log command without any pre-configured scope.","commands":{"allow":["aws_task_log"],"deny":[]}},"allow-aws-workspace-status":{"identifier":"allow-aws-workspace-status","description":"Enables the aws_workspace_status command without any pre-configured scope.","commands":{"allow":["aws_workspace_status"],"deny":[]}},"allow-build-gateway-image":{"identifier":"allow-build-gateway-image","description":"Enables the build_gateway_image command without any pre-configured scope.","commands":{"allow":["build_gateway_image"],"deny":[]}},"allow-build-image":{"identifier":"allow-build-image","description":"Enables the build_image command without any pre-configured scope.","commands":{"allow":["build_image"],"deny":[]}},"allow-build-stt-image":{"identifier":"allow-build-stt-image","description":"Enables the build_stt_image command without any pre-configured scope.","commands":{"allow":["build_stt_image"],"deny":[]}},"allow-cancel-aws-deployment":{"identifier":"allow-cancel-aws-deployment","description":"Enables the cancel_aws_deployment command without any pre-configured scope.","commands":{"allow":["cancel_aws_deployment"],"deny":[]}},"allow-cancel-claude-token":{"identifier":"allow-cancel-claude-token","description":"Enables the cancel_claude_token cLine truncated
{"__app-acl__":{"default_permission":null,"permissions":{"allow-acquire-claude-token":{"identifier":"allow-acquire-claude-token","description":"Enables the acquire_claude_token command without any pre-configured scope.","commands":{"allow":["acquire_claude_token"],"deny":[]}},"allow-add-aws-keys-connection":{"identifier":"allow-add-aws-keys-connection","description":"Enables the add_aws_keys_connection command without any pre-configured scope.","commands":{"allow":["add_aws_keys_connection"],"deny":[]}},"allow-add-aws-profile-connection":{"identifier":"allow-add-aws-profile-connection","description":"Enables the add_aws_profile_connection command without any pre-configured scope.","commands":{"allow":["add_aws_profile_connection"],"deny":[]}},"allow-add-marketplace":{"identifier":"allow-add-marketplace","description":"Enables the add_marketplace command without any pre-configured scope.","commands":{"allow":["add_marketplace"],"deny":[]}},"allow-add-marketplace-gh-host-account":{"identifier":"allow-add-marketplace-gh-host-account","description":"Enables the add_marketplace_gh_host_account command without any pre-configured scope.","commands":{"allow":["add_marketplace_gh_host_account"],"deny":[]}},"allow-add-marketplace-token-account":{"identifier":"allow-add-marketplace-token-account","description":"Enables the add_marketplace_token_account command without any pre-configured scope.","commands":{"allow":["add_marketplace_token_account"],"deny":[]}},"allow-add-project":{"identifier":"allow-add-project","description":"Enables the add_project command without any pre-configured scope.","commands":{"allow":["add_project"],"deny":[]}},"allow-add-scheduled-task":{"identifier":"allow-add-scheduled-task","description":"Enables the add_scheduled_task command without any pre-configured scope.","commands":{"allow":["add_scheduled_task"],"deny":[]}},"allow-apply-marketplace-now":{"identifier":"allow-apply-marketplace-now","description":"Enables the apply_marketplace_now command without any pre-configured scope.","commands":{"allow":["apply_marketplace_now"],"deny":[]}},"allow-apply-settings-import":{"identifier":"allow-apply-settings-import","description":"Enables the apply_settings_import command without any pre-configured scope.","commands":{"allow":["apply_settings_import"],"deny":[]}},"allow-aws-connection-identity":{"identifier":"allow-aws-connection-identity","description":"Enables the aws_connection_identity command without any pre-configured scope.","commands":{"allow":["aws_connection_identity"],"deny":[]}},"allow-aws-connection-sso-login":{"identifier":"allow-aws-connection-sso-login","description":"Enables the aws_connection_sso_login command without any pre-configured scope.","commands":{"allow":["aws_connection_sso_login"],"deny":[]}},"allow-aws-cost-estimate":{"identifier":"allow-aws-cost-estimate","description":"Enables the aws_cost_estimate command without any pre-configured scope.","commands":{"allow":["aws_cost_estimate"],"deny":[]}},"allow-aws-ec2-capacity":{"identifier":"allow-aws-ec2-capacity","description":"Enables the aws_ec2_capacity command without any pre-configured scope.","commands":{"allow":["aws_ec2_capacity"],"deny":[]}},"allow-aws-environment-max-hours":{"identifier":"allow-aws-environment-max-hours","description":"Enables the aws_environment_max_hours command without any pre-configured scope.","commands":{"allow":["aws_environment_max_hours"],"deny":[]}},"allow-aws-policy-document":{"identifier":"allow-aws-policy-document","description":"Enables the aws_policy_document command without any pre-configured scope.","commands":{"allow":["aws_policy_document"],"deny":[]}},"allow-aws-sso-refresh":{"identifier":"allow-aws-sso-refresh","description":"Enables the aws_sso_refresh command without any pre-configured scope.","commands":{"allow":["aws_sso_refresh"],"deny":[]}},"allow-aws-task-log":{"identifier":"allow-aws-task-log","description":"Enables the aws_task_log command without any pre-configured scope.","commands":{"allow":["aws_task_log"],"deny":[]}},"allow-aws-workspace-status":{"identifier":"allow-aws-workspace-status","description":"Enables the aws_workspace_status command without any pre-configured scope.","commands":{"allow":["aws_workspace_status"],"deny":[]}},"allow-build-gateway-image":{"identifier":"allow-build-gateway-image","description":"Enables the build_gateway_image command without any pre-configured scope.","commands":{"allow":["build_gateway_image"],"deny":[]}},"allow-build-image":{"identifier":"allow-build-image","description":"Enables the build_image command without any pre-configured scope.","commands":{"allow":["build_image"],"deny":[]}},"allow-build-stt-image":{"identifier":"allow-build-stt-image","description":"Enables the build_stt_image command without any pre-configured scope.","commands":{"allow":["build_stt_image"],"deny":[]}},"allow-cancel-aws-deployment":{"identifier":"allow-cancel-aws-deployment","description":"Enables the cancel_aws_deployment command withoutLine truncated
+1 -1
View File
@@ -1 +1 @@
{"default":{"identifier":"default","description":"Default capabilities for Triple-C. Every entry here is an IPC command a compromised webview can call directly, so the set is an enumeration of what `app/src` actually invokes — plugin and core grants verified against tauri 2.11.0's `PLUGINS` table in tauri's own `build.rs` rather than assumed from a plugin's `default` set, app-command grants (the bare `allow-*` entries) cross-checked by this crate's `build.rs` against `generate_handler!`. `core:default` in particular is NOT used: it is an alias for `core:{path,event,window,webview,app,image,resources,menu,tray}:default`, and `core:image:default` carries `allow-from-path`, whose handler (`tauri-2.11.0/src/image/plugin.rs:41` → `src/image/mod.rs:96`) is a bare `std::fs::read(path)` with no scope mechanism of any kind. Nothing imports `@tauri-apps/api/image`, so the whole plugin is dropped rather than scoped — there is nothing to scope it with. `core:menu` and `core:tray` are dropped for the same reason (no menu, no tray icon); `core:window` and `core:path` because nothing imports them; `core:resources:allow-close` because no frontend value is a `Resource`; and `core:event`'s `allow-emit`/`allow-emit-to` because the frontend only ever *listens* — every emit in this app originates in Rust. Three notes on what is deliberately kept or accepted: (1) `core:webview:allow-internal-toggle-devtools` is not called by `app/src` at all — it is called by Tauri's own injected `toggle-devtools.js`, which binds Ctrl/Cmd+Shift+I. Both that script and the command behind it are `#[cfg(any(debug_assertions, feature = \"devtools\"))]`, so this grant is a `tauri dev` convenience that does not exist in a release bundle. (2) `opener:allow-open-url` is **gone**. It could not be narrowed by host — `TerminalView`'s `WebLinksAddon` opens links Claude printed inside the container, which are arbitrary by construction, so a host allowlist would have deleted the feature rather than bounded it — and it was carried here as an accepted residual risk: a compromised webview could make the OS open an attacker-chosen http(s) URL, an outbound channel. That risk is now closed rather than recorded. Every host-browser open in the app goes through the `open_url_external` command in `url_open.rs`, which exists because the AppImage environment leaks into a cold-launched browser on Linux (triple-c#34) and which re-validates the URL in Rust — scheme allowlist, no embedded credentials, no control characters, length cap, ASCII asserted before `execvp`. On macOS and Windows that command reaches the same plugin as before, via `OpenerExt::open_url`, whose desktop implementation calls `crate::open::open` directly and is therefore not gated by this file at all (`tauri-plugin-opener-2.5.3/src/lib.rs:60`). The plugin stays a dependency for exactly that reason; what is removed is the webview's ability to reach it without passing the Rust validation. (3) `drag:allow-start-drag` is **gone**, together with the OS drag-out it existed for. It could not be scoped — `tauri-plugin-drag` takes the item paths from the caller and has no scope mechanism, so a compromised webview could call `startDrag({ item: ['~/.ssh/id_rsa'] })` against any host path the user can read — and it was carried as an accepted residual risk for one gesture. Drag-out was held back for separate hardening (see branch `hold/disk-and-dragout`) and the plugin is no longer a dependency. Getting a file *out* of a container is either \"Back up container\" on the project's Overview tab, which archives a tree through the Docker API, or the Files tab's per-row \"Save to host…\", which copies one file; getting one *in* is a drop on the Terminal or the Files tab's \"Upload…\". None of the four touches this permission. The Files tab's two are worth separating out here, because they are the only host-path commands in the app whose dialog is opened by **Rust** rather than by the webview — `pick_save_path` and `pick_files_to_upload` in `commands/file_commands.rs` drive `tauri-plugin-dialog` from the backend, so a compromised webview can ask for a picker and nothing more: it cannot name a host path as an *input* to either command. Be precise about the limit of that claim — host paths do still travel outward in error text (`Failed to create /home/j/Documents/x.txt.triple-c-part-1a2b3c4d: Permission denied`), including canonicalized ones, which disclose symlink targets. That is accepted; the app already hands the webview the project paths. What is closed is the direction that mattered — the webview naming where bytes go. That is the shape an earlier revision of this file named as the honest one if the Files tab ever regained host I/O, and it is the shape it regained it in. The `dialog:allow-open` / `dialog:allow-save` grants below are therefore *not* what those two use; they remain for the frontend pickers in Add Project, the Config tab's workspace and access sections, the CA-certificate fielLine truncated
{"default":{"identifier":"default","description":"Default capabilities for Triple-C. Every entry here is an IPC command a compromised webview can call directly, so the set is an enumeration of what `app/src` actually invokes — plugin and core grants verified against tauri 2.11.0's `PLUGINS` table in tauri's own `build.rs` rather than assumed from a plugin's `default` set, app-command grants (the bare `allow-*` entries) cross-checked by this crate's `build.rs` against `generate_handler!`. `core:default` in particular is NOT used: it is an alias for `core:{path,event,window,webview,app,image,resources,menu,tray}:default`, and `core:image:default` carries `allow-from-path`, whose handler (`tauri-2.11.0/src/image/plugin.rs:41` → `src/image/mod.rs:96`) is a bare `std::fs::read(path)` with no scope mechanism of any kind. Nothing imports `@tauri-apps/api/image`, so the whole plugin is dropped rather than scoped — there is nothing to scope it with. `core:menu` and `core:tray` are dropped for the same reason (no menu, no tray icon); `core:window` and `core:path` because nothing imports them; `core:resources:allow-close` because no frontend value is a `Resource`; and `core:event`'s `allow-emit`/`allow-emit-to` because the frontend only ever *listens* — every emit in this app originates in Rust. Three notes on what is deliberately kept or accepted: (1) `core:webview:allow-internal-toggle-devtools` is not called by `app/src` at all — it is called by Tauri's own injected `toggle-devtools.js`, which binds Ctrl/Cmd+Shift+I. Both that script and the command behind it are `#[cfg(any(debug_assertions, feature = \"devtools\"))]`, so this grant is a `tauri dev` convenience that does not exist in a release bundle. (2) `opener:allow-open-url` is **gone**. It could not be narrowed by host — `TerminalView`'s `WebLinksAddon` opens links Claude printed inside the container, which are arbitrary by construction, so a host allowlist would have deleted the feature rather than bounded it — and it was carried here as an accepted residual risk: a compromised webview could make the OS open an attacker-chosen http(s) URL, an outbound channel. That risk is now closed rather than recorded. Every host-browser open in the app goes through the `open_url_external` command in `url_open.rs`, which exists because the AppImage environment leaks into a cold-launched browser on Linux (triple-c#34) and which re-validates the URL in Rust — scheme allowlist, no embedded credentials, no control characters, length cap, ASCII asserted before `execvp`. On macOS and Windows that command reaches the same plugin as before, via `OpenerExt::open_url`, whose desktop implementation calls `crate::open::open` directly and is therefore not gated by this file at all (`tauri-plugin-opener-2.5.3/src/lib.rs:60`). The plugin stays a dependency for exactly that reason; what is removed is the webview's ability to reach it without passing the Rust validation. (3) `drag:allow-start-drag` is **gone**, together with the OS drag-out it existed for. It could not be scoped — `tauri-plugin-drag` takes the item paths from the caller and has no scope mechanism, so a compromised webview could call `startDrag({ item: ['~/.ssh/id_rsa'] })` against any host path the user can read — and it was carried as an accepted residual risk for one gesture. Drag-out was held back for separate hardening (see branch `hold/disk-and-dragout`) and the plugin is no longer a dependency. Getting a file *out* of a container is either \"Back up container\" on the project's Overview tab, which archives a tree through the Docker API, or the Files tab's per-row \"Save to host…\", which copies one file; getting one *in* is a drop on the Terminal or the Files tab's \"Upload…\". None of the four touches this permission. The Files tab's two are worth separating out here, because they are the only host-path commands in the app whose dialog is opened by **Rust** rather than by the webview — `pick_save_path` and `pick_files_to_upload` in `commands/file_commands.rs` drive `tauri-plugin-dialog` from the backend, so a compromised webview can ask for a picker and nothing more: it cannot name a host path as an *input* to either command. Be precise about the limit of that claim — host paths do still travel outward in error text (`Failed to create /home/j/Documents/x.txt.triple-c-part-1a2b3c4d: Permission denied`), including canonicalized ones, which disclose symlink targets. That is accepted; the app already hands the webview the project paths. What is closed is the direction that mattered — the webview naming where bytes go. That is the shape an earlier revision of this file named as the honest one if the Files tab ever regained host I/O, and it is the shape it regained it in. The `dialog:allow-open` / `dialog:allow-save` grants below are therefore *not* what those two use; they remain for the frontend pickers in Add Project, the Config tab's workspace and access sections, the CA-certificate fielLine truncated
@@ -422,6 +422,12 @@
"const": "allow-aws-cost-estimate",
"markdownDescription": "Enables the aws_cost_estimate command without any pre-configured scope."
},
{
"description": "Enables the aws_ec2_capacity command without any pre-configured scope.",
"type": "string",
"const": "allow-aws-ec2-capacity",
"markdownDescription": "Enables the aws_ec2_capacity command without any pre-configured scope."
},
{
"description": "Enables the aws_environment_max_hours command without any pre-configured scope.",
"type": "string",
@@ -644,6 +650,12 @@
"const": "allow-download-container-file",
"markdownDescription": "Enables the download_container_file command without any pre-configured scope."
},
{
"description": "Enables the ec2_fit_problem command without any pre-configured scope.",
"type": "string",
"const": "allow-ec2-fit-problem",
"markdownDescription": "Enables the ec2_fit_problem command without any pre-configured scope."
},
{
"description": "Enables the export_settings command without any pre-configured scope.",
"type": "string",
@@ -824,6 +836,12 @@
"const": "allow-list-container-files",
"markdownDescription": "Enables the list_container_files command without any pre-configured scope."
},
{
"description": "Enables the list_ec2_instance_types command without any pre-configured scope.",
"type": "string",
"const": "allow-list-ec2-instance-types",
"markdownDescription": "Enables the list_ec2_instance_types command without any pre-configured scope."
},
{
"description": "Enables the list_marketplace_snapshots command without any pre-configured scope.",
"type": "string",
@@ -1394,6 +1412,12 @@
"const": "deny-aws-cost-estimate",
"markdownDescription": "Denies the aws_cost_estimate command without any pre-configured scope."
},
{
"description": "Denies the aws_ec2_capacity command without any pre-configured scope.",
"type": "string",
"const": "deny-aws-ec2-capacity",
"markdownDescription": "Denies the aws_ec2_capacity command without any pre-configured scope."
},
{
"description": "Denies the aws_environment_max_hours command without any pre-configured scope.",
"type": "string",
@@ -1616,6 +1640,12 @@
"const": "deny-download-container-file",
"markdownDescription": "Denies the download_container_file command without any pre-configured scope."
},
{
"description": "Denies the ec2_fit_problem command without any pre-configured scope.",
"type": "string",
"const": "deny-ec2-fit-problem",
"markdownDescription": "Denies the ec2_fit_problem command without any pre-configured scope."
},
{
"description": "Denies the export_settings command without any pre-configured scope.",
"type": "string",
@@ -1796,6 +1826,12 @@
"const": "deny-list-container-files",
"markdownDescription": "Denies the list_container_files command without any pre-configured scope."
},
{
"description": "Denies the list_ec2_instance_types command without any pre-configured scope.",
"type": "string",
"const": "deny-list-ec2-instance-types",
"markdownDescription": "Denies the list_ec2_instance_types command without any pre-configured scope."
},
{
"description": "Denies the list_marketplace_snapshots command without any pre-configured scope.",
"type": "string",
@@ -422,6 +422,12 @@
"const": "allow-aws-cost-estimate",
"markdownDescription": "Enables the aws_cost_estimate command without any pre-configured scope."
},
{
"description": "Enables the aws_ec2_capacity command without any pre-configured scope.",
"type": "string",
"const": "allow-aws-ec2-capacity",
"markdownDescription": "Enables the aws_ec2_capacity command without any pre-configured scope."
},
{
"description": "Enables the aws_environment_max_hours command without any pre-configured scope.",
"type": "string",
@@ -644,6 +650,12 @@
"const": "allow-download-container-file",
"markdownDescription": "Enables the download_container_file command without any pre-configured scope."
},
{
"description": "Enables the ec2_fit_problem command without any pre-configured scope.",
"type": "string",
"const": "allow-ec2-fit-problem",
"markdownDescription": "Enables the ec2_fit_problem command without any pre-configured scope."
},
{
"description": "Enables the export_settings command without any pre-configured scope.",
"type": "string",
@@ -824,6 +836,12 @@
"const": "allow-list-container-files",
"markdownDescription": "Enables the list_container_files command without any pre-configured scope."
},
{
"description": "Enables the list_ec2_instance_types command without any pre-configured scope.",
"type": "string",
"const": "allow-list-ec2-instance-types",
"markdownDescription": "Enables the list_ec2_instance_types command without any pre-configured scope."
},
{
"description": "Enables the list_marketplace_snapshots command without any pre-configured scope.",
"type": "string",
@@ -1394,6 +1412,12 @@
"const": "deny-aws-cost-estimate",
"markdownDescription": "Denies the aws_cost_estimate command without any pre-configured scope."
},
{
"description": "Denies the aws_ec2_capacity command without any pre-configured scope.",
"type": "string",
"const": "deny-aws-ec2-capacity",
"markdownDescription": "Denies the aws_ec2_capacity command without any pre-configured scope."
},
{
"description": "Denies the aws_environment_max_hours command without any pre-configured scope.",
"type": "string",
@@ -1616,6 +1640,12 @@
"const": "deny-download-container-file",
"markdownDescription": "Denies the download_container_file command without any pre-configured scope."
},
{
"description": "Denies the ec2_fit_problem command without any pre-configured scope.",
"type": "string",
"const": "deny-ec2-fit-problem",
"markdownDescription": "Denies the ec2_fit_problem command without any pre-configured scope."
},
{
"description": "Denies the export_settings command without any pre-configured scope.",
"type": "string",
@@ -1796,6 +1826,12 @@
"const": "deny-list-container-files",
"markdownDescription": "Denies the list_container_files command without any pre-configured scope."
},
{
"description": "Denies the list_ec2_instance_types command without any pre-configured scope.",
"type": "string",
"const": "deny-list-ec2-instance-types",
"markdownDescription": "Denies the list_ec2_instance_types command without any pre-configured scope."
},
{
"description": "Denies the list_marketplace_snapshots command without any pre-configured scope.",
"type": "string",
+345 -11
View File
@@ -29,8 +29,9 @@ use serde_json::{json, Map, Value};
use super::error::{classify_sdk_error, AwsError};
use super::identity::CallerIdentity;
use super::tofu_modules::{
environment_parameter_name, owner_key, state_bucket_name, state_key, user_role_names,
validate_environment_name, Deployment, BOUNDARY_POLICY_NAME,
environment_parameter_name, owner_key, state_bucket_name, state_key, user_ec2_stem,
user_record_parameter_name, user_role_names, validate_ec2_user_tag, validate_environment_name,
Deployment, BOUNDARY_POLICY_NAME, EC2_MAX_INSTANCES,
};
use crate::models::host_connection::validate_supported_region;
@@ -309,15 +310,97 @@ pub enum EnvironmentAction {
#[serde(tag = "state", rename_all = "snake_case")]
pub enum OnboardingState {
/// Their task role exists and, where the bucket is readable, their state
/// records a finished apply (its `onboarding_complete` output).
Joined,
/// records a finished apply (its `onboarding_complete` output). `ec2` is
/// their EC2 capacity, from their user record.
Joined { ec2: Option<UserCompute> },
NotJoined,
/// A local runner, or a recent lock.
InProgress { lock: Option<LockInfo> },
/// The last onboarding run didn't finish: a stale lock, a failed run on
/// this machine, or a state without the completion output. Join is a
/// re-apply and finishes it.
Incomplete { lock: Option<LockInfo> },
/// re-apply and finishes it; `ec2` is what their record still holds, so
/// the re-apply can keep it.
Incomplete {
lock: Option<LockInfo>,
ec2: Option<UserCompute>,
},
}
impl OnboardingState {
fn with_ec2(self, compute: Option<UserCompute>) -> Self {
match self {
OnboardingState::Joined { .. } => OnboardingState::Joined { ec2: compute },
OnboardingState::Incomplete { lock, .. } => OnboardingState::Incomplete { lock, ec2: compute },
other => other,
}
}
}
/// A user's EC2 capacity in one environment: their validated user record
/// (`tofu_modules::user_record_parameter_name`). A missing record means no
/// NEW EC2 runs there, not that nothing runs (teardown removes it first).
#[derive(Serialize, Deserialize, Clone, Debug, PartialEq, Eq)]
pub struct UserCompute {
pub instance_type: String,
pub max_instances: u32,
/// The `triple-c-user` value the cluster is named after.
pub user_tag: String,
pub cluster_arn: String,
pub capacity_provider: String,
}
/// Validate a user record read from `user_record_parameter_name(env, …)`:
/// schema 1, a type the app's table knows, 1 to `EC2_MAX_INSTANCES`
/// instances (absent in records written before the field: 1), a valid user
/// tag, and the cluster and capacity provider of exactly that tag's per-user
/// names in `account` and `region`. The error names the field, never its
/// value.
pub fn parse_user_record(
env: &str,
value: &str,
account: &str,
region: &str,
) -> Result<UserCompute, String> {
if value.len() > MAX_UNTRUSTED_BYTES {
return Err("the record is larger than 4 KiB".into());
}
let parsed: Value =
serde_json::from_str(value).map_err(|_| "the record is not JSON".to_string())?;
let obj = parsed
.as_object()
.ok_or_else(|| "the record is not a JSON object".to_string())?;
if field(obj, "schema")?.as_u64() != Some(RECORD_SCHEMA) {
return Err("schema is not 1".into());
}
let instance_type = str_field(obj, "instance_type")?;
if crate::models::project::ec2_instance_spec(instance_type).is_none() {
return Err("instance_type is not a supported instance type".into());
}
let max_instances = match obj.get("max_instances") {
None => 1,
Some(v) => v
.as_u64()
.and_then(|v| u32::try_from(v).ok())
.filter(|v| (1..=EC2_MAX_INSTANCES).contains(v))
.ok_or_else(|| format!("max_instances is not a number from 1 to {EC2_MAX_INSTANCES}"))?,
};
let user_tag = validate_ec2_user_tag(str_field(obj, "user_tag")?)
.map_err(|_| "user_tag is not a valid triple-c-user value".to_string())?;
let stem = user_ec2_stem(env, &user_tag);
let cluster_arn = str_field(obj, "cluster_arn")?;
if cluster_arn != format!("arn:aws:ecs:{region}:{account}:cluster/{stem}") {
return Err("cluster_arn is not this user's cluster in this account and region".into());
}
if str_field(obj, "capacity_provider")? != stem {
return Err("capacity_provider is not this user's capacity provider".into());
}
Ok(UserCompute {
instance_type: instance_type.to_string(),
max_instances,
user_tag,
cluster_arn: cluster_arn.to_string(),
capacity_provider: stem,
})
}
#[derive(Serialize, Clone, Debug, PartialEq)]
@@ -565,9 +648,12 @@ fn onboarding_from_runs(
Some(OnboardingState::InProgress { lock })
} else if lock.is_some() {
actions.push(Unlock);
Some(OnboardingState::Incomplete { lock })
Some(OnboardingState::Incomplete { lock, ec2: None })
} else if run.failed || run.errored_state {
Some(OnboardingState::Incomplete { lock: None })
Some(OnboardingState::Incomplete {
lock: None,
ec2: None,
})
} else {
None
};
@@ -597,6 +683,8 @@ pub trait DetectApi: Send + Sync {
account: &str,
) -> impl Future<Output = Result<Knowable, AwsError>> + Send;
fn role_exists(&self, name: &str) -> impl Future<Output = Result<Knowable, AwsError>> + Send;
/// An SSM String parameter (a user record).
fn parameter(&self, name: &str) -> impl Future<Output = Result<Param, AwsError>> + Send;
/// A state object's contents, capped; `None` when it is missing, can't
/// be read or is too large.
fn state_document(
@@ -675,8 +763,13 @@ pub(crate) async fn scan_in_bucket<D: DetectApi>(
// role), which is not Joined.
None if objects.as_ref().is_some_and(|o| o.has_state(&user_state)) => {
match d.state_document(bucket, &user_state).await? {
Some(text) if onboarding_complete(&text) => OnboardingState::Joined,
_ => OnboardingState::Incomplete { lock: None },
Some(text) if onboarding_complete(&text) => {
OnboardingState::Joined { ec2: None }
}
_ => OnboardingState::Incomplete {
lock: None,
ec2: None,
},
}
}
// Without a state here (a workspace user, who can't list the
@@ -685,11 +778,20 @@ pub(crate) async fn scan_in_bucket<D: DetectApi>(
None => {
let (task_role, _) = user_role_names(&env.name, &key);
match d.role_exists(&task_role).await? {
Knowable::Yes => OnboardingState::Joined,
Knowable::Yes => OnboardingState::Joined { ec2: None },
Knowable::No | Knowable::Unknown => OnboardingState::NotJoined,
}
}
});
if let Some(state) = env.onboarding.take() {
let compute = match state {
OnboardingState::Joined { .. } | OnboardingState::Incomplete { .. } => {
user_compute(d, &env.name, &key, &who.account, region).await?
}
_ => None,
};
env.onboarding = Some(state.with_ec2(compute));
}
}
Ok(EnvironmentScan {
@@ -703,6 +805,234 @@ pub(crate) async fn scan_in_bucket<D: DetectApi>(
})
}
/// An SSM parameter read: its value, missing, or this caller can't read it.
#[derive(Clone, Debug, PartialEq)]
pub enum Param {
Value(String),
Missing,
Denied,
}
/// A user record as read: none, valid, or present but unusable (unreadable
/// or failing validation; the reason names a field, never a value).
#[derive(Clone, Debug, PartialEq)]
pub enum UserRecord {
Absent,
Valid(UserCompute),
Unusable(String),
}
impl UserRecord {
pub fn valid(&self) -> Option<&UserCompute> {
match self {
UserRecord::Valid(c) => Some(c),
_ => None,
}
}
}
/// [`UserRecord`] from a read of `user_record_parameter_name(env, …)`.
pub fn user_record_from(read: Param, env: &str, account: &str, region: &str) -> UserRecord {
match read {
Param::Missing => UserRecord::Absent,
Param::Denied => UserRecord::Unusable("it can't be read with these credentials".into()),
Param::Value(text) => match parse_user_record(env, &text, account, region) {
Ok(c) => UserRecord::Valid(c),
Err(reason) => UserRecord::Unusable(reason),
},
}
}
/// The caller's EC2 capacity in `env` from their user record; `None` when
/// there is no usable record (an unusable one is logged, never echoed).
pub(crate) async fn user_compute<D: DetectApi>(
d: &D,
env: &str,
owner_key: &str,
account: &str,
region: &str,
) -> Result<Option<UserCompute>, AwsError> {
let read = d.parameter(&user_record_parameter_name(env, owner_key)).await?;
match user_record_from(read, env, account, region) {
UserRecord::Valid(c) => Ok(Some(c)),
UserRecord::Absent => Ok(None),
UserRecord::Unusable(reason) => {
log::warn!("Ignoring the EC2 user record for environment {env}: {reason}");
Ok(None)
}
}
}
/// Who holds a per-user EC2 name (`tofu_modules::user_ec2_stem`): an ECS
/// cluster or Auto Scaling group of that name and its `triple-c:owner` tag.
#[derive(Clone, Debug, PartialEq)]
pub struct NameHolder {
/// "ECS cluster" or "Auto Scaling group".
pub kind: &'static str,
pub owner: Option<String>,
}
/// An ECS cluster or Auto Scaling group found by name: its status (clusters
/// only) and tags.
#[derive(Clone, Debug, PartialEq, Default)]
pub struct NamedResource {
pub status: Option<String>,
pub tags: Vec<(String, String)>,
}
/// What an onboarding apply asks AWS before it starts (R3): the caller's
/// user record and whether the per-user EC2 names are taken. The cluster
/// read needs `ecs:DescribeClusters` (tags included); the group read
/// `autoscaling:DescribeAutoScalingGroups`, which only the environment-admin
/// tier has.
pub trait OnboardApi: Send + Sync {
fn parameter(&self, name: &str) -> impl Future<Output = Result<Param, AwsError>> + Send;
fn cluster(
&self,
name: &str,
) -> impl Future<Output = Result<Option<NamedResource>, AwsError>> + Send;
fn group(
&self,
name: &str,
) -> impl Future<Output = Result<Option<NamedResource>, AwsError>> + Send;
}
const OWNER_TAG: &str = "triple-c:owner";
fn owner_of(r: &NamedResource) -> Option<String> {
r.tags
.iter()
.find(|(k, _)| k == OWNER_TAG)
.map(|(_, v)| v.clone())
}
/// The cluster holding `stem`, if any. Every status but INACTIVE (a deleted
/// cluster) holds the name: PROVISIONING, DEPROVISIONING and FAILED too.
pub async fn cluster_holder<A: OnboardApi>(
api: &A,
stem: &str,
) -> Result<Option<NameHolder>, AwsError> {
Ok(api
.cluster(stem)
.await?
.filter(|c| c.status.as_deref() != Some("INACTIVE"))
.map(|c| NameHolder {
kind: "ECS cluster",
owner: owner_of(&c),
}))
}
/// Everything holding `stem`: its cluster and its Auto Scaling group.
pub async fn ec2_name_holders<A: OnboardApi>(
api: &A,
stem: &str,
) -> Result<Vec<NameHolder>, AwsError> {
let mut out: Vec<NameHolder> = cluster_holder(api, stem).await?.into_iter().collect();
if let Some(g) = api.group(stem).await? {
out.push(NameHolder {
kind: "Auto Scaling group",
owner: owner_of(&g),
});
}
Ok(out)
}
/// `ssm:GetParameter` as a [`Param`].
pub(crate) async fn get_parameter(ssm: &aws_sdk_ssm::Client, name: &str) -> Result<Param, AwsError> {
let res = ssm
.get_parameter()
.name(name)
.with_decryption(false)
.send()
.await;
match res {
Ok(out) => Ok(out
.parameter()
.and_then(|p| p.value())
.map_or(Param::Missing, |v| Param::Value(v.to_string()))),
Err(e) if e.as_service_error().is_some_and(|s| s.is_parameter_not_found()) => {
Ok(Param::Missing)
}
Err(e) => {
let (code, status) = error_parts(&e);
match lookup_outcome(code.as_deref(), status) {
Lookup::Missing => Ok(Param::Missing),
Lookup::Denied => Ok(Param::Denied),
Lookup::Error => Err(classify_sdk_error("ssm:GetParameter", &e)),
}
}
}
}
/// The real [`OnboardApi`], from a connection's `SdkConfig`.
pub struct SdkOnboardApi {
ssm: aws_sdk_ssm::Client,
ecs: aws_sdk_ecs::Client,
autoscaling: aws_sdk_autoscaling::Client,
}
impl SdkOnboardApi {
pub fn new(cfg: &SdkConfig) -> Self {
SdkOnboardApi {
ssm: aws_sdk_ssm::Client::new(cfg),
ecs: aws_sdk_ecs::Client::new(cfg),
autoscaling: aws_sdk_autoscaling::Client::new(cfg),
}
}
}
impl OnboardApi for SdkOnboardApi {
async fn parameter(&self, name: &str) -> Result<Param, AwsError> {
get_parameter(&self.ssm, name).await
}
async fn cluster(&self, name: &str) -> Result<Option<NamedResource>, AwsError> {
// A missing name is a `failures` entry, not an error.
let out = self
.ecs
.describe_clusters()
.clusters(name)
.include(aws_sdk_ecs::types::ClusterField::Tags)
.send()
.await
.map_err(|e| classify_sdk_error("ecs:DescribeClusters", &e))?;
Ok(out
.clusters()
.iter()
.find(|c| c.cluster_name() == Some(name))
.map(|c| NamedResource {
status: c.status().map(str::to_string),
tags: c
.tags()
.iter()
.filter_map(|t| Some((t.key()?.to_string(), t.value()?.to_string())))
.collect(),
}))
}
async fn group(&self, name: &str) -> Result<Option<NamedResource>, AwsError> {
let out = self
.autoscaling
.describe_auto_scaling_groups()
.auto_scaling_group_names(name)
.send()
.await
.map_err(|e| classify_sdk_error("autoscaling:DescribeAutoScalingGroups", &e))?;
Ok(out
.auto_scaling_groups()
.iter()
.find(|g| g.auto_scaling_group_name() == Some(name))
.map(|g| NamedResource {
status: None,
tags: g
.tags()
.iter()
.filter_map(|t| Some((t.key()?.to_string(), t.value()?.to_string())))
.collect(),
}))
}
}
/// An OpenTofu state that manages nothing (what `tofu destroy` leaves):
/// valid JSON whose `resources` is an empty list. Anything else, unreadable
/// contents included, counts as a deployment.
@@ -1015,6 +1345,10 @@ impl DetectApi for SdkDetectApi {
knowable("iam:GetRole", res)
}
async fn parameter(&self, name: &str) -> Result<Param, AwsError> {
get_parameter(&self.ssm, name).await
}
async fn state_document(&self, bucket: &str, key: &str) -> Result<Option<String>, AwsError> {
const ACTION: &str = "s3:GetObject";
let res = self
+180 -5
View File
@@ -494,6 +494,8 @@ struct FakeDetect {
roles: HashMap<String, Knowable>,
/// State objects' contents, by key.
documents: HashMap<String, String>,
/// SSM parameters (user records), by name.
parameters: HashMap<String, String>,
calls: Mutex<Vec<String>>,
}
@@ -506,6 +508,7 @@ impl FakeDetect {
boundary: Knowable::Yes,
roles: HashMap::new(),
documents: HashMap::new(),
parameters: HashMap::new(),
calls: Mutex::new(Vec::new()),
}
}
@@ -548,6 +551,10 @@ impl DetectApi for FakeDetect {
self.log(format!("role {name}"));
Ok(self.roles.get(name).cloned().unwrap_or(Knowable::No))
}
async fn parameter(&self, name: &str) -> Result<Param, AwsError> {
self.log(format!("parameter {name}"));
Ok(self.parameters.get(name).cloned().map_or(Param::Missing, Param::Value))
}
async fn state_document(&self, bucket: &str, key: &str) -> Result<Option<String>, AwsError> {
self.log(format!("document {bucket} {key}"));
Ok(self.documents.get(key).cloned())
@@ -612,8 +619,8 @@ fn onboarding_is_found_by_state_then_by_role() {
assert_eq!(scan.owner_id, USER_ID);
assert_eq!(scan.owner_key, owner_key(REGION, USER_ID));
let onboarding = |n: &str| find(&scan.environments, n).onboarding.clone();
assert_eq!(onboarding("bystate"), Some(OnboardingState::Joined));
assert_eq!(onboarding("byrole"), Some(OnboardingState::Joined));
assert_eq!(onboarding("bystate"), Some(OnboardingState::Joined { ec2: None }));
assert_eq!(onboarding("byrole"), Some(OnboardingState::Joined { ec2: None }));
assert_eq!(onboarding("denied"), Some(OnboardingState::NotJoined));
assert_eq!(onboarding("absent"), Some(OnboardingState::NotJoined));
assert!(matches!(
@@ -662,7 +669,7 @@ fn onboarding_is_found_by_state_then_by_role() {
assert_eq!(x.onboarding, Some(OnboardingState::InProgress { lock: None }));
assert_eq!(x.onboarding_actions, vec![EnvironmentAction::Follow]);
let y = find(&scan.environments, "y");
let Some(OnboardingState::Incomplete { lock: Some(lock) }) = &y.onboarding else {
let Some(OnboardingState::Incomplete { lock: Some(lock), ec2: None }) = &y.onboarding else {
panic!("{y:#?}");
};
assert!(!lock.local_process);
@@ -982,12 +989,12 @@ fn a_failed_onboarding_is_incomplete_not_joined() {
let scan = run_scan(&d, &local);
let row = |n: &str| find(&scan.environments, n).clone();
assert_eq!(row("done").onboarding, Some(OnboardingState::Joined));
assert_eq!(row("done").onboarding, Some(OnboardingState::Joined { ec2: None }));
for name in ["partial", "failedhere", "errored"] {
let r = row(name);
assert_eq!(
r.onboarding,
Some(OnboardingState::Incomplete { lock: None }),
Some(OnboardingState::Incomplete { lock: None, ec2: None }),
"{name}"
);
// Join is kept: an idempotent re-apply finishes it.
@@ -1107,3 +1114,171 @@ fn the_tls_statement_is_added_to_a_bucket_policy_never_replacing_it() {
// Something that isn't a policy is never overwritten.
assert!(state_bucket_policy_update(Some("not json"), &bucket).is_err());
}
fn user_record(env: &str, tag: &str) -> Value {
let stem = crate::aws::tofu_modules::user_ec2_stem(env, tag);
json!({
"schema": 1,
"instance_type": "m7g.large",
"cluster_arn": format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/{stem}"),
"capacity_provider": stem,
"user_tag": tag,
"max_instances": 3,
})
}
#[test]
fn a_user_record_reads_as_the_users_compute() {
let rec = user_record("dev", "alice");
let c = parse_user_record("dev", &rec.to_string(), ACCOUNT, REGION).unwrap();
assert_eq!(
c,
UserCompute {
instance_type: "m7g.large".into(),
max_instances: 3,
user_tag: "alice".into(),
cluster_arn: format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/triple-c-dev--u-alice"),
capacity_provider: "triple-c-dev--u-alice".into(),
}
);
// A record written before max_instances existed: the module's default.
let mut old = rec.clone();
old.as_object_mut().unwrap().remove("max_instances");
let c = parse_user_record("dev", &old.to_string(), ACCOUNT, REGION).unwrap();
assert_eq!(c.max_instances, 1);
}
#[test]
fn a_user_record_is_untrusted() {
let bad = |field: &str, value: Value, why: &str| {
let mut rec = user_record("dev", "alice");
rec[field] = value;
let err = parse_user_record("dev", &rec.to_string(), ACCOUNT, REGION).unwrap_err();
assert!(err.contains(why), "{field}: {err}");
assert!(!err.contains("evil"), "{err}");
};
bad("schema", json!(2), "schema");
bad("instance_type", json!("p5.48xlarge"), "instance_type");
bad("instance_type", json!(7), "instance_type");
bad("max_instances", json!(0), "max_instances");
bad("max_instances", json!(5), "max_instances");
bad("max_instances", json!("2"), "max_instances");
bad("user_tag", json!("evil--tag"), "user_tag");
// Another user's cluster, or one in another account or region.
bad(
"cluster_arn",
json!(format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/triple-c-dev--u-bob")),
"cluster_arn",
);
bad(
"cluster_arn",
json!(format!("arn:aws:ecs:us-east-1:{ACCOUNT}:cluster/triple-c-dev--u-alice")),
"cluster_arn",
);
bad(
"cluster_arn",
json!(format!("arn:aws:ecs:{REGION}:999999999999:cluster/triple-c-dev--u-alice")),
"cluster_arn",
);
bad("capacity_provider", json!("evil"), "capacity_provider");
// Another environment's record.
let rec = user_record("prod", "alice").to_string();
assert!(parse_user_record("dev", &rec, ACCOUNT, REGION).is_err());
for text in ["", "not json", "[]", &"x".repeat(5000)] {
assert!(parse_user_record("dev", text, ACCOUNT, REGION).is_err());
}
}
fn compute_of(s: &OnboardingState) -> Option<&UserCompute> {
match s {
OnboardingState::Joined { ec2 } | OnboardingState::Incomplete { ec2, .. } => ec2.as_ref(),
_ => None,
}
}
#[test]
fn onboarding_carries_the_users_compute_from_their_record() {
use crate::aws::tofu_modules::user_record_parameter_name;
let ok = owner_key(REGION, USER_ID);
let mut d = FakeDetect::new()
.with_record("joined", 1)
.with_record("fargate", 1)
.with_record("broken", 1)
.with_record("partial", 1)
.with_record("absent", 1);
d.roles.insert(task_role("joined"), Knowable::Yes);
d.roles.insert(task_role("fargate"), Knowable::Yes);
d.roles.insert(task_role("broken"), Knowable::Yes);
d.objects = Some(StateObjects {
states: vec![
env_key("joined"),
env_key("fargate"),
env_key("broken"),
env_key("partial"),
env_key("absent"),
user_key("partial"),
],
locks: vec![],
});
d.documents.insert(user_key("partial"), "{\"resources\":[{}]}".into());
for env in ["joined", "partial", "absent"] {
d.parameters.insert(
user_record_parameter_name(env, &ok),
user_record(env, "alice").to_string(),
);
}
d.parameters.insert(
user_record_parameter_name("broken", &ok),
user_record("other", "alice").to_string(),
);
// `joined`'s state is not readable here: found by role.
d.objects.as_mut().unwrap().states.retain(|k| *k != user_key("joined"));
let scan = run_scan(&d, &nobody_local);
let onboarding = |n: &str| find(&scan.environments, n).onboarding.clone().unwrap();
let joined = onboarding("joined");
let compute = compute_of(&joined).expect("joined has compute");
assert_eq!(compute.instance_type, "m7g.large");
assert_eq!(compute.max_instances, 3);
assert_eq!(compute.user_tag, "alice");
assert_eq!(onboarding("fargate"), OnboardingState::Joined { ec2: None });
// A record that doesn't validate is no compute.
assert_eq!(onboarding("broken"), OnboardingState::Joined { ec2: None });
// An incomplete onboarding keeps what its record still says.
let partial = onboarding("partial");
assert!(matches!(partial, OnboardingState::Incomplete { .. }), "{partial:?}");
assert_eq!(compute_of(&partial).map(|c| c.max_instances), Some(3));
// Not joined: the record isn't even read.
assert_eq!(onboarding("absent"), OnboardingState::NotJoined);
assert!(!d
.calls()
.contains(&format!("parameter {}", user_record_parameter_name("absent", &ok))));
// Serialized for the UI: the state tag plus `ec2`.
let v = serde_json::to_value(&joined).unwrap();
assert_eq!(v["state"], "joined");
assert_eq!(v["ec2"]["capacity_provider"], "triple-c-joined--u-alice");
assert_eq!(
serde_json::to_value(OnboardingState::Joined { ec2: None }).unwrap(),
json!({"state": "joined", "ec2": null})
);
}
#[test]
fn a_user_record_read_is_absent_valid_or_unusable() {
let ok = user_record("dev", "alice").to_string();
assert_eq!(user_record_from(Param::Missing, "dev", ACCOUNT, REGION), UserRecord::Absent);
assert!(matches!(
user_record_from(Param::Value(ok), "dev", ACCOUNT, REGION),
UserRecord::Valid(_)
));
// Denied or invalid: there may be capacity, so it is not "absent".
assert!(matches!(
user_record_from(Param::Denied, "dev", ACCOUNT, REGION),
UserRecord::Unusable(_)
));
assert!(matches!(
user_record_from(Param::Value("{}".into()), "dev", ACCOUNT, REGION),
UserRecord::Unusable(_)
));
}
+852 -11
View File
@@ -642,6 +642,12 @@ async fn a_workspace_role_reads_fargate_prices_and_a_bare_role_cannot() {
eprintln!("{}: x86 {x86:?}, ARM {arm:?}", e.region);
assert!(x86.vcpu_hour > 0.0 && x86.gb_hour > 0.0 && x86.storage_gb_hour > 0.0);
assert!(arm.vcpu_hour < x86.vcpu_hour, "ARM is cheaper per vCPU");
// E10: an EC2 instance type's on-demand price, with the same rights.
let m7i = super::pricing::fetch_ec2_price(&SdkPricingApi::new(&cfg), &e.region, "m7i.large")
.await
.expect("workspace role reads EC2 prices");
eprintln!("{}: m7i.large {m7i}/hour", e.region);
assert!(m7i > 0.0 && m7i < 1.0, "{m7i}");
let bare = e.role("noprice", None, &[]).await;
let cfg = sdk_config_for_keys(&e.assume_keys(&bare).await, PRICING_REGION);
@@ -1592,7 +1598,7 @@ mod environment {
.run(
ModuleKind::UserOnboard,
Deployment::User { env: &name, owner_key: &okey },
&onboarding_vars(&name, &t.e.region, &owner),
&onboarding_vars(&name, &t.e.region, &owner, None).unwrap(),
TofuOp::Apply,
t.keys_env(&keys),
)
@@ -1601,7 +1607,7 @@ mod environment {
let found = t.scan(&cfg, &who).await;
let env = found.environments.iter().find(|x| x.name == name).unwrap();
// Joined from the state's onboarding_complete output, not just the role.
assert_eq!(env.onboarding, Some(OnboardingState::Joined), "{env:#?}");
assert_eq!(env.onboarding, Some(OnboardingState::Joined { ec2: None }), "{env:#?}");
assert_eq!(env.onboarded_users, Some(1), "{env:#?}");
let (task_role, exec_role) = user_role_names(&name, &okey);
eprintln!("onboarded with region-scoped roles {task_role} / {exec_role}");
@@ -1643,7 +1649,7 @@ mod environment {
.run(
ModuleKind::UserOnboard,
Deployment::User { env: &name, owner_key: &okey },
&onboarding_vars(&name, &t.e.region, &owner),
&onboarding_vars(&name, &t.e.region, &owner, None).unwrap(),
TofuOp::Destroy,
t.keys_env(&keys),
)
@@ -1767,6 +1773,13 @@ mod environment {
config_file_before: Option<Option<std::ffi::OsString>>,
/// `TRIPLEC_E2E_ECR_REPO`: the throwaway repository to force-delete.
ecr_repo: Option<String>,
/// Phase 7: the environment whose per-user EC2 capacity
/// (`triple-c-<env>--u-*`) the guard tears down, and the user's
/// owner key (instance role, user record).
ec2_env: Option<String>,
ec2_okey: Option<String>,
/// Auto Scaling groups a probe made outside that prefix.
probe_groups: Vec<String>,
}
/// The whole phase-5 body: a hang anywhere becomes a panic the guard
@@ -1796,6 +1809,9 @@ mod environment {
Some(None) => std::env::remove_var("AWS_CONFIG_FILE"),
None => {}
}
if let Some(env) = &p.ec2_env {
ec2_cleanup(e, env, p.ec2_okey.as_deref(), &p.probe_groups).await;
}
let ecs = aws_sdk_ecs::Client::new(&e.cfg);
if let Some(cluster) = &p.cluster {
let mut arns = Vec::new();
@@ -1912,6 +1928,211 @@ mod environment {
}
}
/// The AWS CLI as the e2e profile (`keys`: as that session instead):
/// (success, stdout, stderr). The app has no EC2 SDK crate.
async fn aws_cli(keys: Option<&AwsKeys>, region: &str, args: &[&str]) -> (bool, String, String) {
let mut cmd = tokio::process::Command::new("aws");
cmd.args(["--region", region, "--output", "json"]);
match keys {
None => {
cmd.args(["--profile", &super::var("TRIPLEC_AWS_E2E_PROFILE")]);
}
Some(k) => {
cmd.env_remove("AWS_PROFILE")
.env("AWS_ACCESS_KEY_ID", &k.access_key_id)
.env("AWS_SECRET_ACCESS_KEY", &k.secret_access_key);
match &k.session_token {
Some(t) => cmd.env("AWS_SESSION_TOKEN", t),
None => cmd.env_remove("AWS_SESSION_TOKEN"),
};
}
}
cmd.args(args);
match tokio::time::timeout(Duration::from_secs(120), cmd.output()).await {
Ok(Ok(o)) => (
o.status.success(),
String::from_utf8_lossy(&o.stdout).into_owned(),
String::from_utf8_lossy(&o.stderr).into_owned(),
),
Ok(Err(err)) => (false, String::new(), format!("aws cli: {err}")),
Err(_) => (false, String::new(), "aws cli: timed out".into()),
}
}
/// Phase 7's guard, as the account admin: everything of `env`'s per-user
/// EC2 capacity (`triple-c-<env>--u-*`) and the probes' groups, so no
/// instance outlives a failed run. Tasks on the per-user clusters are
/// stopped, groups force-deleted (terminating their instances) and waited
/// for, leftover tagged instances terminated, then capacity providers,
/// container instances, clusters, launch templates, the user record and
/// the instance role/profile deleted. A resource already gone is fine
/// (the modules' destroy then finds nothing). Logs, never fails.
async fn ec2_cleanup(e: &E2e, env: &str, okey: Option<&str>, probe_groups: &[String]) {
let prefix = format!("triple-c-{env}--u-");
let ecs = aws_sdk_ecs::Client::new(&e.cfg);
let asg = aws_sdk_autoscaling::Client::new(&e.cfg);
// Tasks on the per-user clusters.
let mut clusters = Vec::new();
if let Ok(out) = ecs.list_clusters().send().await {
clusters.extend(out.cluster_arns().iter().filter(|a| a.contains(&format!(":cluster/{prefix}"))).cloned());
}
for cluster in &clusters {
if let Ok(out) = ecs.list_tasks().cluster(cluster).send().await {
for arn in out.task_arns() {
eprintln!("cleanup: stopping EC2 task {arn}");
let _ = ecs.stop_task().cluster(cluster).task(arn).reason("triple-c-e2e cleanup").send().await;
}
}
}
// Groups (force: terminates their instances, ignores protection).
let mut groups: Vec<String> = probe_groups.to_vec();
let mut next = None;
loop {
let Ok(out) = asg.describe_auto_scaling_groups().set_next_token(next.take()).send().await else {
eprintln!("cleanup: WARNING: could not list Auto Scaling groups; only the known ones are deleted");
break;
};
for g in out.auto_scaling_groups() {
let n = g.auto_scaling_group_name().unwrap_or_default();
if n.starts_with(&prefix) && !groups.iter().any(|x| x == n) {
groups.push(n.to_string());
}
}
next = out.next_token().map(str::to_string);
if next.is_none() {
break;
}
}
for g in &groups {
match asg.delete_auto_scaling_group().auto_scaling_group_name(g).force_delete(true).send().await {
Ok(_) => eprintln!("cleanup: force-deleting Auto Scaling group {g}"),
Err(err) => {
let msg = format!("{err:?}");
if !msg.contains("not found") && !msg.contains("ValidationError") {
eprintln!("cleanup: delete group {g}: {msg}");
}
}
}
}
if !groups.is_empty() {
for _ in 0..60 {
// A failed read is "not yet gone", never "gone".
let left = asg
.describe_auto_scaling_groups()
.set_auto_scaling_group_names(Some(groups.clone()))
.send()
.await
.map(|o| o.auto_scaling_groups().len())
.unwrap_or(usize::MAX);
if left == 0 {
break;
}
tokio::time::sleep(Duration::from_secs(10)).await;
}
}
// Any instance still tagged with this environment.
let (ok_, out, err) = aws_cli(
None,
&e.region,
&[
"ec2", "describe-instances",
"--filters", &format!("Name=tag:triple-c:environment,Values={env}"),
"Name=instance-state-name,Values=pending,running,stopping,stopped",
"--query", "Reservations[].Instances[].InstanceId",
],
)
.await;
if ok_ {
let ids: Vec<String> = serde_json::from_str(&out).unwrap_or_default();
if !ids.is_empty() {
let mut args = vec!["ec2", "terminate-instances", "--instance-ids"];
args.extend(ids.iter().map(String::as_str));
let r = aws_cli(None, &e.region, &args).await;
eprintln!("cleanup: terminate {ids:?}: ok {}", r.0);
let mut args = vec!["ec2", "wait", "instance-terminated", "--instance-ids"];
args.extend(ids.iter().map(String::as_str));
let _ = aws_cli(None, &e.region, &args).await;
}
} else {
eprintln!("cleanup: describe instances: {err}");
}
// Capacity providers: off their clusters, then deleted.
for cluster in &clusters {
let _ = ecs.put_cluster_capacity_providers().cluster(cluster).set_capacity_providers(Some(vec![])).set_default_capacity_provider_strategy(Some(vec![])).send().await;
}
if let Ok(out) = ecs.describe_capacity_providers().send().await {
for cp in out.capacity_providers() {
let n = cp.name().unwrap_or_default();
if n.starts_with(&prefix) && cp.status() != Some(&aws_sdk_ecs::types::CapacityProviderStatus::Inactive) {
match ecs.delete_capacity_provider().capacity_provider(n).send().await {
Ok(_) => eprintln!("cleanup: deleted capacity provider {n}"),
Err(err) => eprintln!("cleanup: delete capacity provider {n}: {err:?}"),
}
}
}
}
for cluster in &clusters {
if let Ok(out) = ecs.list_container_instances().cluster(cluster).send().await {
for ci in out.container_instance_arns() {
let _ = ecs.deregister_container_instance().cluster(cluster).container_instance(ci).force(true).send().await;
}
}
// Busy (UpdateInProgress) while the provider detaches: retry.
for attempt in 0..12 {
match ecs.delete_cluster().cluster(cluster).send().await {
Ok(_) => {
eprintln!("cleanup: deleted cluster {cluster}");
break;
}
Err(err) if attempt < 11 && format!("{err:?}").contains("UpdateInProgress") => {
tokio::time::sleep(Duration::from_secs(10)).await;
}
Err(err) => {
eprintln!("cleanup: delete cluster {cluster}: {err:?}");
break;
}
}
}
}
// Launch templates.
let (ok_, out, _) = aws_cli(
None,
&e.region,
&["ec2", "describe-launch-templates", "--filters", &format!("Name=launch-template-name,Values={prefix}*"), "--query", "LaunchTemplates[].LaunchTemplateName"],
)
.await;
if ok_ {
let names: Vec<String> = serde_json::from_str(&out).unwrap_or_default();
for n in names {
let r = aws_cli(None, &e.region, &["ec2", "delete-launch-template", "--launch-template-name", &n]).await;
eprintln!("cleanup: delete launch template {n}: ok {} {}", r.0, r.2.trim());
}
}
if let Some(okey) = okey {
let ssm = aws_sdk_ssm::Client::new(&e.cfg);
let param = crate::aws::tofu_modules::user_record_parameter_name(env, okey);
if ssm.delete_parameter().name(&param).send().await.is_ok() {
eprintln!("cleanup: deleted {param}");
}
let role = crate::aws::tofu_modules::user_instance_role_name(env, okey);
if e.iam.remove_role_from_instance_profile().instance_profile_name(&role).role_name(&role).send().await.is_ok() {
eprintln!("cleanup: removed {role} from its instance profile");
}
if e.iam.delete_instance_profile().instance_profile_name(&role).send().await.is_ok() {
eprintln!("cleanup: deleted instance profile {role}");
}
if let Ok(out) = e.iam.list_role_policies().role_name(&role).send().await {
for pn in out.policy_names() {
let _ = e.iam.delete_role_policy().role_name(&role).policy_name(pn).send().await;
}
match e.iam.delete_role().role_name(&role).send().await {
Ok(_) => eprintln!("cleanup: deleted role {role}"),
Err(err) => eprintln!("cleanup: delete role {role}: {err:?}"),
}
}
}
}
/// Poll `f` every `every` until it returns `Some`, or `None` after `within`.
async fn poll<T, F, Fut>(within: Duration, every: Duration, mut f: F) -> Option<T>
where
@@ -1954,6 +2175,8 @@ mod environment {
conn: HostConnection,
ws_cfg: SdkConfig,
ws: CallerIdentity,
/// The environment admin role (fresh sessions via `Env::session`).
admin: Admin,
}
/// The workspace tests' common start: the environment (with
@@ -1967,6 +2190,7 @@ mod environment {
p5: &std::sync::Mutex<Phase5>,
image: &str,
max_workspace_hours: u32,
ec2: Option<&crate::aws::tofu_modules::Ec2Compute>,
) -> Deployed {
use crate::aws::tofu_modules::{environment_cluster_name, user_file_system_token};
@@ -1997,18 +2221,43 @@ mod environment {
let fs_token = user_file_system_token(&name, &okey);
p5_lock(p5).fs_token = Some(fs_token.clone());
// Onboarded by the admin (an onboarding from an admin bundle).
let (keys, _, _) = t.session(&admin).await;
// Onboarded by the admin (an onboarding from an admin bundle). With
// EC2 capacity (phase 7): the user's role carries the triple-c-user
// principal tag, and the vars come from the app's own pre-apply
// step (record, choice, name-collision check) as the admin.
let (keys, admin_cfg, _) = t.session(&admin).await;
let vars = match ec2 {
None => onboarding_vars(&name, &region, &owner, None).unwrap(),
Some(c) => {
use crate::aws::tofu_modules::EC2_USER_TAG_KEY;
use crate::commands::environment_commands::{prepare_onboarding, Ec2Choice};
t.e.tag_role(&user, EC2_USER_TAG_KEY, &c.user_tag).await;
{
let mut g = p5_lock(p5);
g.ec2_env = Some(name.clone());
g.ec2_okey = Some(okey.clone());
}
let api = crate::aws::environments::SdkOnboardApi::new(&admin_cfg);
let t0 = std::time::Instant::now();
let vars = prepare_onboarding(&api, &name, &region, &t.e.account, &owner, Some(Ec2Choice::Ec2(c.clone())))
.await
.unwrap_or_else(|e| panic!("the pre-apply onboarding step: {e}"));
eprintln!("RESULT pre-apply EC2 collision check (free names) passed in {:?}: {vars}", t0.elapsed());
vars
}
};
let t0 = std::time::Instant::now();
let (_, outcome, denied) = t
.run(
ModuleKind::UserOnboard,
Deployment::User { env: &name, owner_key: &okey },
&onboarding_vars(&name, &region, &owner),
&vars,
TofuOp::Apply,
t.keys_env(&keys),
)
.await;
assert_ok("onboard the workspace user", &outcome, &denied);
eprintln!("RESULT onboarding apply: {:?}", t0.elapsed());
let (_, exec_role) = user_role_names(&name, &okey);
// Test-only: pull from the throwaway private ECR repository.
@@ -2056,7 +2305,7 @@ mod environment {
let ws = caller_identity(&ws_cfg).await.expect("workspace user identity");
assert_eq!(ws.user_id, owner, "the profile session is the onboarded owner");
eprintln!("workspace user: {}", ws.arn);
Deployed { region, name, owner, okey, fs_token, conn, ws_cfg, ws }
Deployed { region, name, owner, okey, fs_token, conn, ws_cfg, ws, admin }
}
/// Phase 5: a Fargate workspace's whole life through `ProjectHost::Aws`,
@@ -2113,8 +2362,8 @@ log_stream_prefix, project_secret_name, task_family};
// Boxed: the body's state machine is far larger than a test
// thread's 2 MiB stack once it is moved into the timeout.
let inner = Box::pin(async {
let Deployed { region, name, owner, okey, fs_token, conn, ws_cfg, ws } =
deploy_for_workspace_user(&t, &p5, &image, 0).await;
let Deployed { region, name, owner, okey, fs_token, conn, ws_cfg, ws, .. } =
deploy_for_workspace_user(&t, &p5, &image, 0, None).await;
let mut settings = AppSettings::default();
settings.hosts = vec![conn];
@@ -2573,8 +2822,8 @@ log_stream_prefix, project_secret_name, task_family};
let p5 = std::sync::Mutex::new(Phase5 { ecr_repo, ..Default::default() });
let inner = Box::pin(async {
// ── 1. the environment carries the lifetime cap ──
let Deployed { region, name, owner: _, okey, fs_token: _, conn, ws_cfg, ws } =
deploy_for_workspace_user(&t, &p5, &image, 1).await;
let Deployed { region, name, okey, conn, ws_cfg, ws, .. } =
deploy_for_workspace_user(&t, &p5, &image, 1, None).await;
let env = crate::aws::runtime::read_environment(&SdkRuntimeApi::new(&ws_cfg), &name, &ws.account, &region)
.await
.expect("the environment record, as the workspace user");
@@ -2921,4 +3170,596 @@ log_stream_prefix, project_secret_name, task_family};
})
.await;
}
// ─────────────────────────────────────────────── phase 7: ECS on EC2
/// The phase-7 body: two cold starts, a scale-in wait (managed scaling
/// takes ~15 minutes to remove an idle instance) and three applies.
const PHASE7_TIMEOUT: Duration = Duration::from_secs(110 * 60);
/// A group's (instance count, desired capacity, `id:state` list) as the
/// account admin; `None` when it does not exist.
async fn group_size(asg: &aws_sdk_autoscaling::Client, name: &str) -> Option<(usize, i32, Vec<String>)> {
let out = asg.describe_auto_scaling_groups().auto_scaling_group_names(name).send().await.ok()?;
let g = out.auto_scaling_groups().first()?;
let ids = g
.instances()
.iter()
.map(|i| format!("{}:{}", i.instance_id().unwrap_or_default(), i.lifecycle_state().map(|s| s.as_str()).unwrap_or("?")))
.collect();
Some((g.instances().len(), g.desired_capacity().unwrap_or(-1), ids))
}
/// Instances still tagged with `env` and not terminated, as the account admin.
async fn live_env_instances(region: &str, env: &str) -> Vec<String> {
let (ok_, out, err) = aws_cli(
None,
region,
&[
"ec2", "describe-instances",
"--filters", &format!("Name=tag:triple-c:environment,Values={env}"),
"Name=instance-state-name,Values=pending,running,stopping,stopped,shutting-down",
"--query", "Reservations[].Instances[].[InstanceId,State.Name]",
],
)
.await;
assert!(ok_, "describe-instances: {err}");
let rows: Vec<Vec<String>> = serde_json::from_str(&out).unwrap_or_default();
rows.into_iter().map(|r| r.join(":")).collect()
}
/// Phase 7: an EC2 project's whole life on the user's own capacity
/// (ECS on EC2, bridge networking, scale to zero), as a role holding
/// only the workspace files and carrying the `triple-c-user` principal
/// tag (a role tag: an assumed-role session's principal tags), onboarded
/// with `m7i.large`, at most 1 instance, by the environment admin
/// holding all four admin files. Run with `scripts/aws-e2e-workspace.sh`
/// and `TRIPLEC_E2E_TEST=aws::live_tests::environment::a_phase7_ec2_workspace_round_trips`.
///
/// Covers: the onboarding's resources and its pre-apply collision check
/// (and its refusal for another owner); admin-tier probes of the
/// scaling file (`$Latest`, a pinned control, an unsupported type in a
/// template version, a capacity provider over a non-Triple-C group);
/// a cold start with Docker access, VPN and the sandbox on (instance
/// launch, ECS Exec port forward into the bridge-mode task), `docker
/// ps`, a WireGuard/TUN interface, `src_valid_mark`, bubblewrap, IMDS
/// and the agent's introspection port from the container; RunTask on
/// another user's cluster (the wrong-triple-c-user message); stop and
/// the scale-in to zero; a second start; removal; re-applying the
/// onboarding without EC2 (drain and delete with an instance still up);
/// offboarding and the environment's destroy, all as the admin tier.
/// Its guard (`ec2_cleanup`) force-deletes whatever EC2 capacity is
/// left. About one m7i.large instance-hour.
#[tokio::test]
#[ignore]
async fn a_phase7_ec2_workspace_round_trips() {
use crate::aws::environments::{OnboardApi, SdkOnboardApi, UserCompute, UserRecord};
use crate::aws::runtime::{read_user_record, SdkRuntimeApi};
use crate::aws::tasks::{ec2_run_error, ec2_run_task_request, SdkTaskApi, TaskApi, EC2_TAG_MISMATCH};
use crate::aws::tofu_modules::{user_ec2_stem, user_instance_role_name, Ec2Compute};
use crate::aws::workspace::workspace_tags;
use crate::commands::environment_commands::{prepare_onboarding, Ec2Choice};
use crate::host::aws_host::PROGRESS_INSTANCE;
use crate::host::lifecycle::ProjectHost;
use crate::models::app_settings::ImageSource;
use crate::models::project::{AwsPlacement, AwsWorkspaceConfig, LaunchType, StorageMode};
use crate::models::{AppSettings, Project};
use crate::storage::projects_store::ProjectsStore;
let image = super::var("TRIPLEC_E2E_IMAGE");
let ecr_repo = std::env::var("TRIPLEC_E2E_ECR_REPO").ok().filter(|r| !r.is_empty());
env_e2e(|t| async move {
let p5 = std::sync::Mutex::new(Phase5 { ecr_repo, ..Default::default() });
let inner = Box::pin(async {
// TRIPLEC_E2E_PHASE7_SHORT=1: the life cycle only (no probes, no
// in-container checks or isolation, no scale-in wait; the second
// start is then warm, on the instance still up).
let quick = std::env::var("TRIPLEC_E2E_PHASE7_SHORT").is_ok_and(|v| v == "1");
let short = uuid::Uuid::new_v4().simple().to_string();
let user_tag = format!("e2e{}", &short[..6]);
let compute = Ec2Compute { instance_type: "m7i.large".into(), max_instances: 1, user_tag: user_tag.clone() };
let Deployed { region, name, owner, okey, conn, ws_cfg, ws, admin, .. } =
deploy_for_workspace_user(&t, &p5, &image, 0, Some(&compute)).await;
let account = t.e.account.clone();
let stem = user_ec2_stem(&name, &user_tag);
let asg_admin = aws_sdk_autoscaling::Client::new(&t.e.cfg);
let ecs_admin = aws_sdk_ecs::Client::new(&t.e.cfg);
// ── 1. what the onboarding made ──
let groups = ok("describe the group", asg_admin.describe_auto_scaling_groups().auto_scaling_group_names(&stem).send().await);
let group = groups.auto_scaling_groups().first().expect("the user's Auto Scaling group").clone();
let subnets = group.vpc_zone_identifier().unwrap_or_default().to_string();
let group_tags: Vec<(String, String)> = group
.tags()
.iter()
.map(|t| (t.key().unwrap_or_default().to_string(), t.value().unwrap_or_default().to_string()))
.collect();
eprintln!(
"RESULT group {stem}: min {:?} max {:?} desired {:?}, template {:?}, protect {:?}, tags {group_tags:?}",
group.min_size(),
group.max_size(),
group.desired_capacity(),
group.launch_template().map(|l| (l.launch_template_name(), l.version())),
group.new_instances_protected_from_scale_in()
);
assert_eq!((group.min_size(), group.max_size()), (Some(0), Some(1)));
let cl = ok(
"describe the user's cluster",
ecs_admin.describe_clusters().clusters(&stem).include(aws_sdk_ecs::types::ClusterField::Tags).send().await,
);
let cluster = cl.clusters().first().expect("the user's cluster");
eprintln!(
"RESULT cluster {stem}: {:?}, providers {:?}, default strategy {:?}, tags {:?}",
cluster.status(),
cluster.capacity_providers(),
cluster.default_capacity_provider_strategy().iter().map(|s| s.capacity_provider()).collect::<Vec<_>>(),
cluster.tags().iter().map(|t| (t.key().unwrap_or_default(), t.value().unwrap_or_default())).collect::<Vec<_>>()
);
let cps = ok("describe the capacity provider", ecs_admin.describe_capacity_providers().capacity_providers(&stem).send().await);
let cp = cps.capacity_providers().first().expect("the capacity provider");
eprintln!(
"RESULT capacity provider {stem}: {:?}, managed scaling {:?}, termination protection {:?}",
cp.status(),
cp.auto_scaling_group_provider().and_then(|a| a.managed_scaling()).map(|m| (m.status(), m.target_capacity())),
cp.auto_scaling_group_provider().map(|a| a.managed_termination_protection())
);
let (_, lt, _) = aws_cli(
None,
&region,
&["ec2", "describe-launch-template-versions", "--launch-template-name", &stem, "--query",
"LaunchTemplateVersions[].[VersionNumber,LaunchTemplateData.InstanceType,LaunchTemplateData.ImageId,LaunchTemplateData.MetadataOptions]"],
)
.await;
eprintln!("RESULT launch template {stem}: {}", lt.split_whitespace().collect::<Vec<_>>().join(" "));
let role_name = user_instance_role_name(&name, &okey);
let prof = ok("get the instance profile", t.e.iam.get_instance_profile().instance_profile_name(&role_name).send().await);
eprintln!(
"RESULT instance profile {role_name}: roles {:?}",
prof.instance_profile().map(|p| p.roles().iter().map(|r| r.role_name().to_string()).collect::<Vec<_>>())
);
let record = read_user_record(&SdkRuntimeApi::new(&ws_cfg), &name, &okey, &account, &region).await.expect("read the user record");
eprintln!("RESULT user record, as the workspace user: {record:?}");
let Some(rec) = record.valid().cloned() else { panic!("the user record is not valid: {record:?}") };
assert_eq!((rec.instance_type.as_str(), rec.max_instances, rec.user_tag.as_str()), ("m7i.large", 1, user_tag.as_str()));
assert!(rec.cluster_arn.ends_with(&format!(":cluster/{stem}")), "{}", rec.cluster_arn);
// ── 1b / 6. the collision check's reads, and its refusal ──
let (akeys, acfg, _) = t.session(&admin).await;
let oapi = SdkOnboardApi::new(&acfg);
let held = oapi.cluster(&stem).await.expect("DescribeClusters include=TAGS as the admin tier");
eprintln!("RESULT DescribeClusters include=TAGS as the environment admin: {held:?}");
let held = held.expect("the cluster is found");
assert!(held.tags.iter().any(|(k, v)| k == "triple-c:owner" && *v == owner), "{held:?}");
let missing = oapi.cluster(&format!("{stem}zz")).await;
eprintln!("RESULT DescribeClusters on a missing name: {missing:?}");
assert!(matches!(missing, Ok(None)), "{missing:?}");
let missing_group = oapi.group(&format!("{stem}zz")).await;
eprintln!("RESULT DescribeAutoScalingGroups on a missing name: {missing_group:?}");
assert!(matches!(missing_group, Ok(None)), "{missing_group:?}");
let refused = prepare_onboarding(&oapi, &name, &region, &account, "AROAEXAMPLEOTHER0000:bob", Some(Ec2Choice::Ec2(compute.clone()))).await;
eprintln!("RESULT onboarding another owner with the same triple-c-user value: {refused:?}");
let refused = refused.expect_err("another owner's onboarding with the same tag is refused");
assert!(refused.to_string().contains("another onboarding"), "{refused}");
if !quick {
// ── 6. the scaling file's open points, as the environment admin ──
let asg_env = aws_sdk_autoscaling::Client::new(&acfg);
let ecs_env = aws_sdk_ecs::Client::new(&acfg);
let probe_tags = |n: &str| -> Vec<aws_sdk_autoscaling::types::Tag> {
group_tags
.iter()
.filter(|(k, _)| k.starts_with("triple-c:"))
.map(|(k, v)| {
aws_sdk_autoscaling::types::Tag::builder()
.key(k)
.value(v)
.resource_id(n)
.resource_type("auto-scaling-group")
.propagate_at_launch(false)
.build()
})
.collect()
};
let create_group = |client: aws_sdk_autoscaling::Client, n: String, version: String, tags: Vec<aws_sdk_autoscaling::types::Tag>| {
let subnets = subnets.clone();
let lt_name = stem.clone();
async move {
client
.create_auto_scaling_group()
.auto_scaling_group_name(&n)
.launch_template(
aws_sdk_autoscaling::types::LaunchTemplateSpecification::builder()
.launch_template_name(lt_name)
.version(version)
.build(),
)
.min_size(0)
.max_size(0)
.desired_capacity(0)
.vpc_zone_identifier(subnets)
.set_tags(Some(tags))
.send()
.await
.map(|_| ())
.map_err(|e| classify_sdk_error("autoscaling:CreateAutoScalingGroup", &e).to_string())
}
};
let delete_group = |n: String| {
let client = asg_admin.clone();
async move {
let _ = client.delete_auto_scaling_group().auto_scaling_group_name(&n).force_delete(true).send().await;
}
};
// $Latest must be denied (LaunchTemplateVersionSpecified).
let pa = format!("{stem}-pa");
let r = create_group(asg_env.clone(), pa.clone(), "$Latest".into(), probe_tags(&pa)).await;
eprintln!("RESULT probe: CreateAutoScalingGroup with $Latest as the environment admin: {r:?}");
if r.is_ok() {
eprintln!("RESULT WARNING: $Latest was allowed");
delete_group(pa).await;
}
// The control: the same create pinned to version 1 is allowed
// (InstanceTypes filled from the template).
let pb = format!("{stem}-pb");
let r = create_group(asg_env.clone(), pb.clone(), "1".into(), probe_tags(&pb)).await;
eprintln!("RESULT probe: the same create pinned to version 1 (control): {r:?}");
if r.is_ok() {
let d = asg_env.delete_auto_scaling_group().auto_scaling_group_name(&pb).send().await;
eprintln!("RESULT probe: DeleteAutoScalingGroup of it as the environment admin: {:?}", d.map(|_| ()).map_err(|e| classify_sdk_error("autoscaling:DeleteAutoScalingGroup", &e).to_string()));
delete_group(pb).await;
}
// A template version naming an unsupported type, then a group pinned to it.
let (made, out, err) = aws_cli(
Some(&akeys),
&region,
&["ec2", "create-launch-template-version", "--launch-template-name", &stem, "--source-version", "1",
"--launch-template-data", r#"{"InstanceType":"p5.48xlarge"}"#, "--query", "LaunchTemplateVersion.VersionNumber"],
)
.await;
eprintln!("RESULT probe: CreateLaunchTemplateVersion with p5.48xlarge as the environment admin: ok {made} {} {}", out.trim(), err.trim());
if made {
let pc = format!("{stem}-pc");
let r = create_group(asg_env.clone(), pc.clone(), out.trim().to_string(), probe_tags(&pc)).await;
eprintln!("RESULT probe: CreateAutoScalingGroup pinned to the p5.48xlarge version: {r:?}");
if r.is_ok() {
eprintln!("RESULT WARNING: a group of an unsupported type was allowed");
delete_group(pc).await;
}
}
// A Triple-C capacity provider over a group that isn't Triple-C's.
let foreign_group = format!("e2e-untagged-{}", &short[..10]);
p5_lock(&p5).probe_groups.push(foreign_group.clone());
let r = create_group(asg_admin.clone(), foreign_group.clone(), "1".into(), Vec::new()).await;
eprintln!("RESULT probe: account admin creates the untagged group {foreign_group}: {r:?}");
if r.is_ok() {
let arn = ok("describe the untagged group", asg_admin.describe_auto_scaling_groups().auto_scaling_group_names(&foreign_group).send().await)
.auto_scaling_groups()
.first()
.and_then(|g| g.auto_scaling_group_arn().map(str::to_string))
.unwrap();
let pd = format!("{stem}-pd");
let r = ecs_env
.create_capacity_provider()
.name(&pd)
.auto_scaling_group_provider(
aws_sdk_ecs::types::AutoScalingGroupProvider::builder()
.auto_scaling_group_arn(arn)
.managed_termination_protection(aws_sdk_ecs::types::ManagedTerminationProtection::Disabled)
.build()
.unwrap(),
)
.set_tags(Some(
group_tags
.iter()
.filter(|(k, _)| k.starts_with("triple-c:"))
.map(|(k, v)| aws_sdk_ecs::types::Tag::builder().key(k).value(v).build())
.collect(),
))
.send()
.await;
eprintln!(
"RESULT probe: CreateCapacityProvider {pd} over the untagged group, as the environment admin: {:?}",
r.as_ref().map(|_| ()).map_err(|e| classify_sdk_error("ecs:CreateCapacityProvider", e).to_string())
);
if r.is_ok() {
eprintln!("RESULT WARNING: a Triple-C capacity provider over a foreign group was allowed");
let _ = ecs_admin.delete_capacity_provider().capacity_provider(&pd).send().await;
}
delete_group(foreign_group).await;
}
}
// ── 2. start 1: cold, with an instance launch ──
let mut settings = AppSettings::default();
settings.hosts = vec![conn];
settings.image_source = ImageSource::Custom;
settings.custom_image_name = Some(image.clone());
let mut project = Project::new("e2e-ec2".into(), Vec::new());
project.host_id = Some("e2e-wsuser".into());
project.aws_placement = Some(AwsPlacement { region: region.clone(), environment: name.clone() });
project.allow_docker_access = true;
project.vpn_support_enabled = true;
project.sandbox_mode_enabled = true;
project.aws = Some(AwsWorkspaceConfig {
launch_type: LaunchType::Ec2,
cpu: 1024,
memory_mib: 4096,
storage: StorageMode::Hybrid,
wip_push_on_stop: true,
..Default::default()
});
let id = project.id.clone();
{
let mut g = p5_lock(&p5);
g.project_id = Some(id.clone());
g.secrets.push(crate::aws::tofu_modules::project_secret_name(&name, &okey, &id));
}
let store_dir = tempfile::tempdir().unwrap();
let store = ProjectsStore::over_file_for_test(store_dir.path().join("projects.json"), vec![project.clone()]);
let host = ProjectHost::for_project(&project);
let current = |store: &ProjectsStore| store.get(&id).unwrap();
let api = SdkTaskApi::new(&ws_cfg);
let steps: Arc<std::sync::Mutex<Vec<(Duration, String)>>> = Default::default();
let t0 = std::sync::Mutex::new(std::time::Instant::now());
let progress = |m: &str| {
let at = t0.lock().unwrap().elapsed();
eprintln!(" progress +{at:?}: {m}");
steps.lock().unwrap().push((at, m.to_string()));
};
let started = host.ensure_started(&current(&store), &settings, &store, &progress).await;
if let Err(err) = &started {
// Debugging aid: hold the workspace for an admin to look
// at before the guard tears it down.
let pause: u64 = std::env::var("TRIPLEC_E2E_PAUSE_SECS").ok().and_then(|v| v.parse().ok()).unwrap_or(0);
if pause > 0 {
eprintln!("RESULT start 1 failed ({err}); pausing {pause}s for inspection (task {:?})", current(&store).aws.and_then(|a| a.task_arn));
tokio::time::sleep(Duration::from_secs(pause)).await;
}
}
let target = started.expect("EC2 start 1");
let first_start = t0.lock().unwrap().elapsed();
let s1: Vec<(Duration, String)> = std::mem::take(&mut *steps.lock().unwrap());
eprintln!("RESULT EC2 start 1 (cold, instance launch): {first_start:?}; steps {s1:?}");
assert!(s1.iter().any(|(_, m)| m == PROGRESS_INSTANCE), "no \"Starting an instance\" step: {s1:?}");
let aws1 = current(&store).aws.unwrap();
let task1 = aws1.task_arn.clone().expect("the task ARN is stored");
let revision1 = aws1.task_definition_arn.clone().expect("the revision is stored");
assert!(task1.contains(&format!(":task/{stem}/")), "the task runs on the user's cluster: {task1}");
let d = api.describe(&rec.cluster_arn, std::slice::from_ref(&task1)).await.expect("describe task 1");
eprintln!("RESULT task 1: {:?}", d.first().map(|t| (&t.last_status, &t.container_instance_arn)));
eprintln!("RESULT group after start 1: {:?}", group_size(&asg_admin, &stem).await);
let td = ok("describe revision 1", ecs_admin.describe_task_definition().task_definition(&revision1).send().await);
let td = td.task_definition().unwrap();
eprintln!(
"RESULT revision 1: network {:?}, compat {:?}, volumes {:?}",
td.network_mode(),
td.requires_compatibilities(),
td.volumes().iter().map(|v| v.name().unwrap_or_default()).collect::<Vec<_>>()
);
// The cold start's first (ECS Exec off) task is gone: exactly one
// live task of the project on the user's cluster.
let family = crate::aws::tofu_modules::task_family(&name, &okey, &id);
let live = api.list_project_tasks(&rec.cluster_arn, &family).await.expect("list the project's tasks");
eprintln!("RESULT live tasks on the user's cluster after start 1: {live:?}");
assert_eq!(live, vec![task1.clone()], "exactly one live task");
// ── through the agent (ECS Exec port forward into the bridge task) ──
let (out, code) = sh(&target, "claude", "echo hello-from-ec2; id").await;
eprintln!("RESULT agent tunnel over ECS Exec into the bridge-mode task: exit {code}: {}", out.trim());
assert_eq!(code, 0, "{out}");
// awslogs on EC2 writes with the execution role (the instance
// role has no log rights), into the user's own streams only.
let env_rec = crate::aws::runtime::read_environment(&SdkRuntimeApi::new(&ws_cfg), &name, &account, &region)
.await
.expect("the environment record, as the workspace user");
let stream = crate::aws::tasks::task_log_stream(&crate::aws::tofu_modules::log_stream_prefix(&okey, &id), &task1);
let log1 = api.log_tail(&env_rec.log_group, &stream, 20).await.expect("read task 1's log");
eprintln!("RESULT task 1 log ({stream}): {} lines, last {:?}", log1.len(), log1.last());
assert!(!log1.is_empty(), "task 1 wrote no log lines: awslogs could not write with the execution role");
// E8: the sandbox is forced off in the task's settings.
let (settings_json, code) = sh(&target, "claude", "printenv CLAUDE_CODE_SETTINGS_JSON").await;
assert_eq!(code, 0, "no CLAUDE_CODE_SETTINGS_JSON");
let v: serde_json::Value = serde_json::from_str(settings_json.trim()).expect("settings JSON");
eprintln!("RESULT sandbox in the task's Claude Code settings: {}", v["sandbox"]);
assert_eq!(v["sandbox"]["enabled"], false, "the sandbox is off on EC2");
let (mut docker_ok, mut vpn_ok, mut src_valid_mark, mut bwrap_ok) = (true, true, true, false);
if !quick {
let (out, code) = sh(&target, "claude", "ls -l /var/run/docker.sock; docker ps 2>&1; docker version --format '{{.Server.Version}}' 2>&1").await;
eprintln!("RESULT docker ps as claude (exit {code}):\n{}", out.trim());
docker_ok = code == 0;
let (out, code) = sh(
&target,
"root",
"ip link add wg0 type wireguard && ip link show wg0 && ip link del wg0 && echo WG-OK; \
ip tuntap add tun0 mode tun && ip link show tun0 && ip link del tun0 && echo TUN-OK; \
ls -l /dev/net/tun; sysctl net.ipv4.conf.all.src_valid_mark; grep CapEff /proc/self/status",
)
.await;
eprintln!("RESULT VPN pieces as root (exit {code}):\n{}", out.trim());
let (sudo_out, _) = sh(&target, "claude", "sudo -n ip link add wg1 type wireguard && sudo -n ip link del wg1 && echo SUDO-WG-OK 2>&1").await;
eprintln!("RESULT wireguard via sudo as claude: {}", sudo_out.trim());
vpn_ok = out.contains("WG-OK") || out.contains("TUN-OK");
src_valid_mark = out.contains("src_valid_mark = 1");
let (bwrap, _) = sh(
&target,
"claude",
"bwrap --unshare-user --ro-bind / / true 2>&1; echo \"unshare-user rc=$?\"; \
bwrap --ro-bind / / --dev /dev --proc /proc --unshare-all --die-with-parent true 2>&1; echo \"unshare-all rc=$?\"; \
cat /proc/sys/user/max_user_namespaces 2>&1; uname -r",
)
.await;
eprintln!("RESULT E8 bwrap on EC2:\n{}", bwrap.trim());
bwrap_ok = bwrap.contains("unshare-user rc=0");
let (imds, _) = sh(
&target,
"claude",
"tok=$(curl -s -m 3 -X PUT http://169.254.169.254/latest/api/token -H 'X-aws-ec2-metadata-token-ttl-seconds: 60'); \
echo \"token rc=$? len=${#tok}\"; \
curl -s -m 3 -o /dev/null -w 'imdsv1 http=%{http_code}\\n' http://169.254.169.254/latest/meta-data/; \
gw=$(ip route | awk '/default/ {print $3}'); echo \"gateway $gw\"; \
for h in \"$gw\" 127.0.0.1; do curl -s -m 3 -o /dev/null -w \"introspection $h:51678 http=%{http_code}\\n\" http://$h:51678/v1/metadata; done; \
curl -s -m 3 http://$gw:51678/v1/metadata | head -c 300; echo; \
test -n \"$AWS_CONTAINER_CREDENTIALS_RELATIVE_URI\" && curl -s -m 3 http://169.254.170.2$AWS_CONTAINER_CREDENTIALS_RELATIVE_URI | grep -q AccessKeyId && echo TASK-ROLE-CREDS-OK || echo task-role-creds-missing",
)
.await;
eprintln!("RESULT IMDS / introspection from the container:\n{}", imds.trim());
assert!(imds.contains("len=0"), "the container got an IMDS token: {imds}");
// ── 4. isolation: RunTask on another triple-c-user's cluster ──
let other = format!("e2eother{}", &short[6..10]);
let other_cluster = user_ec2_stem(&name, &other);
let made = ok("admin creates another user's (empty) cluster", ecs_admin.create_cluster().cluster_name(&other_cluster).send().await);
let other_arn = made.cluster().and_then(|c| c.cluster_arn()).unwrap().to_string();
let fake = UserCompute {
instance_type: "m7i.large".into(),
max_instances: 1,
user_tag: other.clone(),
cluster_arn: other_arn.clone(),
capacity_provider: other_cluster.clone(),
};
let r = api.run_task(ec2_run_task_request(&fake, &revision1, &workspace_tags(&owner, &id, &name))).await.map_err(ec2_run_error);
eprintln!("RESULT RunTask on {other_cluster} as triple-c-user {user_tag}: {r:?}");
match r {
Err(crate::aws::error::AwsError::AccessDenied { message, .. }) => assert!(message.contains(EC2_TAG_MISMATCH), "{message}"),
other => panic!("RunTask on another user's cluster: expected AccessDenied, got {other:?}"),
}
let _ = ecs_admin.delete_cluster().cluster(&other_arn).send().await;
}
// ── 5. stop 1, then the scale-in to zero ──
let t2 = std::time::Instant::now();
host.stop_project(&current(&store), &settings).await.expect("stop 1");
poll(Duration::from_secs(300), Duration::from_secs(5), || async {
let d = api.describe(&rec.cluster_arn, std::slice::from_ref(&task1)).await.ok()?;
(d.first()?.last_status == "STOPPED").then_some(())
})
.await
.expect("task 1 never stopped");
let stopped_after = t2.elapsed();
eprintln!("RESULT stop 1: STOPPED after {stopped_after:?}");
store.set_aws_runtime(&id, None, aws1.access_point_id.clone(), Some(revision1.clone()), None).unwrap();
let mut scale_in = Duration::ZERO;
if !quick {
let mut last: Option<(usize, i32)> = None;
let mut scaled = None;
let deadline = std::time::Instant::now() + Duration::from_secs(40 * 60);
while std::time::Instant::now() < deadline {
let now = group_size(&asg_admin, &stem).await;
let key = now.as_ref().map(|n| (n.0, n.1));
if key != last {
eprintln!(" group +{:?} after STOPPED: {now:?}", t2.elapsed() - stopped_after);
last = key;
}
if key == Some((0, 0)) {
scaled = Some(());
break;
}
tokio::time::sleep(Duration::from_secs(30)).await;
}
scale_in = t2.elapsed() - stopped_after;
eprintln!("RESULT scale-in to 0 instances: {scaled:?} after {:.1} min from STOPPED", scale_in.as_secs_f64() / 60.0);
assert!(scaled.is_some(), "the group did not scale back to zero within 40 minutes");
}
// ── start 2 ──
*t0.lock().unwrap() = std::time::Instant::now();
let target2 = host.ensure_started(&current(&store), &settings, &store, &progress).await.expect("EC2 start 2");
let second_start = t0.lock().unwrap().elapsed();
let s2: Vec<(Duration, String)> = std::mem::take(&mut *steps.lock().unwrap());
eprintln!("RESULT EC2 start 2 ({}): {second_start:?}; steps {s2:?}", if quick { "warm: right after stop 1" } else { "after scale-in" });
let (out, code) = sh(&target2, "claude", "echo again; docker ps -q | wc -l").await;
assert_eq!(code, 0, "{out}");
let aws2 = current(&store).aws.unwrap();
let task2 = aws2.task_arn.clone().unwrap();
host.stop_project(&current(&store), &settings).await.expect("stop 2");
poll(Duration::from_secs(300), Duration::from_secs(5), || async {
let d = api.describe(&rec.cluster_arn, std::slice::from_ref(&task2)).await.ok()?;
(d.first()?.last_status == "STOPPED").then_some(())
})
.await
.expect("task 2 never stopped");
store.set_aws_runtime(&id, None, aws2.access_point_id.clone(), aws2.task_definition_arn.clone(), None).unwrap();
let removed = host.remove_aws_resources(&current(&store), &settings, true).await;
eprintln!("RESULT remove_aws_resources: {:?}", removed.as_ref().map_err(|e| e.chars().take(400).collect::<String>()));
removed.expect("remove the project's resources");
// ── 7. the onboarding re-applied without EC2, an instance still up ──
eprintln!("RESULT group before the re-apply without EC2: {:?}", group_size(&asg_admin, &stem).await);
let (keys, acfg, _) = t.session(&admin).await;
let vars = prepare_onboarding(&SdkOnboardApi::new(&acfg), &name, &region, &account, &owner, Some(Ec2Choice::None))
.await
.expect("the pre-apply step without EC2");
let t5 = std::time::Instant::now();
let (_, outcome, denied) = t
.run(ModuleKind::UserOnboard, Deployment::User { env: &name, owner_key: &okey }, &vars, TofuOp::Apply, t.keys_env(&keys))
.await;
eprintln!("RESULT re-apply without EC2: {:?} in {:?}", outcome.summary, t5.elapsed());
assert_ok("re-apply the onboarding without EC2", &outcome, &denied);
let g = group_size(&asg_admin, &stem).await;
let inst = live_env_instances(&region, &name).await;
let cps = ecs_admin.describe_capacity_providers().capacity_providers(&stem).send().await;
let cl = ecs_admin.describe_clusters().clusters(&stem).send().await;
let (_, lt, lt_err) = aws_cli(None, &region, &["ec2", "describe-launch-templates", "--launch-template-names", &stem]).await;
let prof = t.e.iam.get_instance_profile().instance_profile_name(&role_name).send().await;
let role = t.e.iam.get_role().role_name(&role_name).send().await;
let record = read_user_record(&SdkRuntimeApi::new(&ws_cfg), &name, &okey, &account, &region).await;
eprintln!(
"RESULT after the re-apply: group {g:?}; live instances {inst:?}; provider {:?}; cluster {:?}; template {} {}; profile gone {}; role gone {}; record {record:?}",
cps.as_ref().map(|o| o.capacity_providers().iter().map(|c| c.status().cloned()).collect::<Vec<_>>()),
cl.as_ref().map(|o| o.clusters().iter().map(|c| c.status().map(str::to_string)).collect::<Vec<_>>()),
lt.trim().chars().take(100).collect::<String>(),
lt_err.trim().chars().take(100).collect::<String>(),
prof.is_err(),
role.is_err(),
);
assert!(g.is_none(), "the group is gone");
assert!(inst.is_empty(), "no instance is left: {inst:?}");
assert!(prof.is_err() && role.is_err(), "the instance profile and role are gone");
assert!(matches!(record, Ok(UserRecord::Absent)), "{record:?}");
// ── offboard, then destroy the environment, as the environment admin ──
if let Some(role) = p5_lock(&p5).exec_role.take() {
let _ = t.e.iam.delete_role_policy().role_name(&role).policy_name(ECR_POLICY).send().await;
}
let (keys, _, _) = t.session(&admin).await;
let (_, outcome, denied) = t
.run(ModuleKind::UserOnboard, Deployment::User { env: &name, owner_key: &okey }, &vars, TofuOp::Destroy, t.keys_env(&keys))
.await;
assert_ok("offboard", &outcome, &denied);
let (_, outcome, denied) = t
.run(
ModuleKind::Environment,
Deployment::Environment(&name),
&environment_vars(&name, &region, &NetworkChoice::Public, 0).unwrap(),
TofuOp::Destroy,
t.keys_env(&keys),
)
.await;
assert_ok("destroy the environment", &outcome, &denied);
eprintln!(
"RESULT summary: docker {docker_ok}, vpn {vpn_ok}, src_valid_mark {src_valid_mark}, bwrap {bwrap_ok}, start1 {first_start:?}, scale-in {:.1} min, start2 {second_start:?}",
scale_in.as_secs_f64() / 60.0
);
let _ = ws;
assert!(docker_ok, "docker ps failed as claude");
// E8, stable in every run: Docker's seccomp profile refuses it.
assert!(quick || !bwrap_ok, "bwrap worked in the task: revisit sandbox_allowed (E8)");
assert!(vpn_ok && src_valid_mark, "the VPN pieces are missing");
});
let body = async {
if tokio::time::timeout(PHASE7_TIMEOUT, inner).await.is_err() {
panic!("the phase-7 body did not finish in {PHASE7_TIMEOUT:?}");
}
};
let result = std::panic::AssertUnwindSafe(body).catch_unwind().await;
let guard = std::mem::take(&mut *p5_lock(&p5));
phase5_cleanup(&t.e, &guard).await;
if let Err(panic) = result {
std::panic::resume_unwind(panic);
}
})
.await;
}
}
+15 -1
View File
@@ -37,8 +37,20 @@ impl ContextEntry {
pub fn is_list(&self) -> bool {
self.values.len() > 1
}
/// Keys IAM types as Boolean (a `Bool` condition does not match a
/// String-typed simulation value).
pub fn is_boolean(&self) -> bool {
BOOLEAN_KEYS.contains(&self.key.as_str())
}
}
/// The Boolean condition keys the checks send.
const BOOLEAN_KEYS: &[&str] = &[
"autoscaling:LaunchTemplateVersionSpecified",
"ec2:IsLaunchTemplateResource",
];
/// Builds the simulation for `check`: the resource and every literal context
/// value rendered with `p`, `OwnerId` as `user_id`, and `aws:userid` =
/// `user_id` added. A key appearing more than once becomes one entry holding
@@ -175,7 +187,9 @@ impl PermissionApi for SdkPermissionApi {
SdkEntry::builder()
.context_key_name(&e.key)
.set_context_key_values(Some(e.values.clone()))
.context_key_type(if e.is_list() {
.context_key_type(if e.is_boolean() {
ContextKeyTypeEnum::Boolean
} else if e.is_list() {
ContextKeyTypeEnum::StringList
} else {
ContextKeyTypeEnum::String
@@ -344,6 +344,29 @@ fn a_repeated_context_key_is_one_string_list_entry() {
);
assert!(req.context[0].is_list());
assert!(!req.context[1].is_list());
assert!(!req.context[1].is_boolean());
}
/// Every Boolean-typed key a check sends is one `is_boolean` knows, so the
/// simulator gets it as Boolean (a `Bool` condition never matches a String).
#[test]
fn boolean_context_keys_are_sent_as_booleans() {
let mut seen = 0;
for tier in [Tier::WorkspaceUser, Tier::EnvironmentAdmin] {
for c in checks(tier) {
for (key, value) in c.context {
if matches!(value, ContextValue::Literal("true" | "false")) {
let e = ContextEntry {
key: key.to_string(),
values: vec!["true".into()],
};
assert!(e.is_boolean(), "{key} is sent as a String");
seen += 1;
}
}
}
}
assert!(seen >= 2, "only {seen} Boolean keys");
}
#[test]
File diff suppressed because it is too large. Load diff
+34 -2
View File
@@ -208,10 +208,11 @@ pub(crate) fn conditions_hold(stmt: &Value, req: &Request) -> bool {
// A negated operator holds when the key is absent, and
// otherwise when no value matches.
let (matches, negated): (fn(&str, &str) -> bool, bool) = match base {
"StringEquals" => (|p, v| p == v, false),
// IAM's Bool compares "true"/"false"; the checks send strings.
"StringEquals" | "Bool" => (|p, v| p == v, false),
"StringLike" | "ArnLike" => (glob, false),
"StringNotEquals" => (|p, v| p == v, true),
"StringNotLike" => (glob, true),
"StringNotLike" | "ArnNotLike" => (glob, true),
other => panic!("{}: unsupported condition operator {other}", sid(stmt)),
};
entries
@@ -268,6 +269,37 @@ mod tests {
assert!(!glob(&literal, "arn:aws:ec2:us-west-2:1:elastic-ip/eipalloc-1"));
}
#[test]
fn bool_needs_the_key_and_bool_if_exists_does_not() {
let stmt = |op: &str| -> Value {
serde_json::json!({
"Sid": "T", "Effect": "Allow", "Action": "x:Y", "Resource": "*",
"Condition": { op: { "k:Pinned": "true" } }
})
};
let holds = |op: &str, context| {
let c = Check {
action: "x:Y",
resource: "*",
context,
write: true,
};
conditions_hold(
&stmt(op),
&Request {
check: &c,
caller: ME,
owner: ME,
},
)
};
assert!(holds("Bool", &[("k:Pinned", Literal("true"))]));
assert!(!holds("Bool", &[("k:Pinned", Literal("false"))]));
assert!(!holds("Bool", &[]));
assert!(holds("BoolIfExists", &[]));
assert!(!holds("BoolIfExists", &[("k:Pinned", Literal("false"))]));
}
#[test]
fn for_all_values_is_vacuous_on_an_absent_key() {
let stmt: Value = serde_json::json!({
+326 -47
View File
@@ -1,6 +1,9 @@
//! The estimated hourly cost of a Fargate workspace (Q5): on-demand rates
//! from the Price List API (`pricing:GetProducts`, service `AmazonECS`,
//! served from us-east-1 only), cached on disk per region for a day.
//! served from us-east-1 only), cached on disk per region for a day. An EC2
//! workspace (phase 7 E10) is priced as its instance: the `AmazonEC2`
//! on-demand Linux, shared-tenancy price of the user's instance type, cached
//! per region and type the same way.
//!
//! Everything here is best effort. A denied call, no network, a slow answer
//! or a price list we can't read all end in "no estimate" (`None`), never an
@@ -13,6 +16,7 @@ use std::path::{Path, PathBuf};
use std::sync::Mutex;
use std::time::{Duration, SystemTime, UNIX_EPOCH};
use serde::de::DeserializeOwned;
use serde::{Deserialize, Serialize};
use serde_json::Value;
@@ -22,6 +26,8 @@ use crate::models::project::{Arch, AwsWorkspaceConfig, StorageMode};
/// eu-central-1; any of them serves every region's prices).
pub const PRICING_REGION: &str = "us-east-1";
pub const SERVICE_CODE: &str = "AmazonECS";
/// EC2 instances' prices (phase 7 E10).
pub const EC2_SERVICE_CODE: &str = "AmazonEC2";
/// How long a region's rates are used before they are fetched again.
pub const CACHE_TTL: Duration = Duration::from_secs(24 * 60 * 60);
/// The whole fetch (every page), so the command answers quickly offline.
@@ -45,6 +51,79 @@ pub const FILTERS: &[(&str, &str, &str)] = &[
("CONTAINS", "usagetype", "Fargate"),
];
/// `GetProducts` filters for one EC2 instance type's on-demand Linux price
/// (`{region}`, `{instance_type}` filled in). Verified against the live API
/// (2026-11-16): exactly one row per type and region (`USW2-BoxUsage:m7i.large`,
/// $0.1008/hour), carrying both its OnDemand and Reserved terms.
pub const EC2_FILTERS: &[(&str, &str, &str)] = &[
("TERM_MATCH", "regionCode", "{region}"),
("TERM_MATCH", "instanceType", "{instance_type}"),
("TERM_MATCH", "operatingSystem", "Linux"),
("TERM_MATCH", "tenancy", "Shared"),
("TERM_MATCH", "preInstalledSw", "NA"),
("TERM_MATCH", "capacitystatus", "Used"),
("TERM_MATCH", "licenseModel", "No License required"),
];
/// `GetProducts` filters for gp3 EBS storage in `{region}`. Verified against
/// the live API (2026-11-16): one row (`USW2-EBS:VolumeUsage.gp3`, $0.08 per
/// GB-month).
pub const GP3_FILTERS: &[(&str, &str, &str)] = &[
("TERM_MATCH", "regionCode", "{region}"),
("TERM_MATCH", "productFamily", "Storage"),
("TERM_MATCH", "volumeApiName", "gp3"),
];
/// The EC2 instance's root volume (`ec2.tf`'s `volume_size`, gp3), billed
/// while the instance exists.
pub const EC2_ROOT_VOLUME_GIB: u32 = 50;
/// Hours in a billing month, as AWS converts GB-month prices.
pub const HOURS_PER_MONTH: f64 = 730.0;
/// One `GetProducts` query: a service and its filters, values filled in.
#[derive(Debug, Clone, PartialEq)]
pub struct Query {
pub service_code: &'static str,
pub filters: Vec<(&'static str, &'static str, String)>,
}
fn fill(filters: &[(&'static str, &'static str, &str)], vars: &[(&str, &str)]) -> Vec<(&'static str, &'static str, String)> {
filters
.iter()
.map(|(kind, field, value)| {
let value = vars
.iter()
.fold(value.to_string(), |v, (name, with)| v.replace(&format!("{{{name}}}"), with));
(*kind, *field, value)
})
.collect()
}
/// The Fargate rates of `region` ([`FILTERS`]).
pub fn fargate_query(region: &str) -> Query {
Query {
service_code: SERVICE_CODE,
filters: fill(FILTERS, &[("region", region)]),
}
}
/// The on-demand price of `instance_type` in `region` ([`EC2_FILTERS`]).
pub fn ec2_query(region: &str, instance_type: &str) -> Query {
Query {
service_code: EC2_SERVICE_CODE,
filters: fill(EC2_FILTERS, &[("region", region), ("instance_type", instance_type)]),
}
}
/// The gp3 storage price of `region` ([`GP3_FILTERS`]).
pub fn gp3_query(region: &str) -> Query {
Query {
service_code: EC2_SERVICE_CODE,
filters: fill(GP3_FILTERS, &[("region", region)]),
}
}
/// One architecture's on-demand rates in a region, in USD.
#[derive(Debug, Clone, Copy, PartialEq, Serialize, Deserialize)]
pub struct FargateRates {
@@ -74,12 +153,19 @@ impl RegionRates {
/// runs. A gateway sidecar shares the task's CPU and memory, so it adds
/// nothing; Spot isn't in the Price List API (`note_spot`: say it is
/// usually cheaper); EFS is billed per GB-month, not per hour
/// (`efs_billed_separately`).
/// (`efs_billed_separately`). `ec2_instance`: the figure is the user's EC2
/// instance's hourly price, whatever the project reserves of it, and
/// instances scale to zero when no workspace is running (E10);
/// `root_volume_included`: it also has the instance's 50 GiB gp3 root
/// volume (when its price could be read). An instance's public IPv4
/// address (public-mode environments) is never in it.
#[derive(Debug, Clone, Copy, PartialEq, Serialize)]
pub struct HourlyEstimate {
pub usd_per_hour: f64,
pub note_spot: bool,
pub efs_billed_separately: bool,
pub ec2_instance: bool,
pub root_volume_included: bool,
}
pub fn estimate(cfg: &AwsWorkspaceConfig, rates: &FargateRates) -> HourlyEstimate {
@@ -92,9 +178,77 @@ pub fn estimate(cfg: &AwsWorkspaceConfig, rates: &FargateRates) -> HourlyEstimat
+ billed_storage * rates.storage_gb_hour,
note_spot: cfg.spot,
efs_billed_separately: cfg.storage != StorageMode::Ephemeral,
ec2_instance: false,
root_volume_included: false,
}
}
/// An EC2 project's estimate: its instance's on-demand hourly price — the
/// instance is billed whole while it runs, so the reservations, the
/// ephemeral size and Spot (none on EC2) change nothing — plus its
/// [`EC2_ROOT_VOLUME_GIB`] gp3 root volume at `gp3_usd_per_gb_month` (per
/// hour, over [`HOURS_PER_MONTH`]) when that price is known.
pub fn estimate_ec2(
cfg: &AwsWorkspaceConfig,
instance_usd_per_hour: f64,
gp3_usd_per_gb_month: Option<f64>,
) -> HourlyEstimate {
let root = gp3_usd_per_gb_month.map(|p| f64::from(EC2_ROOT_VOLUME_GIB) * p / HOURS_PER_MONTH);
HourlyEstimate {
usd_per_hour: instance_usd_per_hour + root.unwrap_or(0.0),
note_spot: false,
efs_billed_separately: cfg.storage != StorageMode::Ephemeral,
ec2_instance: true,
root_volume_included: root.is_some(),
}
}
/// The gp3 storage price per GB-month in a [`GP3_FILTERS`] price list for
/// `region`; `None` when no row matches.
pub fn parse_gp3_price(region: &str, items: &[String]) -> Option<f64> {
items.iter().find_map(|item| {
let product = serde_json::from_str::<Value>(item).ok()?;
let attrs = &product["product"]["attributes"];
let is = |k: &str, v: &str| attrs[k].as_str() == Some(v);
if !(is("regionCode", region) && is("volumeApiName", "gp3")) {
return None;
}
on_demand_usd(&product).filter(|p| *p > 0.0)
})
}
/// The on-demand hourly price of `instance_type` (Linux, shared tenancy) in
/// an `AmazonEC2` price list for `region`. The filters already narrow it to
/// one row; the attributes are checked again so a looser answer can't price
/// the wrong product. `None` when no row matches.
pub fn parse_ec2_price(region: &str, instance_type: &str, items: &[String]) -> Option<f64> {
items.iter().find_map(|item| {
let product = serde_json::from_str::<Value>(item).ok()?;
let attrs = &product["product"]["attributes"];
let is = |k: &str, v: &str| attrs[k].as_str() == Some(v);
let matches = is("regionCode", region)
&& is("instanceType", instance_type)
&& is("operatingSystem", "Linux")
&& is("tenancy", "Shared")
&& is("preInstalledSw", "NA")
&& is("capacitystatus", "Used")
&& is("licenseModel", "No License required");
if !matches {
return None;
}
on_demand_usd(&product).filter(|p| *p > 0.0)
})
}
/// Whether `instance_type` is safe in a cache file name and a filter value
/// (`m7i.large`): lowercase letters, digits, one dot.
fn plain_instance_type(instance_type: &str) -> bool {
instance_type.split_once('.').is_some_and(|(family, size)| {
let ok = |s: &str| !s.is_empty() && s.bytes().all(|b| b.is_ascii_lowercase() || b.is_ascii_digit());
ok(family) && ok(size)
}) && instance_type.len() <= 32
}
/// The rates in a `GetProducts` price list (each item one product's JSON)
/// for `region`. `None` when no architecture is complete.
pub fn parse_price_list(region: &str, items: &[String]) -> Option<RegionRates> {
@@ -174,10 +328,10 @@ pub struct Page {
/// The Price List API, faked in tests.
pub trait PricingApi {
/// One page of `AmazonECS` products matching [`FILTERS`] for `region`.
/// One page of the products `query` matches.
fn get_products(
&self,
region: &str,
query: &Query,
next_token: Option<String>,
) -> impl Future<Output = Result<Page, String>> + Send;
}
@@ -197,20 +351,20 @@ impl SdkPricingApi {
}
impl PricingApi for SdkPricingApi {
async fn get_products(&self, region: &str, next_token: Option<String>) -> Result<Page, String> {
async fn get_products(&self, query: &Query, next_token: Option<String>) -> Result<Page, String> {
use aws_sdk_pricing::types::{Filter, FilterType};
let mut req = self
.client
.get_products()
.service_code(SERVICE_CODE)
.service_code(query.service_code)
.format_version("aws_v1")
.max_results(100)
.set_next_token(next_token);
for (kind, field, value) in FILTERS {
for (kind, field, value) in &query.filters {
let filter = Filter::builder()
.r#type(FilterType::from(*kind))
.field(*field)
.value(value.replace("{region}", region))
.value(value)
.build()
.map_err(|e| e.to_string())?;
req = req.filters(filter);
@@ -226,31 +380,79 @@ impl PricingApi for SdkPricingApi {
}
}
/// Every page for `region`, parsed.
pub async fn fetch_rates<A: PricingApi>(api: &A, region: &str) -> Result<RegionRates, String> {
/// Every page `query` matches.
async fn fetch_items<A: PricingApi>(api: &A, query: &Query) -> Result<Vec<String>, String> {
let mut items = Vec::new();
let mut token = None;
for _ in 0..MAX_PAGES {
let page = api.get_products(region, token).await?;
let page = api.get_products(query, token).await?;
items.extend(page.items);
token = page.next_token.filter(|t| !t.is_empty());
if token.is_none() {
return parse_price_list(region, &items)
.ok_or_else(|| format!("no Fargate rates for {region} in the price list"));
return Ok(items);
}
}
Err(format!(
"the {region} price list has more than {MAX_PAGES} pages"
))
Err(format!("the {} price list has more than {MAX_PAGES} pages", query.service_code))
}
/// `<dir>/<region>.json`.
/// Every page for `region`, parsed (the live tests' direct call).
#[cfg_attr(not(test), allow(dead_code))]
pub async fn fetch_rates<A: PricingApi>(api: &A, region: &str) -> Result<RegionRates, String> {
let items = fetch_items(api, &fargate_query(region)).await?;
parse_price_list(region, &items).ok_or_else(|| format!("no Fargate rates for {region} in the price list"))
}
/// The on-demand hourly price of `instance_type` in `region` (the live
/// tests' direct call).
#[cfg_attr(not(test), allow(dead_code))]
pub async fn fetch_ec2_price<A: PricingApi>(api: &A, region: &str, instance_type: &str) -> Result<f64, String> {
if !plain_instance_type(instance_type) {
return Err(format!("{instance_type:?} is not an instance type"));
}
let items = fetch_items(api, &ec2_query(region, instance_type)).await?;
parse_ec2_price(region, instance_type, &items)
.ok_or_else(|| format!("no {instance_type} price for {region} in the price list"))
}
/// One cached answer: Fargate rates under `<region>.json`, an instance's
/// price under `<region>--ec2-<type>.json`. `key` is what the file must say
/// it holds (the region, or `<region> <type>`), and what failures are
/// remembered under.
struct CacheKey {
key: String,
file: String,
}
impl CacheKey {
fn fargate(region: &str) -> Self {
Self {
key: region.to_string(),
file: format!("{region}.json"),
}
}
fn ec2(region: &str, instance_type: &str) -> Self {
Self {
key: format!("{region} {instance_type}"),
file: format!("{region}--ec2-{instance_type}.json"),
}
}
fn gp3(region: &str) -> Self {
Self {
key: format!("{region} gp3"),
file: format!("{region}--ebs-gp3.json"),
}
}
}
/// A cache file. `key` was `region` before EC2 prices were cached too (an
/// older file still reads).
#[derive(Debug, Serialize, Deserialize)]
struct CacheFile {
region: String,
struct CacheFile<V> {
#[serde(alias = "region")]
key: String,
/// Unix seconds.
fetched_at: u64,
rates: RegionRates,
rates: V,
}
/// The per-region rate files and the memory of recent failures.
@@ -281,18 +483,38 @@ impl RatesCache {
&self.dir
}
/// The region must already be validated (it names a file).
fn path(&self, region: &str) -> PathBuf {
self.dir.join(format!("{region}.json"))
}
/// The cached rates and whether they are still fresh at `now`. A
/// missing, unreadable or corrupt file, or one for another region, is
/// no cache.
/// no cache. The region must already be validated (it names a file).
pub fn read(&self, region: &str, now: SystemTime) -> Option<(RegionRates, bool)> {
let text = std::fs::read_to_string(self.path(region)).ok()?;
let file: CacheFile = serde_json::from_str(&text).ok()?;
if file.region != region {
self.read_key(&CacheKey::fargate(region), now)
}
#[cfg(test)]
pub fn write(&self, region: &str, rates: &RegionRates, now: SystemTime) -> Result<(), String> {
self.write_key(&CacheKey::fargate(region), rates, now)
}
/// [`Self::read`] for an instance type's hourly price.
pub fn read_ec2(&self, region: &str, instance_type: &str, now: SystemTime) -> Option<(f64, bool)> {
if !plain_instance_type(instance_type) {
return None;
}
self.read_key(&CacheKey::ec2(region, instance_type), now)
}
#[cfg(test)]
pub fn write_ec2(&self, region: &str, instance_type: &str, usd_per_hour: f64, now: SystemTime) -> Result<(), String> {
if !plain_instance_type(instance_type) {
return Err(format!("{instance_type:?} is not an instance type"));
}
self.write_key(&CacheKey::ec2(region, instance_type), &usd_per_hour, now)
}
fn read_key<V: DeserializeOwned>(&self, key: &CacheKey, now: SystemTime) -> Option<(V, bool)> {
let text = std::fs::read_to_string(self.dir.join(&key.file)).ok()?;
let file: CacheFile<V> = serde_json::from_str(&text).ok()?;
if file.key != key.key {
return None;
}
let now = unix_secs(now);
@@ -301,20 +523,20 @@ impl RatesCache {
Some((file.rates, fresh))
}
pub fn write(&self, region: &str, rates: &RegionRates, now: SystemTime) -> Result<(), String> {
fn write_key<V: Serialize>(&self, key: &CacheKey, rates: &V, now: SystemTime) -> Result<(), String> {
std::fs::create_dir_all(&self.dir).map_err(|e| e.to_string())?;
let file = CacheFile {
region: region.to_string(),
key: key.key.clone(),
fetched_at: unix_secs(now),
rates: *rates,
rates,
};
let text = serde_json::to_string_pretty(&file).map_err(|e| e.to_string())?;
// Write then rename, so a concurrent reader never sees half a file.
let tmp = self
.dir
.join(format!(".{region}.json.tmp-{}", std::process::id()));
.join(format!(".{}.tmp-{}", key.file, std::process::id()));
std::fs::write(&tmp, text).map_err(|e| e.to_string())?;
std::fs::rename(&tmp, self.path(region)).map_err(|e| {
std::fs::rename(&tmp, self.dir.join(&key.file)).map_err(|e| {
let _ = std::fs::remove_file(&tmp);
e.to_string()
})
@@ -360,17 +582,74 @@ impl RatesCache {
region: &str,
now: SystemTime,
) -> Option<RegionRates> {
let cached = self.read(region, now);
if let Some((rates, true)) = cached {
return Some(rates);
let parse = |items: &[String]| {
parse_price_list(region, items).ok_or_else(|| format!("no Fargate rates for {region} in the price list"))
};
self.cached::<A, B, _>(&CacheKey::fargate(region), api, fargate_query(region), parse, now)
.await
}
/// [`Self::rates_with`] for `instance_type`'s on-demand hourly price in
/// `region` (E10): the same cache rules, failure back-off and timeout.
pub async fn ec2_price_with<A: PricingApi, B: std::borrow::Borrow<A>>(
&self,
api: impl Future<Output = Option<B>>,
region: &str,
instance_type: &str,
now: SystemTime,
) -> Option<f64> {
if !plain_instance_type(instance_type) {
return None;
}
let stale = cached.map(|(rates, _)| rates);
if self.recently_failed(region, now) {
let parse = |items: &[String]| {
parse_ec2_price(region, instance_type, items)
.ok_or_else(|| format!("no {instance_type} price for {region} in the price list"))
};
let query = ec2_query(region, instance_type);
self.cached::<A, B, _>(&CacheKey::ec2(region, instance_type), api, query, parse, now)
.await
}
/// [`Self::rates_with`] for `region`'s gp3 storage price per GB-month
/// (an EC2 instance's root volume).
pub async fn gp3_price_with<A: PricingApi, B: std::borrow::Borrow<A>>(
&self,
api: impl Future<Output = Option<B>>,
region: &str,
now: SystemTime,
) -> Option<f64> {
let parse = |items: &[String]| {
parse_gp3_price(region, items).ok_or_else(|| format!("no gp3 price for {region} in the price list"))
};
self.cached::<A, B, _>(&CacheKey::gp3(region), api, gp3_query(region), parse, now)
.await
}
/// Fresh cache, else a fetch of `query` (client build included, within
/// [`FETCH_TIMEOUT`]) parsed by `parse` that refreshes the cache, else
/// the stale cache, else `None`; a key whose fetch failed is not asked
/// again for [`RETRY_AFTER_FAILURE`].
async fn cached<A: PricingApi, B: std::borrow::Borrow<A>, V: Serialize + DeserializeOwned + Clone>(
&self,
key: &CacheKey,
api: impl Future<Output = Option<B>>,
query: Query,
parse: impl FnOnce(&[String]) -> Result<V, String>,
now: SystemTime,
) -> Option<V> {
let cached = self.read_key::<V>(key, now);
let stale = match cached {
Some((rates, true)) => return Some(rates),
Some((rates, false)) => Some(rates),
None => None,
};
let what = &key.key;
if self.recently_failed(what, now) {
return stale;
}
let fetch = async {
let api = api.await?;
Some(fetch_rates(api.borrow(), region).await)
Some(fetch_items(api.borrow(), &query).await.and_then(|items| parse(&items)))
};
let fetched = match tokio::time::timeout(FETCH_TIMEOUT, fetch).await {
Ok(Some(r)) => r,
@@ -378,23 +657,23 @@ impl RatesCache {
// one reads credentials, which can be slow), and the stale cache
// answers.
Ok(None) => {
log::info!("No Fargate rates for {region}: no Pricing client could be built");
self.note_failure(region, now, true);
log::info!("No prices for {what}: no Pricing client could be built");
self.note_failure(what, now, true);
return stale;
}
Err(_) => Err(format!("timed out after {}s", FETCH_TIMEOUT.as_secs())),
};
match fetched {
Ok(rates) => {
self.note_failure(region, now, false);
if let Err(e) = self.write(region, &rates, now) {
log::warn!("Could not cache the {region} Fargate rates: {e}");
self.note_failure(what, now, false);
if let Err(e) = self.write_key(key, &rates, now) {
log::warn!("Could not cache the prices for {what}: {e}");
}
Some(rates)
}
Err(e) => {
log::info!("No Fargate rates for {region}: {e}");
self.note_failure(region, now, true);
log::info!("No prices for {what}: {e}");
self.note_failure(what, now, true);
stale
}
}
@@ -0,0 +1,6 @@
{
"PriceList": [
"{\"product\":{\"productFamily\":\"Storage\",\"attributes\":{\"maxThroughputvolume\":\"2000 MiB/s\",\"volumeType\":\"General Purpose\",\"maxIopsvolume\":\"80000\",\"usagetype\":\"USW2-EBS:VolumeUsage.gp3\",\"locationType\":\"AWS Region\",\"maxVolumeSize\":\"64 TiB\",\"storageMedia\":\"SSD-backed\",\"regionCode\":\"us-west-2\",\"servicecode\":\"AmazonEC2\",\"volumeApiName\":\"gp3\",\"location\":\"US West (Oregon)\",\"servicename\":\"Amazon Elastic Compute Cloud\",\"operation\":\"\",\"group\":\"EBS Storage\"},\"sku\":\"BB8UJWJ4XPFJB95G\"},\"serviceCode\":\"AmazonEC2\",\"terms\":{\"OnDemand\":{\"BB8UJWJ4XPFJB95G.JRTCKXETXF\":{\"priceDimensions\":{\"BB8UJWJ4XPFJB95G.JRTCKXETXF.6YS6EN2CT7\":{\"unit\":\"GB-Mo\",\"endRange\":\"Inf\",\"description\":\"$0.08 per GB-month of General Purpose (gp3) provisioned storage - US West (Oregon)\",\"appliesTo\":[],\"rateCode\":\"BB8UJWJ4XPFJB95G.JRTCKXETXF.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0800000000\"}}},\"sku\":\"BB8UJWJ4XPFJB95G\",\"effectiveDate\":\"2026-11-01T00:00:00Z\",\"offerTermCode\":\"JRTCKXETXF\",\"termAttributes\":{}}}},\"version\":\"20261114012926\",\"publicationDate\":\"2026-11-14T01:29:26Z\"}"
],
"FormatVersion": "aws_v1"
}
@@ -0,0 +1,7 @@
{
"FormatVersion": "aws_v1",
"PriceList": [
"{\"product\":{\"productFamily\":\"Compute Instance\",\"attributes\":{\"enhancedNetworkingSupported\":\"Yes\",\"intelTurboAvailable\":\"Yes\",\"memory\":\"8 GiB\",\"dedicatedEbsThroughput\":\"Up to 10000 Mbps\",\"vcpu\":\"2\",\"classicnetworkingsupport\":\"false\",\"capacitystatus\":\"Used\",\"locationType\":\"AWS Region\",\"storage\":\"EBS only\",\"instanceFamily\":\"General purpose\",\"operatingSystem\":\"Linux\",\"intelAvx2Available\":\"Yes\",\"instanceFamilyCategory\":\"General Purpose\",\"regionCode\":\"us-west-2\",\"physicalProcessor\":\"Intel Xeon Scalable (Sapphire Rapids)\",\"clockSpeed\":\"3.2 GHz\",\"ecu\":\"NA\",\"networkPerformance\":\"Up to 12500 Megabit\",\"servicename\":\"Amazon Elastic Compute Cloud\",\"gpuMemory\":\"NA\",\"dedicatedEbsThroughputDescription\":\"650 Mbps\",\"vpcnetworkingsupport\":\"true\",\"instanceType\":\"m7i.large\",\"tenancy\":\"Shared\",\"usagetype\":\"USW2-BoxUsage:m7i.large\",\"normalizationSizeFactor\":\"4\",\"intelAvxAvailable\":\"Yes\",\"processorFeatures\":\"Intel AVX; Intel AVX2; Intel AVX512; Intel Turbo; Intel AMX\",\"servicecode\":\"AmazonEC2\",\"licenseModel\":\"No License required\",\"currentGeneration\":\"Yes\",\"preInstalledSw\":\"NA\",\"location\":\"US West (Oregon)\",\"processorArchitecture\":\"64-bit\",\"marketoption\":\"OnDemand\",\"operation\":\"RunInstances\",\"availabilityzone\":\"NA\"},\"sku\":\"76SYQ5S6SHTVH6CA\"},\"serviceCode\":\"AmazonEC2\",\"terms\":{\"OnDemand\":{\"76SYQ5S6SHTVH6CA.JRTCKXETXF\":{\"priceDimensions\":{\"76SYQ5S6SHTVH6CA.JRTCKXETXF.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"$0.1008 per On Demand Linux m7i.large Instance Hour\",\"appliesTo\":[],\"rateCode\":\"76SYQ5S6SHTVH6CA.JRTCKXETXF.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.1008000000\"}}},\"sku\":\"76SYQ5S6SHTVH6CA\",\"effectiveDate\":\"2026-11-01T00:00:00Z\",\"offerTermCode\":\"JRTCKXETXF\",\"termAttributes\":{}}},\"Reserved\":{\"76SYQ5S6SHTVH6CA.VJWZNREJX2\":{\"priceDimensions\":{\"76SYQ5S6SHTVH6CA.VJWZNREJX2.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"USD 0.0 per Linux/UNIX (Amazon VPC), m7i.large reserved instance applied\",\"appliesTo\":[],\"rateCode\":\"76SYQ5S6SHTVH6CA.VJWZNREJX2.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0000000000\"}},\"76SYQ5S6SHTVH6CA.VJWZNREJX2.2TG2D8R56U\":{\"unit\":\"Quantity\",\"description\":\"Upfront Fee\",\"appliesTo\":[],\"rateCode\":\"76SYQ5S6SHTVH6CA.VJWZNREJX2.2TG2D8R56U\",\"pricePerUnit\":{\"USD\":\"605\"}}},\"sku\":\"76SYQ5S6SHTVH6CA\",\"effectiveDate\":\"2026-11-14T01:50:11Z\",\"offerTermCode\":\"VJWZNREJX2\",\"termAttributes\":{\"LeaseContractLength\":\"1yr\",\"OfferingClass\":\"convertible\",\"PurchaseOption\":\"All Upfront\"}},\"76SYQ5S6SHTVH6CA.BPH4J8HBKS\":{\"priceDimensions\":{\"76SYQ5S6SHTVH6CA.BPH4J8HBKS.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"Linux/UNIX (Amazon VPC), m7i.large reserved instance applied\",\"appliesTo\":[],\"rateCode\":\"76SYQ5S6SHTVH6CA.BPH4J8HBKS.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0457200000\"}}},\"sku\":\"76SYQ5S6SHTVH6CA\",\"effectiveDate\":\"2026-11-14T01:50:09Z\",\"offerTermCode\":\"BPH4J8HBKS\",\"termAttributes\":{\"LeaseContractLength\":\"3yr\",\"OfferingClass\":\"standard\",\"PurchaseOption\":\"No Upfront\"}},\"76SYQ5S6SHTVH6CA.NQ3QZPMQV9\":{\"priceDimensions\":{\"76SYQ5S6SHTVH6CA.NQ3QZPMQV9.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"USD 0.0 per Linux/UNIX (Amazon VPC), m7i.large reserved instance applied\",\"appliesTo\":[],\"rateCode\":\"76SYQ5S6SHTVH6CA.NQ3QZPMQV9.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0000000000\"}},\"76SYQ5S6SHTVH6CA.NQ3QZPMQV9.2TG2D8R56U\":{\"unit\":\"Quantity\",\"description\":\"Upfront Fee\",\"appliesTo\":[],\"rateCode\":\"76SYQ5S6SHTVH6CA.NQ3QZPMQV9.2TG2D8R56U\",\"pricePerUnit\":{\"USD\":\"1046\"}}},\"sku\":\"76SYQ5S6SHTVH6CA\",\"effectiveDate\":\"2026-11-14T01:50:12Z\",\"offerTermCode\":\"NQ3QZPMQV9\",\"termAttributes\":{\"LeaseContractLength\":\"3yr\",\"OfferingClass\":\"standard\",\"PurchaseOption\":\"All Upfront\"}},\"76SYQ5S6SHTVH6CA.HU7G6KETJZ\":{\"priceDimensions\":{\"76SYQ5S6SHTVH6CA.HU7G6KETJZ.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"Linux/UNIX (Amazon VPC), m7i.large reserved instance applied\",\"appliesTo\":[],\"rateCode\":\"76SYQ5S6SHTVH6CA.HU7G6KETJZ.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0317500000\"}},\"76SYQ5S6SHTVH6CA.HU7G6KETJZ.2TG2D8R56U\":{\"unit\":\"Quantity\",\"description\":\"Upfront Fee\",\"appliesTo\":[],\"rateCode\":\"76SYQ5S6SHTVH6CA.HU7G6KETJZ.2TG2D8R56U\",\"pricePerUnit\":{\"USD\":\"278\"}}},\"sku\":\"76SYQ5S6SHTVH6CA\",\"effectiveDate\":\"2026-11-14T01:50:08Z\",\"offerTermCode\":\"HU7G6KETJZ\",\"termAttributes\":{\"LeaseContractLength\":\"1yr\",\"OfferingClass\":\"standard\",\"PurchaseOption\":\"Partial Upfront\"}},\"76SYQ5S6SHTVH6CA.38NPMPTW36\":{\"priceDimensions\":{\"76SYQ5S6SHTVH6CA.Line truncated
"{\"product\":{\"productFamily\":\"Compute Instance\",\"attributes\":{\"enhancedNetworkingSupported\":\"Yes\",\"intelTurboAvailable\":\"No\",\"memory\":\"8 GiB\",\"dedicatedEbsThroughput\":\"Up to 10000 Mbps\",\"vcpu\":\"2\",\"classicnetworkingsupport\":\"false\",\"capacitystatus\":\"Used\",\"locationType\":\"AWS Region\",\"storage\":\"EBS only\",\"instanceFamily\":\"General purpose\",\"operatingSystem\":\"Linux\",\"intelAvx2Available\":\"No\",\"instanceFamilyCategory\":\"General Purpose\",\"regionCode\":\"us-west-2\",\"physicalProcessor\":\"AWS Graviton3 Processor\",\"clockSpeed\":\"2.5 GHz\",\"ecu\":\"NA\",\"networkPerformance\":\"Up to 12500 Megabit\",\"servicename\":\"Amazon Elastic Compute Cloud\",\"gpuMemory\":\"NA\",\"dedicatedEbsThroughputDescription\":\"630 Mbps\",\"vpcnetworkingsupport\":\"true\",\"instanceType\":\"m7g.large\",\"tenancy\":\"Shared\",\"usagetype\":\"USW2-BoxUsage:m7g.large\",\"normalizationSizeFactor\":\"4\",\"intelAvxAvailable\":\"No\",\"servicecode\":\"AmazonEC2\",\"licenseModel\":\"No License required\",\"currentGeneration\":\"Yes\",\"preInstalledSw\":\"NA\",\"location\":\"US West (Oregon)\",\"processorArchitecture\":\"64-bit\",\"marketoption\":\"OnDemand\",\"operation\":\"RunInstances\",\"availabilityzone\":\"NA\"},\"sku\":\"NWX8JZYSP4RSV4Q2\"},\"serviceCode\":\"AmazonEC2\",\"terms\":{\"OnDemand\":{\"NWX8JZYSP4RSV4Q2.JRTCKXETXF\":{\"priceDimensions\":{\"NWX8JZYSP4RSV4Q2.JRTCKXETXF.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"$0.0816 per On Demand Linux m7g.large Instance Hour\",\"appliesTo\":[],\"rateCode\":\"NWX8JZYSP4RSV4Q2.JRTCKXETXF.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0816000000\"}}},\"sku\":\"NWX8JZYSP4RSV4Q2\",\"effectiveDate\":\"2026-11-01T00:00:00Z\",\"offerTermCode\":\"JRTCKXETXF\",\"termAttributes\":{}}},\"Reserved\":{\"NWX8JZYSP4RSV4Q2.HU7G6KETJZ\":{\"priceDimensions\":{\"NWX8JZYSP4RSV4Q2.HU7G6KETJZ.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"Linux/UNIX (Amazon VPC), m7g.large reserved instance applied\",\"appliesTo\":[],\"rateCode\":\"NWX8JZYSP4RSV4Q2.HU7G6KETJZ.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0257000000\"}},\"NWX8JZYSP4RSV4Q2.HU7G6KETJZ.2TG2D8R56U\":{\"unit\":\"Quantity\",\"description\":\"Upfront Fee\",\"appliesTo\":[],\"rateCode\":\"NWX8JZYSP4RSV4Q2.HU7G6KETJZ.2TG2D8R56U\",\"pricePerUnit\":{\"USD\":\"225\"}}},\"sku\":\"NWX8JZYSP4RSV4Q2\",\"effectiveDate\":\"2026-11-14T01:50:10Z\",\"offerTermCode\":\"HU7G6KETJZ\",\"termAttributes\":{\"LeaseContractLength\":\"1yr\",\"OfferingClass\":\"standard\",\"PurchaseOption\":\"Partial Upfront\"}},\"NWX8JZYSP4RSV4Q2.R5XV2EPZQZ\":{\"priceDimensions\":{\"NWX8JZYSP4RSV4Q2.R5XV2EPZQZ.2TG2D8R56U\":{\"unit\":\"Quantity\",\"description\":\"Upfront Fee\",\"appliesTo\":[],\"rateCode\":\"NWX8JZYSP4RSV4Q2.R5XV2EPZQZ.2TG2D8R56U\",\"pricePerUnit\":{\"USD\":\"503\"}},\"NWX8JZYSP4RSV4Q2.R5XV2EPZQZ.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"Linux/UNIX (Amazon VPC), m7g.large reserved instance applied\",\"appliesTo\":[],\"rateCode\":\"NWX8JZYSP4RSV4Q2.R5XV2EPZQZ.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0192000000\"}}},\"sku\":\"NWX8JZYSP4RSV4Q2\",\"effectiveDate\":\"2026-11-14T01:50:09Z\",\"offerTermCode\":\"R5XV2EPZQZ\",\"termAttributes\":{\"LeaseContractLength\":\"3yr\",\"OfferingClass\":\"convertible\",\"PurchaseOption\":\"Partial Upfront\"}},\"NWX8JZYSP4RSV4Q2.38NPMPTW36\":{\"priceDimensions\":{\"NWX8JZYSP4RSV4Q2.38NPMPTW36.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"Linux/UNIX (Amazon VPC), m7g.large reserved instance applied\",\"appliesTo\":[],\"rateCode\":\"NWX8JZYSP4RSV4Q2.38NPMPTW36.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0171000000\"}},\"NWX8JZYSP4RSV4Q2.38NPMPTW36.2TG2D8R56U\":{\"unit\":\"Quantity\",\"description\":\"Upfront Fee\",\"appliesTo\":[],\"rateCode\":\"NWX8JZYSP4RSV4Q2.38NPMPTW36.2TG2D8R56U\",\"pricePerUnit\":{\"USD\":\"450\"}}},\"sku\":\"NWX8JZYSP4RSV4Q2\",\"effectiveDate\":\"2026-11-14T01:50:09Z\",\"offerTermCode\":\"38NPMPTW36\",\"termAttributes\":{\"LeaseContractLength\":\"3yr\",\"OfferingClass\":\"standard\",\"PurchaseOption\":\"Partial Upfront\"}},\"NWX8JZYSP4RSV4Q2.Z2E3P23VKM\":{\"priceDimensions\":{\"NWX8JZYSP4RSV4Q2.Z2E3P23VKM.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"Linux/UNIX (Amazon VPC), m7g.large reserved instance applied\",\"appliesTo\":[],\"rateCode\":\"NWX8JZYSP4RSV4Q2.Z2E3P23VKM.6YS6EN2CT7\",\"beginRange\":\"0\",\"pricePerUnit\":{\"USD\":\"0.0414000000\"}}},\"sku\":\"NWX8JZYSP4RSV4Q2\",\"effectiveDate\":\"2026-11-14T01:50:12Z\",\"offerTermCode\":\"Z2E3P23VKM\",\"termAttributes\":{\"LeaseContractLength\":\"3yr\",\"OfferingClass\":\"convertible\",\"PurchaseOption\":\"No Upfront\"}},\"NWX8JZYSP4RSV4Q2.NQ3QZPMQV9\":{\"priceDimensions\":{\"NWX8JZYSP4RSV4Q2.NQ3QZPMQV9.6YS6EN2CT7\":{\"unit\":\"Hrs\",\"endRange\":\"Inf\",\"description\":\"USD 0.0 per Linux/UNIX (Amazon VPC),Line truncated
]
}
+233 -2
View File
@@ -173,6 +173,7 @@ fn the_filters_name_the_region_on_demand_and_fargate() {
struct FakeApi {
pages: Mutex<Vec<Result<Page, String>>>,
calls: AtomicUsize,
queries: Mutex<Vec<Query>>,
hang: bool,
}
@@ -181,6 +182,7 @@ impl FakeApi {
Self {
pages: Mutex::new(pages),
calls: AtomicUsize::new(0),
queries: Mutex::new(Vec::new()),
hang: false,
}
}
@@ -205,8 +207,12 @@ impl FakeApi {
}
impl PricingApi for FakeApi {
async fn get_products(&self, region: &str, _token: Option<String>) -> Result<Page, String> {
assert_eq!(region, "us-west-2");
async fn get_products(&self, query: &Query, _token: Option<String>) -> Result<Page, String> {
assert!(
query.filters.contains(&("TERM_MATCH", "regionCode", "us-west-2".to_string())),
"{query:?}"
);
self.queries.lock().unwrap().push(query.clone());
self.calls.fetch_add(1, Ordering::SeqCst);
if self.hang {
std::future::pending::<()>().await;
@@ -416,3 +422,228 @@ async fn no_client_counts_as_a_failure_and_is_not_asked_again_for_a_while() {
assert_eq!(c.rates(Some(&api), "us-west-2", at(later)).await, Some(RATES));
assert_eq!(api.calls(), 1);
}
// ---- EC2 (phase 7 E10) ----
/// Real `AmazonEC2` `GetProducts` answers for us-west-2 with [`EC2_FILTERS`]
/// (recorded 2026-11-16): one row each for m7i.large and m7g.large.
const EC2_FIXTURE: &str = include_str!("fixtures/ec2-get-products-us-west-2.json");
const M7I_LARGE: f64 = 0.1008;
const M7G_LARGE: f64 = 0.0816;
fn ec2_items() -> Vec<String> {
let v: Value = serde_json::from_str(EC2_FIXTURE).unwrap();
v["PriceList"]
.as_array()
.unwrap()
.iter()
.map(|s| s.as_str().unwrap().to_string())
.collect()
}
fn ec2_cfg(cpu: u32, memory_mib: u32, storage: StorageMode) -> AwsWorkspaceConfig {
AwsWorkspaceConfig {
launch_type: crate::models::project::LaunchType::Ec2,
cpu,
memory_mib,
storage,
..Default::default()
}
}
#[test]
fn the_recorded_ec2_price_list_gives_each_instance_types_hourly_price() {
let items = ec2_items();
assert_eq!(parse_ec2_price("us-west-2", "m7i.large", &items), Some(M7I_LARGE));
assert_eq!(parse_ec2_price("us-west-2", "m7g.large", &items), Some(M7G_LARGE));
// Not a type in the list, or another region: nothing.
assert_eq!(parse_ec2_price("us-west-2", "c7i.large", &items), None);
assert_eq!(parse_ec2_price("eu-west-1", "m7i.large", &items), None);
}
#[test]
fn an_ec2_row_for_other_software_tenancy_or_capacity_never_counts() {
let row = ec2_items().into_iter().find(|s| s.contains("m7i.large")).unwrap();
for (from, to) in [
("\"tenancy\":\"Shared\"", "\"tenancy\":\"Dedicated\""),
("\"preInstalledSw\":\"NA\"", "\"preInstalledSw\":\"SQL Web\""),
("\"capacitystatus\":\"Used\"", "\"capacitystatus\":\"UnusedCapacityReservation\""),
("\"operatingSystem\":\"Linux\"", "\"operatingSystem\":\"Windows\""),
("\"licenseModel\":\"No License required\"", "\"licenseModel\":\"Bring your own license\""),
] {
assert!(row.contains(from), "{from}");
let other = row.replace(from, to);
assert_eq!(parse_ec2_price("us-west-2", "m7i.large", &[other]), None, "{to}");
}
// A zero price (what an unused reservation row says) is not a price.
let zero = row.replace("0.1008000000", "0.0000000000");
assert_eq!(parse_ec2_price("us-west-2", "m7i.large", &[zero]), None);
assert_eq!(parse_ec2_price("us-west-2", "m7i.large", &["garbage".into()]), None);
}
#[test]
fn the_ec2_filters_name_linux_shared_tenancy_on_demand_for_the_type() {
assert_eq!(EC2_SERVICE_CODE, "AmazonEC2");
let q = ec2_query("us-west-2", "m7i.large");
assert_eq!(q.service_code, "AmazonEC2");
let f = |field: &'static str, value: &str| ("TERM_MATCH", field, value.to_string());
assert_eq!(
q.filters,
vec![
f("regionCode", "us-west-2"),
f("instanceType", "m7i.large"),
f("operatingSystem", "Linux"),
f("tenancy", "Shared"),
f("preInstalledSw", "NA"),
f("capacitystatus", "Used"),
f("licenseModel", "No License required"),
]
);
// The Fargate query is unchanged.
let fq = fargate_query("us-west-2");
assert_eq!(fq.service_code, "AmazonECS");
assert_eq!(
fq.filters,
vec![
("TERM_MATCH", "regionCode", "us-west-2".to_string()),
("TERM_MATCH", "termType", "OnDemand".to_string()),
("CONTAINS", "usagetype", "Fargate".to_string()),
]
);
}
#[test]
fn an_ec2_estimate_is_the_instances_price_whatever_the_reservations() {
let small = estimate_ec2(&ec2_cfg(512, 1024, StorageMode::Ephemeral), M7I_LARGE, None);
let big = estimate_ec2(&ec2_cfg(1920, 7373, StorageMode::Ephemeral), M7I_LARGE, None);
assert!(close(small.usd_per_hour, 0.1008) && close(big.usd_per_hour, 0.1008));
assert!(small.ec2_instance && !small.note_spot && !small.efs_billed_separately);
assert!(!small.root_volume_included);
// Ephemeral size and Spot don't apply on EC2; EFS is still billed apart.
let mut c = ec2_cfg(1024, 4096, StorageMode::Hybrid);
c.ephemeral_gib = 200;
c.spot = true;
let e = estimate_ec2(&c, M7G_LARGE, None);
assert!(close(e.usd_per_hour, 0.0816));
assert!(e.efs_billed_separately && !e.note_spot && e.ec2_instance);
// Fargate estimates say they are not an instance's.
assert!(!estimate(&cfg(1024, 4096, 21), &X86).ec2_instance);
}
fn ec2_api() -> FakeApi {
FakeApi::answering(vec![Ok(Page {
items: ec2_items(),
next_token: None,
})])
}
#[tokio::test]
async fn an_ec2_price_is_fetched_once_and_cached_per_region_and_type() {
let (_d, c) = cache();
let api = ec2_api();
let price = c.ec2_price_with::<FakeApi, _>(async { Some(&api) }, "us-west-2", "m7i.large", at(T0)).await;
assert_eq!(price, Some(M7I_LARGE));
assert_eq!(api.calls(), 1);
assert_eq!(api.queries.lock().unwrap()[0], ec2_query("us-west-2", "m7i.large"));
assert!(c.dir().join("us-west-2--ec2-m7i.large.json").is_file());
assert_eq!(c.read_ec2("us-west-2", "m7i.large", at(T0 + DAY - 1)), Some((M7I_LARGE, true)));
assert_eq!(c.read_ec2("us-west-2", "m7i.large", at(T0 + DAY)), Some((M7I_LARGE, false)));
// Fresh: no call. Another type, or the region's Fargate rates, are not it.
let again = c
.ec2_price_with::<FakeApi, FakeApi>(async { panic!("no client for a fresh cache") }, "us-west-2", "m7i.large", at(T0 + 60))
.await;
assert_eq!(again, Some(M7I_LARGE));
assert_eq!(c.read_ec2("us-west-2", "m7g.large", at(T0)), None);
assert_eq!(c.read("us-west-2", at(T0)), None);
// A file renamed to another type's name is no cache.
std::fs::copy(
c.dir().join("us-west-2--ec2-m7i.large.json"),
c.dir().join("us-west-2--ec2-m7g.large.json"),
)
.unwrap();
assert_eq!(c.read_ec2("us-west-2", "m7g.large", at(T0)), None);
}
#[tokio::test]
async fn an_ec2_price_backs_off_after_a_failure_and_falls_back_to_the_stale_cache() {
let (_d, c) = cache();
let denied = FakeApi::denied();
assert_eq!(c.ec2_price_with::<FakeApi, _>(async { Some(&denied) }, "us-west-2", "m7i.large", at(T0)).await, None);
assert_eq!(c.ec2_price_with::<FakeApi, _>(async { Some(&denied) }, "us-west-2", "m7i.large", at(T0 + 60)).await, None);
assert_eq!(denied.calls(), 1, "a recent failure is not retried");
// The failure is per type: another type is still asked.
let other = FakeApi::denied();
let _ = c.ec2_price_with::<FakeApi, _>(async { Some(&other) }, "us-west-2", "m7g.large", at(T0 + 60)).await;
assert_eq!(other.calls(), 1);
// A stale price answers when the refresh fails.
c.write_ec2("us-west-2", "c7i.large", 0.0893, at(T0)).unwrap();
let r = c.ec2_price_with::<FakeApi, _>(async { Some(&denied) }, "us-west-2", "c7i.large", at(T0 + 2 * DAY)).await;
assert_eq!(r, Some(0.0893));
}
#[tokio::test(start_paused = true)]
async fn a_hanging_ec2_price_times_out_to_none() {
let (_d, c) = cache();
let api = FakeApi::hanging();
let started = tokio::time::Instant::now();
assert_eq!(c.ec2_price_with::<FakeApi, _>(async { Some(&api) }, "us-west-2", "m7i.large", at(T0)).await, None);
assert_eq!(started.elapsed(), FETCH_TIMEOUT);
}
#[tokio::test]
async fn an_instance_type_that_could_name_another_file_is_never_used() {
let (_d, c) = cache();
for bad in ["../x", "m7i.large/..", "M7I.large", "m7i", "", "m7i.large.json", "m7i .large"] {
let r = c
.ec2_price_with::<FakeApi, FakeApi>(async { panic!("no client for {bad:?}") }, "us-west-2", bad, at(T0))
.await;
assert_eq!(r, None, "{bad}");
assert_eq!(c.read_ec2("us-west-2", bad, at(T0)), None);
assert!(c.write_ec2("us-west-2", bad, 1.0, at(T0)).is_err(), "{bad}");
assert!(fetch_ec2_price(&ec2_api(), "us-west-2", bad).await.is_err());
}
assert_eq!(fetch_ec2_price(&ec2_api(), "us-west-2", "m7g.large").await, Ok(M7G_LARGE));
}
#[test]
fn a_fargate_cache_file_written_before_ec2_prices_still_reads() {
let (_d, c) = cache();
std::fs::create_dir_all(c.dir()).unwrap();
let old = serde_json::json!({ "region": "us-west-2", "fetched_at": T0, "rates": RATES });
std::fs::write(c.dir().join("us-west-2.json"), old.to_string()).unwrap();
assert_eq!(c.read("us-west-2", at(T0)), Some((RATES, true)));
}
const GP3_FIXTURE: &str = include_str!("fixtures/ebs-gp3-get-products-us-west-2.json");
fn gp3_items() -> Vec<String> {
let v: Value = serde_json::from_str(GP3_FIXTURE).unwrap();
v["PriceList"].as_array().unwrap().iter().map(|s| s.as_str().unwrap().to_string()).collect()
}
/// The root volume (50 GiB gp3, `ec2.tf`) is in an EC2 estimate when its
/// price is known: $0.08/GB-month in us-west-2 (recorded 2026-11-16) is
/// 50 × 0.08 / 730 ≈ $0.0055/h on top of the instance.
#[test]
fn an_ec2_estimate_adds_the_gp3_root_volume_when_its_price_is_known() {
assert_eq!(parse_gp3_price("us-west-2", &gp3_items()), Some(0.08));
assert_eq!(parse_gp3_price("eu-west-1", &gp3_items()), None, "another region's row");
assert_eq!(parse_gp3_price("us-west-2", &ec2_items()), None, "not a gp3 row");
let e = estimate_ec2(&ec2_cfg(1024, 4096, StorageMode::Ephemeral), M7I_LARGE, Some(0.08));
assert!(close(e.usd_per_hour, 0.1008 + 50.0 * 0.08 / 730.0), "{}", e.usd_per_hour);
assert!(e.root_volume_included && e.ec2_instance);
assert_eq!(gp3_query("us-west-2").service_code, "AmazonEC2");
}
#[tokio::test]
async fn the_gp3_price_is_fetched_once_and_cached_per_region() {
let (_d, c) = cache();
let api = FakeApi::answering(vec![Ok(Page { items: gp3_items(), next_token: None })]);
let price = c.gp3_price_with::<FakeApi, _>(async { Some(&api) }, "us-west-2", at(T0)).await;
assert_eq!(price, Some(0.08));
assert_eq!(api.queries.lock().unwrap()[0], gp3_query("us-west-2"));
assert!(c.dir().join("us-west-2--ebs-gp3.json").is_file());
let again = c.gp3_price_with::<FakeApi, _>(async { Some(&api) }, "us-west-2", at(T0 + 1)).await;
assert_eq!(again, Some(0.08));
assert_eq!(api.calls(), 1);
}
+88 -11
View File
@@ -14,12 +14,15 @@ use std::time::Duration;
use aws_config::SdkConfig;
use aws_sdk_secretsmanager::error::ProvideErrorMetadata;
use super::environments::{knowable, parse_environment_record, EnvironmentInfo, Knowable};
use super::environments::{
get_parameter, knowable, parse_environment_record, user_record_from, EnvironmentInfo, Knowable,
Param, UserRecord,
};
use super::error::{classify_sdk_error, AwsError};
use super::tasks::{find_project_task, TaskApi, STOP_REASON};
use super::tofu_modules::{
access_point_root, workspace_access_point_root, environment_cluster_name, environment_parameter_name, project_secret_name, task_family,
user_file_system_token, user_role_names,
is_user_ec2_cluster, user_file_system_token, user_record_parameter_name, user_role_names,
};
use super::workspace::WorkspaceApi;
@@ -40,6 +43,13 @@ pub trait RuntimeApi: Send + Sync {
&self,
env: &str,
) -> impl Future<Output = Result<Option<String>, AwsError>> + Send;
/// `ssm:GetParameter` of the caller's EC2 record in `env`
/// (`user_record_parameter_name`).
fn user_record(
&self,
env: &str,
owner_key: &str,
) -> impl Future<Output = Result<Param, AwsError>> + Send;
/// `iam:GetRole`.
fn role_exists(&self, name: &str) -> impl Future<Output = Result<Knowable, AwsError>> + Send;
/// `secretsmanager:GetSecretValue`'s `SecretString`.
@@ -80,6 +90,50 @@ pub async fn read_environment<R: RuntimeApi>(
})
}
/// The caller's EC2 record in `env`, validated (untrusted: `parse_user_record`).
/// Absent = no EC2 capacity for NEW runs; a task already running on the
/// user's cluster may outlive the record (teardown removes it first).
pub async fn read_user_record<R: RuntimeApi>(
rt: &R,
env: &str,
owner_key: &str,
account: &str,
region: &str,
) -> Result<UserRecord, AwsError> {
Ok(user_record_from(rt.user_record(env, owner_key).await?, env, account, region))
}
/// The per-user EC2 cluster (ARN) a stored task ARN names, when it is one of
/// `env`'s: where a project that ran on EC2 is still looked for after its
/// user record is gone.
pub fn ec2_cluster_of_task(env: &str, task_arn: &str) -> Option<String> {
cluster_of_task(task_arn)
.filter(|c| is_user_ec2_cluster(env, c))
.and_then(|_| cluster_arn_of_task(task_arn))
}
/// Where a project's task may be: `env_cluster` (the environment's), then
/// the user's EC2 cluster from their record, then the per-user cluster
/// `stored_task_arn` names (the record may be gone while that task runs),
/// without repeats.
pub fn project_task_clusters(
env: &str,
env_cluster: String,
ec2_cluster: Option<String>,
stored_task_arn: Option<&str>,
) -> Vec<String> {
let mut out = vec![env_cluster];
for c in [ec2_cluster, stored_task_arn.and_then(|a| ec2_cluster_of_task(env, a))]
.into_iter()
.flatten()
{
if !out.contains(&c) {
out.push(c);
}
}
out
}
/// Whether the caller's task role exists in the environment: the onboarding
/// signal a workspace user can read. A denied lookup is not an answer, so
/// it is let through (RunTask will say what is wrong).
@@ -237,6 +291,9 @@ pub struct RemoveRequest<'a> {
/// The EFS-mode `/workspace` access point the store has (kept after a
/// switch to Hybrid or Ephemeral, Q1).
pub stored_workspace_access_point_id: Option<&'a str>,
/// The user's EC2 cluster (their record's), looked in too: a project
/// may have run there.
pub ec2_cluster: Option<&'a str>,
/// Delete the EFS access points, home and `/workspace` (ruling R9). The
/// files stay on the user's own file system (only their task role can
/// read it) until offboarding destroys it; a later start of the same
@@ -258,17 +315,33 @@ pub async fn remove_workspace_resources<T: TaskApi, W: WorkspaceApi, R: RuntimeA
let cluster = environment_cluster_name(req.env);
let family = task_family(req.env, req.owner_key, req.project_id);
// 1. The task.
let live =
find_project_task(tasks, &cluster, &family, req.owner_user_id, req.project_id).await?;
let mut to_stop: Vec<String> = live.into_iter().map(|t| t.arn).collect();
if let Some(arn) = req.stored_task_arn {
if cluster_of_task(arn) == Some(cluster.as_str()) && !to_stop.iter().any(|a| a == arn) {
to_stop.push(arn.to_string());
// 1. The task: in the environment's cluster, the user's EC2 cluster (the
// record's, and the stored task's when that is a per-user cluster of
// this environment — the record may be gone).
let clusters = project_task_clusters(
req.env,
cluster.clone(),
req.ec2_cluster.map(str::to_string),
req.stored_task_arn,
);
let mut to_stop: Vec<(String, String)> = Vec::new();
for c in &clusters {
if let Some(t) =
find_project_task(tasks, c, &family, req.owner_user_id, req.project_id).await?
{
to_stop.push((c.clone(), t.arn));
}
}
for arn in &to_stop {
tasks.stop_task(&cluster, arn, STOP_REASON).await?;
if let Some(arn) = req.stored_task_arn {
let ours = cluster_of_task(arn) == Some(cluster.as_str())
|| ec2_cluster_of_task(req.env, arn).is_some();
if ours && !to_stop.iter().any(|(_, a)| a == arn) {
let c = cluster_of_task(arn).map_or_else(|| cluster.clone(), str::to_string);
to_stop.push((c, arn.to_string()));
}
}
for (c, arn) in &to_stop {
tasks.stop_task(c, arn, STOP_REASON).await?;
}
let mut failures: Vec<String> = Vec::new();
@@ -385,6 +458,10 @@ impl RuntimeApi for SdkRuntimeApi {
}
}
async fn user_record(&self, env: &str, owner_key: &str) -> Result<Param, AwsError> {
get_parameter(&self.ssm, &user_record_parameter_name(env, owner_key)).await
}
async fn role_exists(&self, name: &str) -> Result<Knowable, AwsError> {
let res = self.iam.get_role().role_name(name).send().await;
knowable("iam:GetRole", res)
+140 -2
View File
@@ -36,6 +36,8 @@ struct Fake {
revisions: Vec<TaskDefinitionRef>,
access_points: Vec<AccessPointRef>,
fail_secret_delete: bool,
/// The clusters ListTasks was asked about, by name.
lists: Mutex<Vec<String>>,
}
impl Fake {
@@ -52,6 +54,10 @@ impl RuntimeApi for Fake {
self.log(format!("record {env}"));
Ok(self.record.clone())
}
async fn user_record(&self, env: &str, owner_key: &str) -> Result<crate::aws::environments::Param, AwsError> {
self.log(format!("user-record {env} {owner_key}"));
Ok(crate::aws::environments::Param::Missing)
}
async fn role_exists(&self, name: &str) -> Result<Knowable, AwsError> {
self.log(format!("role {name}"));
Ok(self.role.clone().unwrap_or(Knowable::Yes))
@@ -85,8 +91,16 @@ impl TaskApi for Fake {
async fn describe(&self, _c: &str, arns: &[String]) -> Result<Vec<TaskInfo>, AwsError> {
Ok(self.live_tasks.iter().filter(|t| arns.contains(&t.arn)).cloned().collect())
}
async fn list_project_tasks(&self, _c: &str, _f: &str) -> Result<Vec<String>, AwsError> {
Ok(self.live_tasks.iter().map(|t| t.arn.clone()).collect())
async fn list_project_tasks(&self, c: &str, _f: &str) -> Result<Vec<String>, AwsError> {
// By cluster name or ARN, like ECS.
let name = c.rsplit_once(":cluster/").map_or(c, |(_, n)| n);
self.lists.lock().unwrap().push(name.to_string());
Ok(self
.live_tasks
.iter()
.filter(|t| cluster_of_task(&t.arn) == Some(name))
.map(|t| t.arn.clone())
.collect())
}
async fn stop_task(&self, cluster: &str, arn: &str, reason: &str) -> Result<(), AwsError> {
self.log(format!("stop {cluster} {arn} {reason}"));
@@ -262,6 +276,9 @@ fn live(id: &str) -> TaskInfo {
family()
),
runtime_id: None,
container_instance_arn: None,
exec_enabled: true,
created_unix: None,
exec_agent_running: false,
exit_code: None,
owner: Some(USER.into()),
@@ -305,6 +322,7 @@ async fn removal_stops_the_task_leaves_the_revisions_and_schedules_the_secret()
stored_task_arn: Some(&t0),
stored_access_point_id: Some("fsap-stored"),
stored_workspace_access_point_id: None,
ec2_cluster: None,
discard_data: false,
};
remove_workspace_resources(&fake, &fake, &fake, &r).await.unwrap();
@@ -341,6 +359,7 @@ async fn a_failed_step_is_reported_but_the_rest_still_runs() {
stored_task_arn: None,
stored_access_point_id: Some("fsap-1"),
stored_workspace_access_point_id: None,
ec2_cluster: None,
discard_data: true,
};
let e = remove_workspace_resources(&fake, &fake, &fake, &r).await.unwrap_err();
@@ -381,6 +400,7 @@ async fn removal_with_discard_deletes_both_access_points_and_without_keeps_both(
stored_access_point_id: Some("fsap-home"),
// Kept in the store after switching EFS -> Hybrid.
stored_workspace_access_point_id: Some("fsap-ws-stored"),
ec2_cluster: None,
discard_data: false,
};
let keep = Fake { access_points: points.clone(), ..Default::default() };
@@ -402,3 +422,121 @@ async fn removal_with_discard_deletes_both_access_points_and_without_keeps_both(
deleted.sort();
assert_eq!(deleted, vec!["fsap-home", "fsap-ws", "fsap-ws-stored"]);
}
// ------------------------------------------------------------- EC2 (phase 7)
fn ec2_task_arn(id: &str) -> String {
format!("arn:aws:ecs:{REGION}:{ACCOUNT}:task/triple-c-dev--u-alice/{id}")
}
#[test]
fn a_stored_task_on_a_per_user_cluster_names_that_cluster() {
assert_eq!(
ec2_cluster_of_task("dev", &ec2_task_arn("t")).as_deref(),
Some(format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/triple-c-dev--u-alice").as_str())
);
// The environment's own cluster, another environment's per-user
// cluster and anything else are not.
assert_eq!(ec2_cluster_of_task("dev", &task_arn("t")), None);
assert_eq!(ec2_cluster_of_task("prod", &ec2_task_arn("t")), None);
let odd = format!("arn:aws:ecs:{REGION}:{ACCOUNT}:task/triple-c-dev--u-a--b/t");
assert_eq!(ec2_cluster_of_task("dev", &odd), None);
assert_eq!(ec2_cluster_of_task("dev", "not-an-arn"), None);
}
#[tokio::test]
async fn the_user_record_is_read_and_validated() {
struct Rec(Param);
impl Rec {
async fn read(&self) -> UserRecord {
let fake = RecFake(self.0.clone(), Mutex::new(Vec::new()));
let r = read_user_record(&fake, "dev", &ok(), ACCOUNT, REGION).await.unwrap();
assert_eq!(*fake.1.lock().unwrap(), vec![format!("dev {}", ok())]);
r
}
}
let valid = serde_json::json!({
"schema": 1, "instance_type": "m7i.large", "max_instances": 2, "user_tag": "alice",
"cluster_arn": format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/triple-c-dev--u-alice"),
"capacity_provider": "triple-c-dev--u-alice",
})
.to_string();
let UserRecord::Valid(c) = Rec(Param::Value(valid)).read().await else { panic!() };
assert_eq!((c.instance_type.as_str(), c.max_instances), ("m7i.large", 2));
assert_eq!(Rec(Param::Missing).read().await, UserRecord::Absent);
assert!(matches!(Rec(Param::Denied).read().await, UserRecord::Unusable(_)));
assert!(matches!(Rec(Param::Value("{}".into())).read().await, UserRecord::Unusable(_)));
}
/// Only `user_record` is asked of it.
struct RecFake(Param, Mutex<Vec<String>>);
impl RuntimeApi for RecFake {
async fn environment_record(&self, _env: &str) -> Result<Option<String>, AwsError> {
unreachable!()
}
async fn user_record(&self, env: &str, owner_key: &str) -> Result<Param, AwsError> {
self.1.lock().unwrap().push(format!("{env} {owner_key}"));
Ok(self.0.clone())
}
async fn role_exists(&self, _name: &str) -> Result<Knowable, AwsError> {
unreachable!()
}
async fn secret_string(&self, _name: &str) -> Result<String, AwsError> {
unreachable!()
}
async fn access_point_state(&self, _id: &str) -> Result<Option<String>, AwsError> {
unreachable!()
}
async fn delete_secret(&self, _name: &str) -> Result<(), AwsError> {
unreachable!()
}
async fn delete_access_point(&self, _id: &str) -> Result<(), AwsError> {
unreachable!()
}
}
#[tokio::test]
async fn removal_finds_and_stops_a_task_on_the_users_ec2_cluster() {
let key = ok();
let user_cluster = format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/triple-c-dev--u-alice");
let base = RemoveRequest {
env: "dev",
region: REGION,
owner_user_id: USER,
owner_key: &key,
project_id: PROJECT,
stored_task_arn: None,
stored_access_point_id: None,
stored_workspace_access_point_id: None,
ec2_cluster: Some(&user_cluster),
discard_data: false,
};
// Found through the record's cluster.
let mut t = live("e1");
t.arn = ec2_task_arn("e1");
let fake = Fake { live_tasks: vec![t.clone()], ..Default::default() };
remove_workspace_resources(&fake, &fake, &fake, &base).await.unwrap();
assert_eq!(*fake.lists.lock().unwrap(), ["triple-c-dev", "triple-c-dev--u-alice"]);
let calls = fake.calls();
assert!(
calls.contains(&format!("stop {user_cluster} {} {STOP_REASON}", ec2_task_arn("e1"))),
"{calls:?}"
);
// The record is gone: the stored ARN still leads to its cluster.
let stored = ec2_task_arn("e1");
let fake = Fake { live_tasks: vec![t], ..Default::default() };
let r = RemoveRequest { ec2_cluster: None, stored_task_arn: Some(&stored), ..base };
remove_workspace_resources(&fake, &fake, &fake, &r).await.unwrap();
let stops: Vec<String> = fake.calls().into_iter().filter(|c| c.starts_with("stop")).collect();
assert_eq!(stops.len(), 1, "{stops:?}");
assert!(stops[0].contains(&ec2_task_arn("e1")), "{stops:?}");
// A stored ARN on someone else's kind of cluster is not stopped.
let foreign = format!("arn:aws:ecs:{REGION}:{ACCOUNT}:task/other-cluster/x");
let fake = Fake::default();
let r = RemoveRequest { ec2_cluster: None, stored_task_arn: Some(&foreign), ..base };
remove_workspace_resources(&fake, &fake, &fake, &r).await.unwrap();
assert!(!fake.calls().iter().any(|c| c.starts_with("stop")));
}
+127 -8
View File
@@ -25,7 +25,7 @@ use aws_sdk_ecs::types::{
ManagedAgentName, NetworkConfiguration, Tag, Task, TaskField,
};
use super::environments::EnvironmentInfo;
use super::environments::{EnvironmentInfo, UserCompute};
use super::error::{classify_sdk_error, AwsError};
pub use super::tofu_modules::task_log_stream;
use super::workspace::{workspace_tags, CONTAINER_NAME, TAG_OWNER, TAG_PROJECT};
@@ -62,8 +62,17 @@ pub struct TaskInfo {
pub task_definition_arn: String,
/// The `workspace` container's runtime id (needed for an SSM target).
pub runtime_id: Option<String>,
/// The EC2 container instance the task was placed on; `None` on Fargate,
/// and on EC2 until an instance has room for it.
pub container_instance_arn: Option<String>,
/// The `workspace` container's ExecuteCommandAgent is RUNNING.
pub exec_agent_running: bool,
/// ECS Exec is on for the task (`enableExecuteCommand`). ECS turns it
/// off on a RunTask that has to wait for an EC2 instance (verified live,
/// phase 7), and such a task can never be reached.
pub exec_enabled: bool,
/// When ECS created the task (Unix seconds), if it says.
pub created_unix: Option<i64>,
/// The `workspace` container's exit code, once it has one.
pub exit_code: Option<i32>,
/// Its `triple-c:owner` / `triple-c:project` tags (DescribeTasks with
@@ -112,6 +121,9 @@ pub(crate) fn task_info(t: &Task) -> TaskInfo {
stopped_reason: t.stopped_reason().map(str::to_string),
task_definition_arn: t.task_definition_arn().unwrap_or_default().to_string(),
runtime_id: workspace.and_then(|c| c.runtime_id()).map(str::to_string),
container_instance_arn: t.container_instance_arn().map(str::to_string),
exec_enabled: t.enable_execute_command(),
created_unix: t.created_at().map(|d| d.secs()),
exec_agent_running: workspace.is_some_and(|c| {
c.managed_agents().iter().any(|a| {
a.name() == Some(&ManagedAgentName::ExecuteCommandAgent)
@@ -216,6 +228,65 @@ pub fn run_task_request(
req.build().expect(BUILT)
}
/// Where a new task runs.
#[derive(Debug, Clone, Copy)]
pub enum RunOn<'a> {
/// The environment's cluster, Fargate (or `FARGATE_SPOT`).
Fargate { spot: bool },
/// The user's own cluster and capacity provider (phase 7), from their
/// record.
Ec2(&'a UserCompute),
}
impl RunOn<'_> {
/// The cluster a task runs on, and is looked for in.
pub fn cluster<'e>(&'e self, env: &'e EnvironmentInfo) -> &'e str {
match self {
RunOn::Fargate { .. } => &env.cluster_arn,
RunOn::Ec2(c) => &c.cluster_arn,
}
}
}
/// The RunTask request on the user's EC2 capacity: the exact revision, on
/// their cluster, placed by their capacity provider (weight 1). No launch
/// type (exclusive with a strategy) and no network configuration: EC2 tasks
/// use `bridge` networking (ruling R2). ECS Exec, tags, the client token and
/// no overrides as on Fargate (P1).
pub fn ec2_run_task_request(
compute: &UserCompute,
task_def_arn: &str,
tags: &[(String, String)],
) -> RunTaskInput {
const BUILT: &str = "every required field is set";
RunTaskInput::builder()
.cluster(&compute.cluster_arn)
.task_definition(task_def_arn)
.count(1)
.enable_execute_command(true)
.capacity_provider_strategy(
CapacityProviderStrategyItem::builder()
.capacity_provider(&compute.capacity_provider)
.weight(1)
.build()
.expect(BUILT),
)
.set_tags(Some(
tags.iter()
.map(|(k, v)| Tag::builder().key(k).value(v).build())
.collect(),
))
.client_token(uuid::Uuid::new_v4().to_string())
.build()
.expect(BUILT)
}
/// RunTask refused on a per-user cluster: the workspace policy lets a caller
/// run tasks only on the cluster named after their own `triple-c-user`
/// principal tag.
pub const EC2_TAG_MISMATCH: &str = "This EC2 capacity belongs to a different triple-c-user value \
than your sign-in's (or your sign-in has no triple-c-user tag). Ask an admin.";
/// The family of `arn:...:task-definition/<family>:<revision>`, when the
/// revision is a number. `None` for a bare family or anything else.
pub fn revision_family(arn: &str) -> Option<&str> {
@@ -271,14 +342,15 @@ pub struct TaskStart {
}
/// The project's live task if there is one (Review Focus 2: never a second
/// task for a project), else RunTask with the exact `task_def_arn`.
/// task for a project), else RunTask with the exact `task_def_arn` where
/// `run_on` says.
pub async fn find_or_run<A: TaskApi>(
api: &A,
env: &EnvironmentInfo,
owner_user_id: &str,
project_id: &str,
task_def_arn: &str,
spot: bool,
run_on: RunOn<'_>,
) -> Result<TaskStart, AwsError> {
let family = revision_family(task_def_arn).ok_or_else(|| {
other(format!(
@@ -291,20 +363,43 @@ pub async fn find_or_run<A: TaskApi>(
)));
}
if let Some(task) =
find_project_task(api, &env.cluster_arn, family, owner_user_id, project_id).await?
find_project_task(api, run_on.cluster(env), family, owner_user_id, project_id).await?
{
return Ok(TaskStart { task, reused: true });
}
let tags = workspace_tags(owner_user_id, project_id, &env.name);
let task = api
.run_task(run_task_request(env, task_def_arn, spot, &tags))
.await?;
let task = match run_on {
RunOn::Fargate { spot } => {
api.run_task(run_task_request(env, task_def_arn, spot, &tags)).await?
}
RunOn::Ec2(compute) => api
.run_task(ec2_run_task_request(compute, task_def_arn, &tags))
.await
.map_err(ec2_run_error)?,
};
Ok(TaskStart {
task,
reused: false,
})
}
/// A RunTask error on the user's EC2 capacity, in its terms: access denied is
/// the `triple-c-user` mismatch ([`EC2_TAG_MISMATCH`]); a placement failure
/// (`RESOURCE:CPU`/`RESOURCE:MEMORY`) means no instance can take the task.
pub fn ec2_run_error(e: AwsError) -> AwsError {
match e {
AwsError::AccessDenied { action, message } => AwsError::AccessDenied {
action,
message: format!("{EC2_TAG_MISMATCH} (AWS said: {message})"),
},
AwsError::Other { message } if message.contains("RESOURCE:") => other(format!(
"{message}. No instance of your EC2 capacity has room for this task: lower its CPU \
or memory reservation, or stop another EC2 workspace."
)),
e => e,
}
}
// ------------------------------------------------------------------ quotas
/// Whether an AWS message says an account limit or quota was hit.
@@ -386,6 +481,30 @@ impl Default for WaitTiming {
}
}
impl WaitTiming {
/// An EC2 task's wait: a cold start (capacity scaled to zero) launches
/// an instance and pulls the image before the task can run.
pub fn ec2() -> Self {
WaitTiming {
timeout: Duration::from_secs(10 * 60),
..Self::default()
}
}
}
/// An EC2 task still waiting for an instance with room: ECS keeps it
/// PROVISIONING (or PENDING) with no container instance while the capacity
/// provider's managed scaling launches one. (A Fargate task never has a
/// container instance, so ask this only of EC2 tasks.)
pub fn waiting_for_instance(t: &TaskInfo) -> bool {
t.container_instance_arn.is_none()
&& t.desired_status == "RUNNING"
&& matches!(t.last_status.as_str(), "PROVISIONING" | "PENDING")
}
/// How a [`wait_ready`] timeout begins.
pub const WAIT_TIMED_OUT: &str = "The task did not become ready within";
/// Where a project's task logs go: the environment's log group and the
/// project's `awslogs-stream-prefix` (Task 4's `log_stream_prefix`).
#[derive(Debug, Clone, PartialEq, Eq)]
@@ -440,7 +559,7 @@ pub async fn wait_ready<A: TaskApi>(
}
if tokio::time::Instant::now() + timing.interval > deadline {
return Err(other(format!(
"The task did not become ready within {} (last status {}{}). It may still \
"{WAIT_TIMED_OUT} {} (last status {}{}). It may still \
be starting; check the ECS console, or stop the project and start it again.",
describe_timeout(timing.timeout),
t.last_status,
+140 -7
View File
@@ -64,6 +64,9 @@ fn info(id: &str, last: &str) -> TaskInfo {
stopped_reason: None,
task_definition_arn: td_arn(3),
runtime_id: None,
container_instance_arn: None,
exec_enabled: true,
created_unix: None,
exec_agent_running: false,
exit_code: None,
owner: Some(USER.into()),
@@ -230,7 +233,7 @@ async fn an_existing_running_task_is_reused_and_nothing_is_run() {
..Default::default()
}
.describing(vec![vec![ready("t1")]]);
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(4), false)
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(4), RunOn::Fargate { spot: false })
.await
.unwrap();
assert!(got.reused);
@@ -246,7 +249,7 @@ async fn a_pending_task_is_reused_too() {
..Default::default()
}
.describing(vec![vec![info("t1", "PENDING")]]);
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), false)
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), RunOn::Fargate { spot: false })
.await
.unwrap();
assert!(got.reused);
@@ -272,7 +275,7 @@ async fn a_task_of_another_owner_or_project_in_the_family_is_never_reused() {
..Default::default()
}
.describing(vec![vec![foreign, other_project, untagged]]);
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), false)
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), RunOn::Fargate { spot: false })
.await
.unwrap();
assert!(!got.reused);
@@ -289,7 +292,7 @@ async fn a_task_already_stopping_is_not_reused() {
..Default::default()
}
.describing(vec![vec![stopping, info("t2", "DEPROVISIONING")]]);
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), false)
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), RunOn::Fargate { spot: false })
.await
.unwrap();
assert!(!got.reused);
@@ -312,7 +315,7 @@ async fn a_running_task_is_preferred_over_a_pending_one() {
#[tokio::test]
async fn nothing_listed_means_no_describe_and_a_run_with_the_exact_revision_and_tags() {
let api = Fake::default();
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(7), true)
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(7), RunOn::Fargate { spot: true })
.await
.unwrap();
assert!(!got.reused);
@@ -345,7 +348,7 @@ async fn a_bare_family_or_foreign_arn_is_refused_before_any_call() {
format!("arn:aws:ecs:{REGION}:{ACCOUNT}:task-definition/{FAMILY}:latest"),
"arn:aws:ecs:eu-west-1:1:task-definition/someone-else:1".to_string(),
] {
let err = find_or_run(&api, &env_info(), USER, PROJECT_ID, &bad, false)
let err = find_or_run(&api, &env_info(), USER, PROJECT_ID, &bad, RunOn::Fargate { spot: false })
.await
.unwrap_err();
assert!(matches!(err, AwsError::Other { .. }), "{bad}: {err:?}");
@@ -358,7 +361,7 @@ async fn a_quota_refusal_is_reported_once_without_retrying() {
let api = Fake::default().running(Err(quota_error(
"You've reached the limit on the number of vCPUs you can run concurrently",
)));
let err = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), false)
let err = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), RunOn::Fargate { spot: false })
.await
.unwrap_err();
assert!(is_quota_error(&err));
@@ -748,3 +751,133 @@ fn task_info_reads_the_workspace_container_and_its_exec_agent() {
assert_eq!(bare.exit_code, None);
assert_eq!(bare.owner, None);
}
// ------------------------------------------------------------- EC2 (phase 7)
fn user_cluster() -> String {
format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/triple-c-dev--u-alice")
}
fn compute() -> UserCompute {
UserCompute {
instance_type: "m7i.large".into(),
max_instances: 2,
user_tag: "alice".into(),
cluster_arn: user_cluster(),
capacity_provider: "triple-c-dev--u-alice".into(),
}
}
#[test]
fn an_ec2_run_uses_the_users_cluster_and_capacity_provider_and_no_network() {
let c = compute();
let req = ec2_run_task_request(&c, &td_arn(3), &tags());
assert_eq!(req.cluster(), Some(user_cluster().as_str()));
assert_eq!(req.task_definition(), Some(td_arn(3).as_str()));
assert_eq!(req.launch_type(), None, "exclusive with a capacity provider strategy");
let strategy = req.capacity_provider_strategy();
assert_eq!(strategy.len(), 1);
assert_eq!(strategy[0].capacity_provider(), "triple-c-dev--u-alice");
assert_eq!(strategy[0].weight(), 1);
assert!(req.network_configuration().is_none(), "bridge networking (R2)");
assert_eq!(req.enable_execute_command(), Some(true));
assert_eq!(req.count(), Some(1));
assert!(req.client_token().is_some_and(|t| !t.is_empty()));
let got: Vec<(String, String)> = req
.tags()
.iter()
.map(|t| (t.key().unwrap().to_string(), t.value().unwrap().to_string()))
.collect();
assert_eq!(got, tags());
assert!(req.overrides().is_none());
assert!(req.propagate_tags().is_none());
assert!(req.enable_ecs_managed_tags().is_none());
}
#[tokio::test]
async fn an_ec2_find_or_run_looks_and_runs_on_the_users_cluster() {
let api = Fake::default();
let c = compute();
let got = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), RunOn::Ec2(&c))
.await
.unwrap();
assert!(!got.reused);
assert_eq!(api.calls(), vec![format!("list {} {FAMILY}", user_cluster()), "run".to_string()]);
let req = &api.run_requests.lock().unwrap()[0];
assert_eq!(req.cluster(), Some(user_cluster().as_str()));
assert_eq!(req.capacity_provider_strategy()[0].capacity_provider(), "triple-c-dev--u-alice");
assert_eq!(RunOn::Ec2(&c).cluster(&env_info()), user_cluster());
assert_eq!(RunOn::Fargate { spot: true }.cluster(&env_info()), cluster());
}
#[tokio::test]
async fn access_denied_on_the_users_cluster_says_the_triple_c_user_value_is_wrong() {
let api = Fake::default().running(Err(AwsError::AccessDenied {
action: "ecs:RunTask".into(),
message: "not authorized to perform ecs:RunTask on cluster/triple-c-dev--u-alice".into(),
}));
let c = compute();
let e = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), RunOn::Ec2(&c))
.await
.unwrap_err();
let AwsError::AccessDenied { action, message } = &e else { panic!("{e:?}") };
assert_eq!(action, "ecs:RunTask");
assert!(message.starts_with(EC2_TAG_MISMATCH), "{message}");
assert!(message.contains("not authorized"), "AWS's words are kept: {message}");
assert!(EC2_TAG_MISMATCH.contains("different triple-c-user value"));
assert!(EC2_TAG_MISMATCH.contains("Ask an admin"));
// The same refusal on Fargate is left as it is.
let api = Fake::default().running(Err(AwsError::AccessDenied {
action: "ecs:RunTask".into(),
message: "nope".into(),
}));
let e = find_or_run(&api, &env_info(), USER, PROJECT_ID, &td_arn(3), RunOn::Fargate { spot: false })
.await
.unwrap_err();
assert_eq!(e, AwsError::AccessDenied { action: "ecs:RunTask".into(), message: "nope".into() });
}
#[test]
fn a_placement_failure_on_ec2_says_no_instance_has_room() {
let e = ec2_run_error(
run_task_outcome(Vec::new(), vec![(Some("RESOURCE:MEMORY".into()), None)]).unwrap_err(),
);
assert!(e.to_string().contains("RESOURCE:MEMORY"), "{e}");
assert!(e.to_string().contains("No instance of your EC2 capacity has room"), "{e}");
// Anything else passes through.
let q = quota_error("limit");
assert_eq!(ec2_run_error(q.clone()), q);
}
#[test]
fn an_ec2_task_without_an_instance_is_waiting_for_one() {
let provisioning = info("t", "PROVISIONING");
assert!(waiting_for_instance(&provisioning));
assert!(waiting_for_instance(&info("t", "PENDING")));
let placed = TaskInfo {
container_instance_arn: Some(format!(
"arn:aws:ecs:{REGION}:{ACCOUNT}:container-instance/triple-c-dev--u-alice/abc"
)),
..info("t", "PENDING")
};
assert!(!waiting_for_instance(&placed));
assert!(!waiting_for_instance(&info("t", "RUNNING")));
assert!(!waiting_for_instance(&info("t", "STOPPED")));
// The EC2 wait leaves room for an instance launch and an image pull.
assert_eq!(WaitTiming::ec2().timeout, Duration::from_secs(600));
assert_eq!(WaitTiming::ec2().interval, WaitTiming::default().interval);
}
#[test]
fn task_info_reads_the_container_instance() {
let t = Task::builder()
.task_arn(task_arn("t"))
.container_instance_arn("arn:aws:ecs:eu-west-1:111122223333:container-instance/c/1")
.build();
assert_eq!(
task_info(&t).container_instance_arn.as_deref(),
Some("arn:aws:ecs:eu-west-1:111122223333:container-instance/c/1")
);
assert_eq!(task_info(&Task::builder().build()).container_instance_arn, None);
}
+139 -7
View File
@@ -16,9 +16,16 @@ use crate::models::host_connection::validate_supported_region;
/// `locals.template_version` (and the `triple-c:template-version` default
/// tag) of `aws/terraform/environment`.
pub const ENVIRONMENT_MODULE_VERSION: u32 = 2;
/// The same for the user-onboarding module.
/// The same for the user-onboarding module. 2: optional per-user EC2
/// capacity (phase 7); an onboarding without it plans as version 1 did,
/// apart from this tag.
#[allow(dead_code)] // the onboarding module's version; checked against its HCL by a test
pub const ONBOARD_MODULE_VERSION: u32 = 1;
pub const ONBOARD_MODULE_VERSION: u32 = 2;
/// The same for the permissions-boundary module. 2: the instance role's ECS
/// agent actions, on its own cluster (phase 7, E4); an account admin
/// re-applies the boundary once.
#[allow(dead_code)] // checked against its HCL by a test; nothing reads the tag yet
pub const BOUNDARY_MODULE_VERSION: u32 = 2;
/// The permissions boundary every module-created role carries (no path).
/// Equal to [`super::policies::PERMISSIONS_BOUNDARY_NAME`]; a test checks it.
pub const BOUNDARY_POLICY_NAME: &str = "triple-c-permissions-boundary";
@@ -52,6 +59,7 @@ const USER_ONBOARD_FILES: &[(&str, &str)] = module!("user-onboard":
"main.tf",
"iam.tf",
"efs.tf",
"ec2.tf",
"outputs.tf",
".terraform.lock.hcl",
);
@@ -132,6 +140,100 @@ pub fn user_role_names(env: &str, owner_key: &str) -> (String, String) {
)
}
/// The principal tag every EC2 user's sign-in carries (phase 7, amended E1):
/// an admin sets it (an IAM Identity Center attribute for access control,
/// or a tag on the IAM user or role), and the workspace policy lets a
/// caller run tasks only on the per-user cluster named after its value.
#[allow(dead_code)] // the onboarding module (Task 3) and the UI read it
pub const EC2_USER_TAG_KEY: &str = "triple-c-user";
/// Checks a `triple-c-user` tag value: 1 to 32 characters of
/// `[A-Za-z0-9-]`, not starting or ending with a hyphen and without `--`
/// (the separator in [`user_ec2_stem`], which environment names can't hold
/// either, so a per-user name can be read back only one way).
pub fn validate_ec2_user_tag(tag: &str) -> Result<String, String> {
let ok = (1..=32).contains(&tag.len())
&& tag.bytes().all(|c| c.is_ascii_alphanumeric() || c == b'-')
&& !tag.starts_with('-')
&& !tag.ends_with('-')
&& !tag.contains("--");
if ok {
Ok(tag.to_string())
} else {
Err(format!(
"\"{}\" is not a valid triple-c-user tag: use 1 to 32 letters, digits and single \
hyphens, not starting or ending with a hyphen",
tag.escape_debug()
))
}
}
/// The name of a user's EC2 cluster, capacity provider, Auto Scaling group
/// and launch template: `triple-c-<env>--u-<user tag>` (at most 65
/// characters). Environment names and user tags never contain `--`, so the
/// workspace policy's `cluster/triple-c-*--u-*` matches exactly the
/// per-user clusters, and `triple-c-*--u-<tag>` exactly that user's.
pub fn user_ec2_stem(env: &str, user_tag: &str) -> String {
format!("triple-c-{env}--u-{user_tag}")
}
/// Whether `cluster` (a cluster *name*) is one of `env`'s per-user EC2
/// clusters ([`user_ec2_stem`] of a valid tag).
pub fn is_user_ec2_cluster(env: &str, cluster: &str) -> bool {
cluster
.strip_prefix(&format!("{}--u-", environment_cluster_name(env)))
.is_some_and(|tag| validate_ec2_user_tag(tag).is_ok())
}
/// A user's EC2 instance role (and instance profile): named by owner key,
/// not the user tag, because role names are account-global and at most 64
/// characters (this is at most 51). The admin's PassRole to EC2 names
/// `role/triple-c-*-instance`.
#[allow(dead_code)] // the onboarding module builds it; pinned by tests
pub fn user_instance_role_name(env: &str, owner_key: &str) -> String {
format!("triple-c-{env}-{owner_key}-instance")
}
/// The SSM parameter holding a user's EC2 record (E11), written by their
/// onboarding module only while it has EC2 capacity: a String of JSON with
/// exactly [`USER_RECORD_FIELDS`]. Workspace users can read it
/// (`parameter/triple-c/*`) though not the state bucket; it lives outside
/// `/triple-c/environments/`, which environment detection lists.
#[allow(dead_code)] // the app reads it from Task 4 on
pub fn user_record_parameter_name(env: &str, owner_key: &str) -> String {
format!("/triple-c/users/{env}/{owner_key}")
}
/// The fields of a user's EC2 record, schema 1: `instance_type`,
/// `cluster_arn` (the user's cluster, [`user_ec2_stem`]),
/// `capacity_provider` (its one provider, the same name) and `user_tag`
/// (the `triple-c-user` value the cluster is named after) and
/// `max_instances` (the group's maximum), so a re-apply can keep them.
/// `environments::parse_user_record` reads them; a test pins the HCL to it.
#[allow(dead_code)]
pub const USER_RECORD_FIELDS: [&str; 6] = [
"schema",
"instance_type",
"cluster_arn",
"capacity_provider",
"user_tag",
"max_instances",
];
/// The most instances a user's EC2 capacity may run at once.
pub const EC2_MAX_INSTANCES: u32 = 4;
/// A user's optional EC2 capacity, as the onboarding module takes it.
#[derive(Clone, Debug, Serialize, Deserialize, PartialEq, Eq)]
pub struct Ec2Compute {
/// One of the types `ec2_instance_spec` knows.
pub instance_type: String,
/// 1 to [`EC2_MAX_INSTANCES`].
pub max_instances: u32,
/// The user's `triple-c-user` principal tag value.
pub user_tag: String,
}
/// Whether a project id is used in names as it is: 1 to 64 characters of
/// `[A-Za-z0-9-]`. Every UUID (how the app makes ids) is.
fn is_clean_project_id(id: &str) -> bool {
@@ -347,16 +449,46 @@ pub fn environment_vars(
}
/// The onboarding module's `terraform.tfvars.json` for the user whose
/// `aws:userid` is `owner_id`. Bedrock stays on the workspace user's own
/// credentials unless a later setting opts the task role in.
pub fn onboarding_vars(env: &str, region: &str, owner_id: &str) -> serde_json::Value {
json!({
/// `aws:userid` is `owner_id`, with their EC2 capacity or none (empty
/// `ec2_instance_type`: the module makes nothing for EC2). Applies the same
/// rules as the module's EC2 variable validations. Bedrock stays on the
/// workspace user's own credentials unless a later setting opts the task
/// role in.
pub fn onboarding_vars(
env: &str,
region: &str,
owner_id: &str,
ec2: Option<&Ec2Compute>,
) -> Result<serde_json::Value, String> {
let (instance_type, max_instances, user_tag) = match ec2 {
None => (String::new(), 1, String::new()),
Some(c) => {
if crate::models::project::ec2_instance_spec(&c.instance_type).is_none() {
return Err(format!(
"\"{}\" is not an instance type Triple-C supports (m7i, m7g, c7i, c7g, r7i, \
r7g from large to 4xlarge; t3, t4g from large to 2xlarge)",
c.instance_type.escape_debug()
));
}
if !(1..=EC2_MAX_INSTANCES).contains(&c.max_instances) {
return Err(format!(
"The number of instances must be from 1 to {EC2_MAX_INSTANCES}"
));
}
let tag = validate_ec2_user_tag(&c.user_tag)?;
(c.instance_type.clone(), c.max_instances, tag)
}
};
Ok(json!({
"environment_name": env,
"region": region,
"owner_id": owner_id,
"owner_key": owner_key(region, owner_id),
"bedrock_on_task_role": false,
})
"ec2_instance_type": instance_type,
"ec2_max_instances": max_instances,
"ec2_user_tag": user_tag,
}))
}
/// The boundary module's `terraform.tfvars.json`: the shipped boundary
File diff suppressed because it is too large. Load diff
+182 -32
View File
@@ -27,10 +27,11 @@ use std::future::Future;
use aws_config::SdkConfig;
use aws_sdk_ecs::operation::register_task_definition::RegisterTaskDefinitionInput;
use aws_sdk_ecs::types::{
Compatibility, ContainerDefinition, CpuArchitecture, EfsAuthorizationConfig,
EfsAuthorizationConfigIam, EfsTransitEncryption, EfsVolumeConfiguration, EphemeralStorage,
HealthCheck, KernelCapabilities, KeyValuePair, LinuxParameters, LogConfiguration, LogDriver,
MountPoint, NetworkMode, OsFamily, RuntimePlatform, Secret, SortOrder, Tag,
Compatibility, ContainerDefinition, CpuArchitecture, Device, DeviceCgroupPermission,
EfsAuthorizationConfig, EfsAuthorizationConfigIam, EfsTransitEncryption,
EfsVolumeConfiguration, EphemeralStorage, HealthCheck, HostVolumeProperties,
KernelCapabilities, KeyValuePair, LinuxParameters, LogConfiguration, LogDriver, MountPoint,
NetworkMode, OsFamily, RuntimePlatform, Secret, SortOrder, SystemControl, Tag,
TaskDefinitionField, TaskDefinitionStatus, Volume,
};
use aws_sdk_secretsmanager::error::ProvideErrorMetadata;
@@ -39,7 +40,7 @@ use rand::RngCore;
use serde::Serialize;
use sha2::{Digest, Sha256};
use super::environments::EnvironmentInfo;
use super::environments::{EnvironmentInfo, UserCompute};
use super::error::{classify_sdk_error, AwsError};
use super::identity::CallerIdentity;
use super::tofu_modules::{
@@ -47,7 +48,10 @@ use super::tofu_modules::{
user_file_system_token, user_role_names, workspace_access_point_root,
};
use crate::host::ssm::AGENT_PORT;
use crate::models::project::{validate_aws_workspace, Arch, RepoSource, SecretsDelivery, StorageMode};
use crate::models::project::{
ec2_fit_problem, fits_instance, validate_aws_workspace, Arch, Fit, LaunchType, RepoSource,
SecretsDelivery, StorageMode,
};
use crate::models::Project;
pub const TAG_OWNER: &str = "triple-c:owner";
@@ -103,6 +107,11 @@ pub const MAX_GATEWAY_CONFIG_BYTES: usize = 8_192;
/// Seconds ECS waits after SIGTERM before SIGKILL (Fargate allows 2–120).
pub const STOP_TIMEOUT_SECS: i32 = 120;
/// The instance's Docker socket, mounted at the same path in an EC2 task
/// with Docker access (E6).
pub const DOCKER_SOCKET_PATH: &str = "/var/run/docker.sock";
const DOCKER_SOCKET_VOLUME: &str = "docker-socket";
/// The task definition's whole `environment` (R2): constants, no project data.
const FIXED_ENVIRONMENT: [(&str, &str); 2] = [("TRIPLEC_HOST", "aws"), ("COLORTERM", "truecolor")];
@@ -205,6 +214,12 @@ pub fn gateway_sidecar_problem(
if !project.aws.as_ref().is_some_and(|a| a.gateway_sidecar) {
return None;
}
if project.aws.as_ref().is_some_and(|a| a.launch_type == LaunchType::Ec2) {
// EC2 tasks use `bridge` networking (ruling R2): the containers of a
// task don't share a loopback, and the sidecar must never listen
// beyond it. Not offered on EC2 yet.
return Some(EC2_NO_SIDECAR.into());
}
if project.backend != crate::models::project::Backend::OpenAiCompatible {
return Some(
"The LLM gateway sidecar needs the OpenAI Compatible backend: choose it in the \
@@ -222,6 +237,10 @@ pub fn gateway_sidecar_problem(
None
}
/// Why an EC2 project can't run the gateway sidecar (yet).
pub const EC2_NO_SIDECAR: &str = "The LLM gateway sidecar is not available on EC2 yet: turn it \
off, or run the project on Fargate";
/// `workspace_env`'s fingerprint of the custom variables' values.
const CUSTOM_ENV_FINGERPRINT: &str = "TRIPLE_C_CUSTOM_ENV";
@@ -473,6 +492,21 @@ pub struct WorkspacePlan {
/// The gateway sidecar's image (Q7) when the project runs it; `None`
/// leaves the request exactly as it was before the sidecar existed.
pub gateway_image: Option<String>,
/// Run on the user's own EC2 instances (phase 7) rather than Fargate;
/// `None` leaves the request exactly as it was before EC2 existed.
pub ec2: Option<Ec2Plan>,
}
/// What an EC2 task definition adds (E6, E7). `cpu`/`memory` are the task's
/// reservations on the instance; arch and ephemeral storage don't apply (the
/// instance decides both).
#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize)]
pub struct Ec2Plan {
/// Mount the instance's Docker socket (`allow_docker_access`, E6).
pub docker_socket: bool,
/// What a VPN client needs (`vpn_support_enabled`, E7): `NET_ADMIN`,
/// `/dev/net/tun`, `net.ipv4.conf.all.src_valid_mark=1`.
pub vpn: bool,
}
/// sha256 (hex) over the task definition `plan` produces, minus the label
@@ -583,12 +617,7 @@ fn build_request(plan: &WorkspacePlan) -> RegisterTaskDefinitionInput {
.build()
.expect(BUILT),
)
.linux_parameters(
LinuxParameters::builder()
.init_process_enabled(true)
.capabilities(KernelCapabilities::builder().add("SYS_PTRACE").build())
.build(),
)
.linux_parameters(linux_parameters(plan.ec2))
.log_configuration(
LogConfiguration::builder()
.log_driver(LogDriver::Awslogs)
@@ -616,29 +645,64 @@ fn build_request(plan: &WorkspacePlan) -> RegisterTaskDefinitionInput {
.build(),
);
}
let docker_socket = plan.ec2.is_some_and(|e| e.docker_socket);
if docker_socket {
container = container.mount_points(
MountPoint::builder()
.source_volume(DOCKER_SOCKET_VOLUME)
.container_path(DOCKER_SOCKET_PATH)
.read_only(false)
.build(),
);
}
if plan.ec2.is_some_and(|e| e.vpn) {
// Namespaced `net.*` sysctls are allowed in `bridge` mode: each
// container has its own network namespace, so this sets the
// workspace's, never the instance's (ECS warns only for `host`, and
// for several containers of one `awsvpc` task).
for (key, value) in crate::docker::container::vpn_requirements().sysctls {
container = container
.system_controls(SystemControl::builder().namespace(key).value(value).build());
}
}
let mut req = RegisterTaskDefinitionInput::builder()
.family(&plan.family)
.task_role_arn(&plan.task_role_arn)
.execution_role_arn(&plan.exec_role_arn)
.network_mode(NetworkMode::Awsvpc)
.requires_compatibilities(Compatibility::Fargate)
.cpu(plan.cpu.to_string())
.memory(plan.memory.to_string())
.runtime_platform(
RuntimePlatform::builder()
.cpu_architecture(match plan.arch {
Arch::X86_64 => CpuArchitecture::X8664,
Arch::Arm64 => CpuArchitecture::Arm64,
})
.operating_system_family(OsFamily::Linux)
.build(),
)
.ephemeral_storage(
EphemeralStorage::builder()
.size_in_gib(plan.ephemeral_gib as i32)
.build(),
)
.execution_role_arn(&plan.exec_role_arn);
req = match plan.ec2 {
// Fargate: exactly the request from before EC2 existed (pinned).
None => req
.network_mode(NetworkMode::Awsvpc)
.requires_compatibilities(Compatibility::Fargate)
.cpu(plan.cpu.to_string())
.memory(plan.memory.to_string())
.runtime_platform(
RuntimePlatform::builder()
.cpu_architecture(match plan.arch {
Arch::X86_64 => CpuArchitecture::X8664,
Arch::Arm64 => CpuArchitecture::Arm64,
})
.operating_system_family(OsFamily::Linux)
.build(),
)
.ephemeral_storage(
EphemeralStorage::builder()
.size_in_gib(plan.ephemeral_gib as i32)
.build(),
),
// EC2 (ruling R2: `bridge`, the instance is the user's alone). No
// runtime platform — the instance's architecture decides and the
// image is multi-arch — and no ephemeral storage (the instance's
// disk). Task-level cpu/memory are the reservations `fits_instance`
// checked against the user's instance type.
Some(_) => req
.network_mode(NetworkMode::Bridge)
.requires_compatibilities(Compatibility::Ec2)
.cpu(plan.cpu.to_string())
.memory(plan.memory.to_string()),
};
let mut req = req
.container_definitions(container.build())
.set_tags(Some(
plan.tags
@@ -655,9 +719,45 @@ fn build_request(plan: &WorkspacePlan) -> RegisterTaskDefinitionInput {
{
req = req.volumes(efs_volume(name, fs, ap));
}
if docker_socket {
req = req.volumes(
Volume::builder()
.name(DOCKER_SOCKET_VOLUME)
.host(HostVolumeProperties::builder().source_path(DOCKER_SOCKET_PATH).build())
.build(),
);
}
req.build().expect(BUILT)
}
/// The workspace container's Linux parameters: docker-init and `SYS_PTRACE`
/// always; on EC2 with VPN support, [`vpn_requirements`]' capability and
/// device too (Fargate allows neither).
///
/// [`vpn_requirements`]: crate::docker::container::vpn_requirements
fn linux_parameters(ec2: Option<Ec2Plan>) -> LinuxParameters {
let mut caps = KernelCapabilities::builder().add("SYS_PTRACE");
let mut lp = LinuxParameters::builder().init_process_enabled(true);
if ec2.is_some_and(|e| e.vpn) {
let vpn = crate::docker::container::vpn_requirements();
for cap in vpn.capabilities {
caps = caps.add(cap);
}
for (host, container, perms) in vpn.devices {
let mut device = Device::builder().host_path(host).container_path(container);
for p in perms.chars() {
device = device.permissions(match p {
'r' => DeviceCgroupPermission::Read,
'w' => DeviceCgroupPermission::Write,
_ => DeviceCgroupPermission::Mknod,
});
}
lp = lp.devices(device.build().expect("every required field is set"));
}
}
lp.capabilities(caps.build()).build()
}
/// `secrets` entries for `keys`, each by `valueFrom: <arn>:<KEY>::`.
fn secrets<'k>(plan: &WorkspacePlan, keys: impl Iterator<Item = &'k String>) -> Vec<Secret> {
keys.map(|k| {
@@ -902,9 +1002,56 @@ pub struct PreparedWorkspace {
pub secret_arn: String,
}
/// Why an EC2 project can't start on `compute` (the user's record, `None`
/// when they have none), or what its task definition adds. A Fargate project
/// gets `Ok(None)` whatever `compute` is.
pub fn ec2_plan(
project: &Project,
env: &str,
compute: Option<&UserCompute>,
) -> Result<Option<Ec2Plan>, String> {
let Some(cfg) = project.aws.as_ref().filter(|c| c.launch_type == LaunchType::Ec2) else {
return Ok(None);
};
if cfg.gateway_sidecar {
return Err(EC2_NO_SIDECAR.into());
}
let compute = compute.ok_or_else(|| no_ec2_capacity(env))?;
match fits_instance(cfg, &compute.instance_type) {
Fit::Ok => {}
Fit::TooBig { .. } => {
let problem = ec2_fit_problem(cfg, &compute.instance_type).unwrap_or_default();
return Err(format!(
"{problem} Lower the reservations in the project's Host settings, or ask an \
admin for a larger instance type."
));
}
// The record is validated against the same table, so this is a
// newer app's type: let ECS decide.
Fit::UnknownType => log::warn!(
"Project {}: instance type {} is not in the app's table; its fit is unchecked",
project.id,
compute.instance_type
),
}
Ok(Some(Ec2Plan {
docker_socket: project.allow_docker_access,
vpn: project.vpn_support_enabled,
}))
}
/// The user has no EC2 record in `env` (none was added, or it was removed).
pub fn no_ec2_capacity(env: &str) -> String {
format!(
"You have no EC2 capacity in {env}; ask an admin to add it to your onboarding (or run \
the project on Fargate)."
)
}
/// Make the project's secret, access points (Hybrid, EFS) and task definition
/// ready. Everything that can be checked locally — placement, sizes, the
/// payload's size and names — is checked before the first write.
/// payload's size and names, an EC2 project's fit on `compute` — is checked
/// before the first write.
#[allow(clippy::too_many_arguments)]
pub async fn prepare_workspace<W: WorkspaceApi>(
api: &W,
@@ -916,6 +1063,7 @@ pub async fn prepare_workspace<W: WorkspaceApi>(
custom_env_keys: &[String],
image: &str,
gateway: Option<GatewayPayload<'_>>,
compute: Option<&UserCompute>,
) -> Result<PreparedWorkspace, AwsError> {
let (Some(placement), Some(cfg)) = (project.aws_placement.as_ref(), project.aws.as_ref())
else {
@@ -940,6 +1088,7 @@ pub async fn prepare_workspace<W: WorkspaceApi>(
)));
}
validate_aws_workspace(cfg).map_err(other)?;
let ec2 = ec2_plan(project, &env.name, compute).map_err(other)?;
if image.trim().is_empty() {
return Err(other("No container image is set for AWS"));
}
@@ -1100,6 +1249,7 @@ pub async fn prepare_workspace<W: WorkspaceApi>(
workspace_access_point_id: workspace_access_point_id.clone(),
tags,
gateway_image: gateway.map(|_| crate::docker::gateway::registry_image_ref()),
ec2,
};
let fingerprint = workspace_fingerprint(&plan);
let task_definition_arn =
+236 -13
View File
@@ -102,6 +102,7 @@ fn sample_plan(storage: StorageMode) -> WorkspacePlan {
.then(|| "fsap-4567efab".to_string()),
tags: workspace_tags(USER, PROJECT_ID, "dev"),
gateway_image: None,
ec2: None,
}
}
@@ -962,7 +963,7 @@ async fn prepare(
p: &Project,
env: &[(String, String)],
) -> Result<PreparedWorkspace, AwsError> {
prepare_workspace(fake, &who(), &env_info(), p, env, &[], &[], IMAGE, None).await
prepare_workspace(fake, &who(), &env_info(), p, env, &[], &[], IMAGE, None, None).await
}
#[tokio::test]
@@ -972,7 +973,7 @@ async fn prepare_puts_the_setup_script_and_the_environments_cap_in_the_secret()
p.aws.as_mut().unwrap().setup_script = Some("sudo apt-get install -y jq".into());
let mut env = env_info();
env.max_workspace_hours = 12;
prepare_workspace(&fake, &who(), &env, &p, &[], &[], &[], IMAGE, None).await.unwrap();
prepare_workspace(&fake, &who(), &env, &p, &[], &[], &[], IMAGE, None, None).await.unwrap();
let m = payload_map(&fake.secrets.lock().unwrap()[0].2);
assert_eq!(m["TRIPLEC_SETUP_SCRIPT"], "sudo apt-get install -y jq");
assert_eq!(m["TRIPLEC_MAX_LIFETIME_MINUTES"], "720");
@@ -1397,6 +1398,7 @@ async fn files_travel_in_the_secret_only() {
&[],
IMAGE,
None,
None,
)
.await
.unwrap();
@@ -1638,7 +1640,7 @@ async fn prepare_leaves_the_credentials_out_of_a_copy_at_connect_secret() {
p.aws.as_mut().unwrap().secrets_delivery = crate::models::project::SecretsDelivery::CopyAtConnect;
let files = vec![("ssh/id_ed25519".to_string(), b"PRIVATE KEY".to_vec())];
let env = pairs(&[("GIT_TOKEN", "ghp_secret"), ("MINE", "x"), ("TZ", "UTC")]);
prepare_workspace(&fake, &who(), &env_info(), &p, &env, &files, &["MINE".to_string()], IMAGE, None)
prepare_workspace(&fake, &who(), &env_info(), &p, &env, &files, &["MINE".to_string()], IMAGE, None, None)
.await
.unwrap();
let m = payload_map(&fake.secrets.lock().unwrap()[0].2);
@@ -1662,7 +1664,7 @@ async fn switching_to_copy_at_connect_replaces_a_secret_that_held_credentials()
Some(r#"{"GIT_TOKEN":"ghp_old","TRIPLEC_AGENT_TOKEN":"t","TZ":"UTC"}"#.into());
*fake.pending_puts.lock().unwrap() = 2; // AWS still finishing the delete
let env = pairs(&[("GIT_TOKEN", "ghp_new"), ("TZ", "UTC")]);
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None)
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None, None)
.await
.unwrap();
assert_eq!(*fake.secret_reads.lock().unwrap(), 1, "read only once it exists");
@@ -1675,7 +1677,7 @@ async fn switching_to_copy_at_connect_replaces_a_secret_that_held_credentials()
*fake.existing_secret.lock().unwrap() = Some(
r#"{"TRIPLEC_AGENT_TOKEN":"t","TRIPLEC_CREDENTIALS":"connect","TZ":"UTC","TRIPLEC_REPOS":"[]"}"#.into(),
);
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None)
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None, None)
.await
.unwrap();
assert!(fake.deleted_secrets.lock().unwrap().is_empty());
@@ -1684,7 +1686,7 @@ async fn switching_to_copy_at_connect_replaces_a_secret_that_held_credentials()
// No secret yet (a first start): created, never read — a read of a
// secret that doesn't exist may be refused — nothing deleted.
let fake = Fake::onboarded();
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None)
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None, None)
.await
.unwrap();
assert_eq!(*fake.creates.lock().unwrap(), 1);
@@ -1737,7 +1739,7 @@ async fn the_start_after_a_timed_out_replace_retries_the_create() {
let env = pairs(&[("TZ", "UTC")]);
let fake = Fake::onboarded();
*fake.pending_puts.lock().unwrap() = 3;
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None)
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None, None)
.await
.unwrap();
assert_eq!(fake.secrets.lock().unwrap().len(), 1, "made once the name is free");
@@ -1752,7 +1754,7 @@ async fn the_start_after_a_timed_out_replace_retries_the_create() {
// Still taken past the limit: try again in a minute, never "restore".
let fake = Fake::onboarded();
*fake.pending_puts.lock().unwrap() = 1000;
let e = prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None)
let e = prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None, None)
.await
.unwrap_err()
.to_string();
@@ -1770,7 +1772,7 @@ async fn a_copy_at_connect_read_of_a_secret_being_deleted_is_retried() {
r#"{"TRIPLEC_AGENT_TOKEN":"t","TRIPLEC_CREDENTIALS":"connect","TZ":"UTC","TRIPLEC_REPOS":"[]"}"#.into(),
);
*fake.pending_reads.lock().unwrap() = 3;
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None)
prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None, None)
.await
.unwrap();
assert_eq!(*fake.secret_reads.lock().unwrap(), 4);
@@ -1780,7 +1782,7 @@ async fn a_copy_at_connect_read_of_a_secret_being_deleted_is_retried() {
let fake = Fake::onboarded();
*fake.existing_secret.lock().unwrap() = Some(r#"{"TZ":"UTC"}"#.into());
*fake.pending_reads.lock().unwrap() = 1000;
let e = prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None)
let e = prepare_workspace(&fake, &who(), &env_info(), &copy_at_connect_project(), &env, &[], &[], IMAGE, None, None)
.await
.unwrap_err()
.to_string();
@@ -1985,7 +1987,7 @@ fn sidecar_project() -> Project {
async fn prepare_registers_the_sidecar_only_when_the_project_runs_it() {
let fake = Fake::onboarded();
let p = sidecar_project();
prepare_workspace(&fake, &who(), &env_info(), &p, &[], &[], &[], IMAGE, Some(gw()))
prepare_workspace(&fake, &who(), &env_info(), &p, &[], &[], &[], IMAGE, Some(gw()), None)
.await
.unwrap();
assert_eq!(fake.registered.lock().unwrap()[0].container_definitions().len(), 2);
@@ -1994,7 +1996,7 @@ async fn prepare_registers_the_sidecar_only_when_the_project_runs_it() {
// Toggle off: a payload handed in anyway is ignored.
let fake = Fake::onboarded();
prepare_workspace(&fake, &who(), &env_info(), &project(StorageMode::Ephemeral), &[], &[], &[], IMAGE, Some(gw()))
prepare_workspace(&fake, &who(), &env_info(), &project(StorageMode::Ephemeral), &[], &[], &[], IMAGE, Some(gw()), None)
.await
.unwrap();
assert_eq!(fake.registered.lock().unwrap()[0].container_definitions().len(), 1);
@@ -2006,7 +2008,7 @@ async fn prepare_registers_the_sidecar_only_when_the_project_runs_it() {
async fn prepare_refuses_a_sidecar_without_its_payload_or_with_an_oversized_config() {
let fake = Fake::onboarded();
let p = sidecar_project();
let e = prepare_workspace(&fake, &who(), &env_info(), &p, &[], &[], &[], IMAGE, None)
let e = prepare_workspace(&fake, &who(), &env_info(), &p, &[], &[], &[], IMAGE, None, None)
.await
.unwrap_err();
assert!(e.to_string().contains("gateway"), "{e}");
@@ -2021,6 +2023,7 @@ async fn prepare_refuses_a_sidecar_without_its_payload_or_with_an_oversized_conf
&[],
IMAGE,
Some(GatewayPayload { config: &big, ..gw() }),
None,
)
.await
.unwrap_err();
@@ -2035,6 +2038,7 @@ async fn prepare_refuses_a_sidecar_without_its_payload_or_with_an_oversized_conf
&[],
IMAGE,
Some(GatewayPayload { api_key: " ", ..gw() }),
None,
)
.await
.unwrap_err();
@@ -2084,3 +2088,222 @@ fn the_sidecars_token_is_not_a_leftover_credential_but_a_dropped_provider_key_is
[("TRIPLEC_AGENT_TOKEN", "t")].iter().map(|(k, v)| (k.to_string(), v.to_string())).collect();
assert!(holds(sidecar_old, &off));
}
// ------------------------------------------------------------- EC2 (phase 7)
fn ec2_sample(storage: StorageMode, docker_socket: bool, vpn: bool) -> WorkspacePlan {
WorkspacePlan { ec2: Some(Ec2Plan { docker_socket, vpn }), ..sample_plan(storage) }
}
fn compute(instance_type: &str) -> UserCompute {
UserCompute {
instance_type: instance_type.into(),
max_instances: 2,
user_tag: "alice".into(),
cluster_arn: format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/triple-c-dev--u-alice"),
capacity_provider: "triple-c-dev--u-alice".into(),
}
}
fn ec2_project(storage: StorageMode, cpu: u32, memory_mib: u32) -> Project {
let mut p = project(storage);
let a = p.aws.as_mut().unwrap();
a.launch_type = crate::models::LaunchType::Ec2;
a.cpu = cpu;
a.memory_mib = memory_mib;
p
}
#[test]
fn an_ec2_request_is_bridge_mode_with_the_reservations_and_no_fargate_fields() {
for storage in [StorageMode::Ephemeral, StorageMode::Hybrid, StorageMode::Efs] {
let plan = ec2_sample(storage, false, false);
let req = task_definition_request(&plan);
assert_eq!(req.requires_compatibilities(), &[aws_sdk_ecs::types::Compatibility::Ec2]);
assert_eq!(req.network_mode(), Some(&aws_sdk_ecs::types::NetworkMode::Bridge));
assert_eq!((req.cpu(), req.memory()), (Some("1024"), Some("4096")));
assert!(req.runtime_platform().is_none(), "the instance's architecture decides");
assert!(req.ephemeral_storage().is_none(), "the instance's disk");
// Everything else as on Fargate: roles, tags, health check, stop
// timeout, init, logs, the EFS volumes and the fingerprint label.
let fargate = task_definition_request(&sample_plan(storage));
assert_eq!(req.task_role_arn(), fargate.task_role_arn());
assert_eq!(req.execution_role_arn(), fargate.execution_role_arn());
assert_eq!(req.tags(), fargate.tags());
assert_eq!(req.volumes(), fargate.volumes());
let (c, f) = (&req.container_definitions()[0], &fargate.container_definitions()[0]);
assert_eq!(c.health_check(), f.health_check());
assert_eq!(c.stop_timeout(), Some(STOP_TIMEOUT_SECS));
assert_eq!(c.mount_points(), f.mount_points());
assert_eq!(c.secrets(), f.secrets());
assert_eq!(c.environment(), f.environment());
assert_eq!(c.log_configuration(), f.log_configuration());
assert_eq!(c.linux_parameters(), f.linux_parameters(), "SYS_PTRACE and init only");
assert!(c.system_controls().is_empty());
assert!(c.port_mappings().is_empty());
assert_eq!(
c.docker_labels().unwrap().get(FINGERPRINT_LABEL),
Some(&workspace_fingerprint(&plan))
);
assert_ne!(workspace_fingerprint(&plan), workspace_fingerprint(&sample_plan(storage)));
}
}
#[test]
fn an_ec2_request_for_each_toggle_combination() {
let mut fingerprints = std::collections::BTreeSet::new();
for (docker, vpn) in [(false, false), (true, false), (false, true), (true, true)] {
let plan = ec2_sample(StorageMode::Hybrid, docker, vpn);
let req = task_definition_request(&plan);
fingerprints.insert(workspace_fingerprint(&plan));
let c = &req.container_definitions()[0];
let lp = c.linux_parameters().unwrap();
assert_eq!(lp.init_process_enabled(), Some(true));
// E6: the instance's Docker socket at the same path.
let socket_volume: Vec<_> =
req.volumes().iter().filter(|v| v.name() == Some("docker-socket")).collect();
let socket_mount: Vec<_> = c
.mount_points()
.iter()
.filter(|m| m.container_path() == Some(DOCKER_SOCKET_PATH))
.collect();
if docker {
assert_eq!(socket_volume.len(), 1);
assert_eq!(
socket_volume[0].host().and_then(|h| h.source_path()),
Some("/var/run/docker.sock")
);
assert!(socket_volume[0].efs_volume_configuration().is_none());
assert_eq!(socket_mount.len(), 1);
assert_eq!(socket_mount[0].source_volume(), Some("docker-socket"));
} else {
assert!(socket_volume.is_empty() && socket_mount.is_empty());
}
// The home access point is mounted either way.
assert!(c.mount_points().iter().any(|m| m.container_path() == Some(HOME_MOUNT_PATH)));
// E7: the three VPN pieces, all or none, plus SYS_PTRACE as always.
let caps = lp.capabilities().unwrap().add();
let sysctls: Vec<(&str, &str)> = c
.system_controls()
.iter()
.map(|s| (s.namespace().unwrap(), s.value().unwrap()))
.collect();
if vpn {
assert_eq!(caps, ["SYS_PTRACE".to_string(), "NET_ADMIN".to_string()]);
assert_eq!(lp.devices().len(), 1);
let d = &lp.devices()[0];
assert_eq!((d.host_path(), d.container_path()), ("/dev/net/tun", Some("/dev/net/tun")));
use aws_sdk_ecs::types::DeviceCgroupPermission as P;
assert_eq!(d.permissions(), [P::Read, P::Write, P::Mknod]);
assert_eq!(sysctls, [("net.ipv4.conf.all.src_valid_mark", "1")]);
} else {
assert_eq!(caps, ["SYS_PTRACE".to_string()]);
assert!(lp.devices().is_empty() && sysctls.is_empty());
}
}
assert_eq!(fingerprints.len(), 4, "every combination is its own revision");
}
/// The VPN pieces are Docker's own (`vpn_host_config`), not a second list.
#[test]
fn the_ec2_vpn_pieces_are_the_docker_ones() {
let vpn = crate::docker::container::vpn_requirements();
assert_eq!(vpn.capabilities, ["NET_ADMIN"]);
assert_eq!(
vpn.devices,
[("/dev/net/tun".to_string(), "/dev/net/tun".to_string(), "rwm".to_string())]
);
assert_eq!(vpn.sysctls, [("net.ipv4.conf.all.src_valid_mark".to_string(), "1".to_string())]);
}
#[test]
fn ec2_plan_checks_the_record_and_the_fit_and_carries_the_toggles() {
// Fargate: nothing, whatever the record says.
let fargate = project(StorageMode::Hybrid);
assert_eq!(ec2_plan(&fargate, "dev", None), Ok(None));
assert_eq!(ec2_plan(&fargate, "dev", Some(&compute("m7i.large"))), Ok(None));
// No record: the onboarding has no EC2 capacity (or it was removed).
let mut p = ec2_project(StorageMode::Hybrid, 1024, 4096);
let e = ec2_plan(&p, "dev", None).unwrap_err();
assert_eq!(
e,
"You have no EC2 capacity in dev; ask an admin to add it to your onboarding (or run the \
project on Fargate)."
);
// Fits: the toggles ride along.
p.allow_docker_access = true;
assert_eq!(
ec2_plan(&p, "dev", Some(&compute("m7i.large"))),
Ok(Some(Ec2Plan { docker_socket: true, vpn: false }))
);
p.vpn_support_enabled = true;
assert_eq!(
ec2_plan(&p, "dev", Some(&compute("m7i.large"))),
Ok(Some(Ec2Plan { docker_socket: true, vpn: true }))
);
// Too big: names what the instance has for a workspace (10 % memory
// margin: ECS registers less than the nominal 8 GiB).
let big = ec2_project(StorageMode::Hybrid, 2048, 8192);
let e = ec2_plan(&big, "dev", Some(&compute("m7i.large"))).unwrap_err();
assert!(e.contains("m7i.large"), "{e}");
assert!(e.contains("1920 CPU units, 7373 MiB"), "{e}");
assert!(ec2_plan(&big, "dev", Some(&compute("m7i.xlarge"))).is_ok());
// The gateway sidecar is not offered on EC2 (bridge networking).
let mut gw_on = ec2_project(StorageMode::Hybrid, 1024, 4096);
gw_on.aws.as_mut().unwrap().gateway_sidecar = true;
assert_eq!(ec2_plan(&gw_on, "dev", Some(&compute("m7i.large"))).unwrap_err(), EC2_NO_SIDECAR);
}
#[test]
fn the_sidecar_is_refused_on_ec2() {
use crate::models::gateway_settings::{GatewayModel, GatewaySettings};
let models = GatewaySettings {
models: vec![GatewayModel { name: "gpt".into(), model_id: "gpt-5".into() }],
..Default::default()
};
let mut p = ec2_project(StorageMode::Hybrid, 1024, 4096);
p.backend = crate::models::project::Backend::OpenAiCompatible;
p.aws.as_mut().unwrap().gateway_sidecar = true;
assert_eq!(gateway_sidecar_problem(&p, &models).as_deref(), Some(EC2_NO_SIDECAR));
p.aws.as_mut().unwrap().gateway_sidecar = false;
assert_eq!(gateway_sidecar_problem(&p, &models), None);
}
#[tokio::test]
async fn an_ec2_project_registers_an_ec2_revision_on_the_users_instance() {
let fake = Fake::onboarded();
let mut p = ec2_project(StorageMode::Hybrid, 1024, 4096);
p.allow_docker_access = true;
let c = compute("m7i.large");
prepare_workspace(&fake, &who(), &env_info(), &p, &[], &[], &[], IMAGE, None, Some(&c))
.await
.unwrap();
let reg = fake.registered.lock().unwrap();
assert_eq!(reg.len(), 1);
assert_eq!(reg[0].requires_compatibilities(), &[aws_sdk_ecs::types::Compatibility::Ec2]);
assert_eq!(reg[0].network_mode(), Some(&aws_sdk_ecs::types::NetworkMode::Bridge));
assert!(reg[0].volumes().iter().any(|v| v.name() == Some("docker-socket")));
}
#[tokio::test]
async fn an_ec2_project_without_capacity_or_room_fails_before_any_write() {
let p = ec2_project(StorageMode::Hybrid, 1024, 4096);
let fake = Fake::onboarded();
let e = prepare_workspace(&fake, &who(), &env_info(), &p, &[], &[], &[], IMAGE, None, None)
.await
.unwrap_err();
assert!(e.to_string().contains("no EC2 capacity in dev"), "{e}");
let big = ec2_project(StorageMode::Hybrid, 2048, 8192);
let c = compute("m7i.large");
let e = prepare_workspace(&fake, &who(), &env_info(), &big, &[], &[], &[], IMAGE, None, Some(&c))
.await
.unwrap_err();
assert!(e.to_string().contains("more than your EC2 instance type"), "{e}");
assert_eq!(fake.writes(), 0);
}
@@ -30,7 +30,7 @@ use std::sync::{Mutex, OnceLock};
use std::time::Duration;
use aws_config::SdkConfig;
use serde::Serialize;
use serde::{Deserialize, Serialize};
use serde_json::Value;
use tauri::{AppHandle, Emitter, State};
@@ -38,17 +38,20 @@ use aws_sdk_s3::error::ProvideErrorMetadata;
use tokio::sync::oneshot;
use crate::aws::environments::{
ensure_state_bucket, scan, state_bucket_policy, LIVE_LOCK_SECS, DetectedEnvironment,
EnvironmentAction, EnvironmentInfo, EnvironmentScan, EnvironmentState, Knowable, LocalRun,
OnboardingState, SdkDetectApi,
cluster_holder, ec2_name_holders, ensure_state_bucket, scan, state_bucket_policy,
user_record_from, DetectedEnvironment, EnvironmentAction, EnvironmentInfo, EnvironmentScan,
EnvironmentState, Knowable, LocalRun, NameHolder, OnboardApi, OnboardingState, SdkDetectApi,
SdkOnboardApi, UserCompute, UserRecord, LIVE_LOCK_SECS,
};
use crate::aws::tofu::{
self, backend_hcl, StartError, TofuEvent, TofuOp, TofuOutcome, Workdir,
};
use crate::aws::tofu_modules::{
boundary_vars, environment_vars, module_files, onboarding_vars, owner_key, state_bucket_name,
state_key, validate_environment_name, Deployment, ModuleKind, NetworkChoice,
state_key, user_ec2_stem, user_record_parameter_name, validate_ec2_user_tag,
validate_environment_name, Deployment, Ec2Compute, ModuleKind, NetworkChoice,
};
use crate::models::project::{ec2_instance_types, Ec2InstanceType, DEFAULT_EC2_INSTANCE_TYPE};
use crate::aws::{caller_identity, sdk_config, AwsError, CallerIdentity};
use crate::models::{validate_supported_region, AwsAuth, HostConnection, HostKind};
use crate::storage::secure;
@@ -272,7 +275,7 @@ pub(crate) fn onboarding_unlock_refusal(
}
let env = find(scan, name).ok_or_else(|| format!("No environment named \"{name}\" was found"))?;
match &env.onboarding {
Some(OnboardingState::Incomplete { lock: Some(l) })
Some(OnboardingState::Incomplete { lock: Some(l), .. })
if !l.local_process && env.onboarding_actions.contains(&EnvironmentAction::Unlock) =>
{
Ok(())
@@ -298,6 +301,139 @@ pub(crate) fn onboard_refusal(scan: &EnvironmentScan, name: &str) -> Result<(),
Ok(())
}
/// The EC2 capacity an onboarding asks for: none (Fargate only), or EC2
/// with these values. A command given no choice keeps what the user's
/// record holds ([`resolve_ec2`]).
#[derive(Deserialize, Serialize, Clone, Debug, PartialEq)]
#[serde(tag = "kind", rename_all = "snake_case")]
pub enum Ec2Choice {
None,
Ec2(Ec2Compute),
}
/// What an onboarding apply uses: the explicit choice, or with none given
/// the recorded capacity (a re-apply keeps it unless changed). With none
/// given and a record that exists but can't be used, it refuses rather than
/// apply without EC2, which would tear the capacity down.
pub(crate) fn resolve_ec2(
choice: Option<Ec2Choice>,
record: &UserRecord,
) -> Result<Option<Ec2Compute>, String> {
match (choice, record) {
(Some(Ec2Choice::None), _) => Ok(None),
(Some(Ec2Choice::Ec2(c)), _) => Ok(Some(c)),
(None, UserRecord::Absent) => Ok(None),
(None, UserRecord::Valid(r)) => Ok(Some(Ec2Compute {
instance_type: r.instance_type.clone(),
max_instances: r.max_instances,
user_tag: r.user_tag.clone(),
})),
(None, UserRecord::Unusable(reason)) => Err(format!(
"Your EC2 record can't be used ({reason}), so Triple-C can't keep your EC2 capacity \
as it is; choose the EC2 capacity explicitly"
)),
}
}
/// R3 (amended): an apply with EC2 capacity would adopt (or fail on) a
/// per-user cluster or Auto Scaling group named `stem` that already exists.
/// Each one found must be the caller's: its `triple-c:owner` tag (the
/// module's default tags) is `owner_id`, or, for an untagged one only, the
/// caller's record names the cluster. A half-built capacity of the caller's
/// own (a failed apply that wrote no record yet) is therefore re-applied;
/// one tagged as someone else's is always refused (a record can't vouch for
/// it), and an untagged one is refused unless the record names it.
pub(crate) fn ec2_collision_refusal(
stem: &str,
holders: &[NameHolder],
owner_id: &str,
recorded: Option<&UserCompute>,
) -> Result<(), String> {
let suffix = format!(":cluster/{stem}");
let recorded_here = recorded.is_some_and(|r| r.cluster_arn.ends_with(&suffix));
for h in holders {
match h.owner.as_deref() {
Some(o) if o == owner_id => continue,
None if recorded_here => continue,
_ => {}
}
let whose = if h.owner.is_some() {
"belongs to another onboarding"
} else {
"exists but isn't tagged as any Triple-C user's"
};
return Err(format!(
"An {} named {stem} already {whose}. The triple-c-user value is probably wrong: use \
the value an admin set as your triple-c-user sign-in tag",
h.kind
));
}
Ok(())
}
/// Everything a self-onboarding apply needs settled before it starts: the
/// caller's record, the capacity ([`resolve_ec2`]), the module variables
/// and, with EC2 capacity, the name check ([`ec2_collision_refusal`]).
/// Check-then-apply leaves a narrow race (another admin creating the same
/// names in between); only admins apply, and the module would then fail on
/// create rather than adopt.
pub(crate) async fn prepare_onboarding<A: OnboardApi>(
api: &A,
env: &str,
region: &str,
account: &str,
owner_id: &str,
choice: Option<Ec2Choice>,
) -> Result<Value, AwsError> {
let okey = owner_key(region, owner_id);
let read = api.parameter(&user_record_parameter_name(env, &okey)).await?;
let record = user_record_from(read, env, account, region);
let compute = resolve_ec2(choice, &record).map_err(other)?;
let vars = onboarding_vars(env, region, owner_id, compute.as_ref()).map_err(other)?;
if let Some(c) = &compute {
// The caller's own triple-c-user tag can't be read (session tags
// aren't visible to GetCallerIdentity); a wrong value fails later
// as access denied.
let stem = user_ec2_stem(env, &c.user_tag);
let holders = ec2_name_holders(api, &stem).await?;
ec2_collision_refusal(&stem, &holders, owner_id, record.valid()).map_err(other)?;
}
Ok(vars)
}
/// The bundle's version of the check, with what a workspace user can read
/// (the cluster, via `ecs:Describe*`, and the record; not the Auto Scaling
/// group): a warning to show, never a refusal. The admin runs the full
/// check from the README.
pub(crate) async fn bundle_ec2_warning<A: OnboardApi>(
api: &A,
env: &str,
region: &str,
account: &str,
owner_id: &str,
compute: &Ec2Compute,
) -> Option<String> {
let stem = user_ec2_stem(env, &compute.user_tag);
let checked = async {
let holder = cluster_holder(api, &stem).await?;
let read = api
.parameter(&user_record_parameter_name(env, &owner_key(region, owner_id)))
.await?;
let record = user_record_from(read, env, account, region);
Ok::<_, AwsError>(
ec2_collision_refusal(&stem, &holder.into_iter().collect::<Vec<_>>(), owner_id, record.valid())
.err(),
)
};
match checked.await {
Ok(warning) => warning,
Err(_) => Some(format!(
"Triple-C couldn't check whether the EC2 name {stem} is already in use; the README has \
the admin check it"
)),
}
}
/// `AlreadyRunning` is told apart before it becomes an `AwsError`.
pub(crate) fn start_error(e: StartError, label: &str) -> AwsError {
match e {
@@ -415,7 +551,54 @@ const README: &str = "README.txt";
const BACKEND: &str = "backend.hcl";
const TFVARS: &str = "terraform.tfvars.json";
fn bundle_readme(d: Deployment, account: &str, region: &str, bucket: &str, key: &str) -> String {
/// The onboarding bundle's EC2 step: before applying with EC2 capacity, the
/// admin checks that the per-user names are free or already this user's
/// (the module can't: a name taken by someone else's onboarding would be
/// adopted, and a collision only fails at create). Ownership is the
/// resources' `triple-c:owner` tag, or failing that the user's record, as
/// in [`ec2_collision_refusal`]. `user_tag` is the bundle's `ec2_user_tag`
/// when set and valid; otherwise the commands show `<tag>` for the admin to
/// fill in.
fn ec2_collision_step(
env: &str,
owner_key: &str,
owner_id: &str,
region: &str,
user_tag: Option<&str>,
) -> String {
let stem = user_ec2_stem(env, user_tag.unwrap_or("<tag>"));
let record = user_record_parameter_name(env, owner_key);
format!(
"EC2 capacity (only if ec2_instance_type is set)
------------------------------------------------
ec2_user_tag must be the value of the user's triple-c-user principal tag.
BEFORE tofu apply, check that the per-user EC2 names are not already in use
by another onboarding. Run:
aws ecs describe-clusters --region {region} --clusters {stem} --include TAGS --query 'clusters[?status!=`INACTIVE`].[clusterName,tags[?key==`triple-c:owner`].value|[0]]' --output text
aws autoscaling describe-auto-scaling-groups --region {region} --auto-scaling-group-names {stem} --query 'AutoScalingGroups[].[AutoScalingGroupName,Tags[?Key==`triple-c:owner`].Value|[0]]' --output text
aws ssm get-parameter --region {region} --name {record} --query Parameter.Value --output text
If either of the first two prints {stem}, apply ONLY if the owner printed
next to it is this user's owner_id ({owner_id}) or the third (this user's
record) succeeds and its cluster_arn ends in cluster/{stem}; otherwise the
name belongs to someone else's onboarding (or to nobody's): stop and
choose the correct ec2_user_tag. (The app runs the same check before
applying; run the check right before the apply, since nothing holds the
name in between.)
"
)
}
fn bundle_readme(
d: Deployment,
account: &str,
region: &str,
bucket: &str,
key: &str,
vars: &Value,
) -> String {
let what = match d {
Deployment::Environment(env) => format!(
"the Triple-C environment \"{env}\" (ECS cluster, security group, KMS key, log \
@@ -442,6 +625,16 @@ fn bundle_readme(d: Deployment, account: &str, region: &str, bucket: &str, key:
format!(r#"{{"LocationConstraint":"{region}",{tags}}}"#)
};
let policy = state_bucket_policy(bucket);
let ec2_step = match d {
Deployment::User { env, owner_key } => {
let tag = vars["ec2_user_tag"]
.as_str()
.filter(|t| validate_ec2_user_tag(t).is_ok());
let owner_id = vars["owner_id"].as_str().unwrap_or("<owner_id>");
ec2_collision_step(env, owner_key, owner_id, region, tag)
}
_ => String::new(),
};
let credentials = match d {
Deployment::Boundary => format!(
"- Credentials of an account administrator for account {account} WITHOUT
@@ -489,7 +682,7 @@ AWS CLI v2 recent enough that create-bucket accepts Tags, e.g. 2.39):
aws s3api put-public-access-block --bucket {bucket} --region {region} --expected-bucket-owner {account} --public-access-block-configuration BlockPublicAcls=true,IgnorePublicAcls=true,BlockPublicPolicy=true,RestrictPublicBuckets=true
aws s3api put-bucket-policy --bucket {bucket} --region {region} --expected-bucket-owner {account} --policy '{policy}'
2. Deploy
{ec2_step}2. Deploy
---------
From this folder:
@@ -529,7 +722,7 @@ fn bundle_files(
let json = serde_json::to_string_pretty(vars)
.map_err(|e| format!("Could not encode the module variables: {e}"))?;
files.push((TFVARS.into(), json + "\n"));
files.push((README.into(), bundle_readme(d, account, &region, &bucket, &key)));
files.push((README.into(), bundle_readme(d, account, &region, &bucket, &key, vars)));
Ok(files)
}
@@ -1317,12 +1510,15 @@ pub async fn cancel_aws_deployment(
}
/// Onboard the caller to `environment`, as the owner their own
/// `aws:userid` names.
/// `aws:userid` names, with the EC2 capacity `ec2` asks for (omitted: keep
/// what their user record holds, none without one). With EC2 capacity the
/// per-user names must be free or already theirs (R3).
#[tauri::command]
pub async fn onboard_aws_user(
connection_id: String,
region: String,
environment: String,
ec2: Option<Ec2Choice>,
app: AppHandle,
state: State<'_, AppState>,
) -> Result<String, AwsError> {
@@ -1331,7 +1527,15 @@ pub async fn onboard_aws_user(
let owner = validate_owner_id(&ctx.who.user_id).map_err(other)?;
let found = detect(&ctx).await?;
onboard_refusal(&found, &env).map_err(other)?;
let vars = onboarding_vars(&env, &ctx.region, &owner);
let vars = prepare_onboarding(
&SdkOnboardApi::new(&ctx.cfg),
&env,
&ctx.region,
&ctx.who.account,
&owner,
ec2,
)
.await?;
let key = DeploymentKey::User {
owner_key: owner_key(&ctx.region, &owner),
env,
@@ -1339,6 +1543,22 @@ pub async fn onboard_aws_user(
opened(launch(app, &ctx, &key, &vars, TofuOp::Apply).await)
}
/// The EC2 instance types onboarding offers (the app's built-in table) and
/// the one it offers first.
#[derive(Serialize, Clone, Debug, PartialEq)]
pub struct Ec2InstanceTable {
pub default_type: &'static str,
pub types: Vec<Ec2InstanceType>,
}
#[tauri::command]
pub fn list_ec2_instance_types() -> Ec2InstanceTable {
Ec2InstanceTable {
default_type: DEFAULT_EC2_INSTANCE_TYPE,
types: ec2_instance_types(),
}
}
/// Deploy (or re-apply) the account's permissions boundary. Refused when
/// the boundary exists but this region's bucket has no state for it.
#[tauri::command]
@@ -1378,8 +1598,10 @@ async fn pick_folder(window: &tauri::Window) -> Option<PathBuf> {
/// Write an admin bundle (D5) for `kind` into a folder the user picks:
/// `environment` names the environment (environment and onboarding),
/// `owner_id` the user to onboard (an `aws:userid`), `network` an
/// environment's network (public when omitted). `None` when the dialog was
/// dismissed; otherwise the folder written.
/// environment's network (public when omitted), `ec2` an onboarding's EC2
/// capacity (none when omitted; the bundle's README has the admin check
/// its names first). `None` when the dialog was dismissed; otherwise the
/// folder written and any warning.
#[allow(clippy::too_many_arguments)]
#[tauri::command]
pub async fn save_aws_admin_bundle(
@@ -1390,9 +1612,10 @@ pub async fn save_aws_admin_bundle(
owner_id: Option<String>,
network: Option<NetworkChoice>,
max_workspace_hours: Option<u32>,
ec2: Option<Ec2Compute>,
window: tauri::Window,
state: State<'_, AppState>,
) -> Result<Option<String>, AwsError> {
) -> Result<Option<SavedBundle>, AwsError> {
let need_env = || {
environment
.as_deref()
@@ -1420,7 +1643,7 @@ pub async fn save_aws_admin_bundle(
.map(str::trim)
.ok_or_else(|| other("The user's owner id (aws:userid) is required"))
.and_then(|o| validate_owner_id(o).map_err(other))?;
let vars = onboarding_vars(&env, &region_ok, &owner);
let vars = onboarding_vars(&env, &region_ok, &owner, ec2.as_ref()).map_err(other)?;
(
DeploymentKey::User {
owner_key: owner_key(&region_ok, &owner),
@@ -1432,11 +1655,29 @@ pub async fn save_aws_admin_bundle(
ModuleKind::PermissionsBoundary => (DeploymentKey::Boundary, boundary_vars(&region_ok)),
};
let ctx = context(&state, &connection_id, &region).await?;
let warning = match (&key, &ec2, &owner_id) {
(DeploymentKey::User { env, .. }, Some(c), Some(owner)) => {
let api = SdkOnboardApi::new(&ctx.cfg);
bundle_ec2_warning(&api, env, &ctx.region, &ctx.who.account, owner.trim(), c).await
}
_ => None,
};
let Some(dir) = pick_folder(&window).await else {
return Ok(None);
};
write_bundle(&dir, key.deployment(), &ctx.who.account, &ctx.region, &vars).map_err(other)?;
Ok(Some(dir.to_string_lossy().into_owned()))
Ok(Some(SavedBundle {
folder: dir.to_string_lossy().into_owned(),
warning,
}))
}
/// A bundle written: its folder and, for an onboarding with EC2 capacity,
/// a warning when its EC2 name looks taken ([`bundle_ec2_warning`]).
#[derive(Serialize, Clone, Debug, PartialEq)]
pub struct SavedBundle {
pub folder: String,
pub warning: Option<String>,
}
#[cfg(test)]
@@ -6,7 +6,7 @@ use serde_json::json;
use super::*;
use crate::aws::tofu::Workdir;
use crate::aws::environments::{
EnvironmentState, Knowable, LockInfo, OnboardingState,
EnvironmentState, Knowable, LockInfo, NamedResource, OnboardingState, Param, UserCompute,
};
use crate::aws::tofu_modules::{
environment_vars, module_files, onboarding_vars, owner_key, state_key, ModuleKind,
@@ -195,7 +195,7 @@ fn bundles_contain_the_module_backend_vars_and_readme() {
let dir = tempfile::tempdir().unwrap();
let ok = owner_key(REGION, USER_ID);
let d = Deployment::User { env: "dev", owner_key: &ok };
let vars = onboarding_vars("dev", REGION, USER_ID);
let vars = onboarding_vars("dev", REGION, USER_ID, None).unwrap();
let written = write_bundle(dir.path(), d, ACCOUNT, REGION, &vars).unwrap();
for (name, text) in module_files(ModuleKind::UserOnboard) {
@@ -253,6 +253,45 @@ fn bundles_contain_the_module_backend_vars_and_readme() {
}
assert!(!readme.contains("put-bucket-tagging"), "a later tag can't satisfy the create");
// The EC2 name-collision check (the app does it too, Task 4): the
// exact per-user names, and this user's record to compare against.
let record = format!("/triple-c/users/dev/{ok}");
for needle in [
"BEFORE tofu apply",
&format!("aws ecs describe-clusters --region {REGION} --clusters triple-c-dev--u-<tag>"),
&format!(
"aws autoscaling describe-auto-scaling-groups --region {REGION} \
--auto-scaling-group-names triple-c-dev--u-<tag>"
),
&format!("aws ssm get-parameter --region {REGION} --name {record}"),
"cluster/triple-c-dev--u-<tag>",
"--include TAGS",
"triple-c:owner",
&format!("owner_id ({USER_ID})"),
] {
assert!(readme.contains(needle), "README lacks {needle:?}:\n{readme}");
}
// With a tag in the variables, the commands name it.
let mut tagged = vars.clone();
tagged["ec2_instance_type"] = "m7i.large".into();
tagged["ec2_user_tag"] = "alice".into();
let other = tempfile::tempdir().unwrap();
write_bundle(other.path(), d, ACCOUNT, REGION, &tagged).unwrap();
let readme_tagged = fs::read_to_string(other.path().join("README.txt")).unwrap();
assert!(readme_tagged.contains("--clusters triple-c-dev--u-alice "), "{readme_tagged}");
assert!(readme_tagged.contains("--auto-scaling-group-names triple-c-dev--u-alice "));
assert!(!readme_tagged.contains("<tag>"));
// The check comes before the apply it guards.
let check = readme.find("BEFORE tofu apply").unwrap();
let apply = readme.find(" tofu apply").unwrap();
assert!(check < apply, "{readme}");
// Not in the other bundles.
let env_dir = tempfile::tempdir().unwrap();
let env_vars = environment_vars("dev", REGION, &NetworkChoice::Public, 0).unwrap();
write_bundle(env_dir.path(), Deployment::Environment("dev"), ACCOUNT, REGION, &env_vars).unwrap();
let env_readme = fs::read_to_string(env_dir.path().join("README.txt")).unwrap();
assert!(!env_readme.contains("EC2 capacity"), "{env_readme}");
// Writing again over our own files is fine...
write_bundle(dir.path(), d, ACCOUNT, REGION, &vars).unwrap();
// ...but an unrelated file (or folder) means the wrong folder.
@@ -507,12 +546,12 @@ fn destroy_is_refused_while_users_are_onboarded() {
fn only_your_own_stale_onboarding_lock_can_be_removed() {
use EnvironmentAction::*;
let mut stale = env("dev", EnvironmentState::Ready, vec![Join]);
stale.onboarding = Some(OnboardingState::Incomplete { lock: Some(lock(9000, false)) });
stale.onboarding = Some(OnboardingState::Incomplete { lock: Some(lock(9000, false)), ec2: None });
stale.onboarding_actions = vec![Unlock];
let mut live = env("live", EnvironmentState::Ready, vec![Join]);
live.onboarding = Some(OnboardingState::InProgress { lock: Some(lock(10, false)) });
let mut none = env("none", EnvironmentState::Ready, vec![Join]);
none.onboarding = Some(OnboardingState::Incomplete { lock: None });
none.onboarding = Some(OnboardingState::Incomplete { lock: None, ec2: None });
let s = scan(Knowable::Yes, vec![stale, live, none]);
let mine = owner_key(REGION, USER_ID);
assert!(onboarding_unlock_refusal(&s, "dev", &mine).is_ok());
@@ -631,3 +670,319 @@ fn ending_a_launch_never_clears_a_following_slot() {
assert_eq!(slots.begin_launch(dir).unwrap_err(), SlotBusy::Following);
slots.end_launch(Path::new("/w/env/none")); // nothing there: no-op
}
fn recorded(env: &str, tag: &str, max: u32) -> UserCompute {
let stem = crate::aws::tofu_modules::user_ec2_stem(env, tag);
UserCompute {
instance_type: "m7g.xlarge".into(),
max_instances: max,
user_tag: tag.into(),
cluster_arn: format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/{stem}"),
capacity_provider: stem,
}
}
#[test]
fn a_reapply_keeps_the_recorded_compute_unless_changed() {
let rec = UserRecord::Valid(recorded("dev", "alice", 3));
let kept = Ec2Compute {
instance_type: "m7g.xlarge".into(),
max_instances: 3,
user_tag: "alice".into(),
};
// Omitted: keep what the record says (or none without one).
assert_eq!(resolve_ec2(None, &rec), Ok(Some(kept.clone())));
assert_eq!(resolve_ec2(None, &UserRecord::Absent), Ok(None));
// Explicitly none: the re-apply removes it.
assert_eq!(resolve_ec2(Some(Ec2Choice::None), &rec), Ok(None));
// A new choice wins.
let new = Ec2Compute {
instance_type: "m7i.large".into(),
max_instances: 1,
user_tag: "alice".into(),
};
assert_eq!(
resolve_ec2(Some(Ec2Choice::Ec2(new.clone())), &rec),
Ok(Some(new.clone()))
);
// A record that exists but can't be used is never read as "none" (that
// would tear the capacity down); an explicit choice still goes.
let bad = UserRecord::Unusable("schema is not 1".into());
let e = resolve_ec2(None, &bad).unwrap_err();
assert!(e.contains("schema is not 1") && e.contains("explicitly"), "{e}");
assert_eq!(resolve_ec2(Some(Ec2Choice::None), &bad), Ok(None));
assert_eq!(resolve_ec2(Some(Ec2Choice::Ec2(new.clone())), &bad), Ok(Some(new.clone())));
// The kept values reach the module's variables unchanged.
let vars = onboarding_vars("dev", REGION, USER_ID, resolve_ec2(None, &rec).unwrap().as_ref()).unwrap();
assert_eq!(vars["ec2_instance_type"], "m7g.xlarge");
assert_eq!(vars["ec2_max_instances"], 3);
assert_eq!(vars["ec2_user_tag"], "alice");
// The UI's JSON for each choice.
let none: Ec2Choice = serde_json::from_value(json!({"kind": "none"})).unwrap();
assert_eq!(none, Ec2Choice::None);
let ec2: Ec2Choice = serde_json::from_value(json!({
"kind": "ec2", "instance_type": "m7i.large", "max_instances": 1, "user_tag": "alice"
}))
.unwrap();
assert_eq!(ec2, Ec2Choice::Ec2(new));
}
fn held(kind: &'static str, owner: Option<&str>) -> NameHolder {
NameHolder {
kind,
owner: owner.map(String::from),
}
}
#[test]
fn ec2_names_held_by_another_owner_are_refused() {
let stem = "triple-c-dev--u-alice";
let me = USER_ID;
// Free names: fine, with or without a record.
assert!(ec2_collision_refusal(stem, &[], me, None).is_ok());
assert!(ec2_collision_refusal(stem, &[], me, Some(&recorded("dev", "bob", 1))).is_ok());
let cluster = |o| held("ECS cluster", o);
let group = |o| held("Auto Scaling group", o);
// Tagged as mine: a half-built capacity (no record yet) is re-applied.
assert!(ec2_collision_refusal(stem, &[cluster(Some(me)), group(Some(me))], me, None).is_ok());
// Another owner's, or untagged: refused, naming the resource.
let e = ec2_collision_refusal(stem, &[cluster(Some("AROAOTHER:bob"))], me, None).unwrap_err();
assert!(e.contains("ECS cluster") && e.contains(stem) && e.contains("another onboarding"), "{e}");
assert!(e.contains("triple-c-user"), "{e}");
let e = ec2_collision_refusal(stem, &[cluster(Some(me)), group(None)], me, None).unwrap_err();
assert!(e.contains("Auto Scaling group") && e.contains("isn't tagged"), "{e}");
// The record naming the cluster is the fallback for an untagged one
// only: one tagged as someone else's is refused even then.
let mine = recorded("dev", "alice", 2);
assert!(ec2_collision_refusal(stem, &[cluster(None), group(None)], me, Some(&mine)).is_ok());
assert!(ec2_collision_refusal(stem, &[cluster(None), group(Some(me))], me, Some(&mine)).is_ok());
let e = ec2_collision_refusal(stem, &[cluster(None), group(Some("x"))], me, Some(&mine)).unwrap_err();
assert!(e.contains("Auto Scaling group") && e.contains("another onboarding"), "{e}");
// A record naming another cluster doesn't help.
assert!(ec2_collision_refusal(stem, &[cluster(None)], me, Some(&recorded("dev", "bob", 1))).is_err());
// A cluster that merely ends in the same text is not the same cluster.
let mut lookalike = recorded("dev", "alice", 1);
lookalike.cluster_arn = format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/x{stem}");
assert!(ec2_collision_refusal(stem, &[cluster(None)], me, Some(&lookalike)).is_err());
}
/// An `OnboardApi` answering from fixed data, recording what it was asked.
#[derive(Default)]
struct FakeOnboard {
record: Option<String>,
record_denied: bool,
cluster: Option<NamedResource>,
group: Option<NamedResource>,
group_denied: bool,
calls: std::sync::Mutex<Vec<String>>,
}
impl FakeOnboard {
fn calls(&self) -> Vec<String> {
self.calls.lock().unwrap().clone()
}
}
impl OnboardApi for FakeOnboard {
async fn parameter(&self, name: &str) -> Result<Param, AwsError> {
self.calls.lock().unwrap().push(format!("parameter {name}"));
Ok(match (&self.record, self.record_denied) {
(_, true) => Param::Denied,
(Some(v), _) => Param::Value(v.clone()),
(None, _) => Param::Missing,
})
}
async fn cluster(&self, name: &str) -> Result<Option<NamedResource>, AwsError> {
self.calls.lock().unwrap().push(format!("cluster {name}"));
Ok(self.cluster.clone())
}
async fn group(&self, name: &str) -> Result<Option<NamedResource>, AwsError> {
self.calls.lock().unwrap().push(format!("group {name}"));
if self.group_denied {
return Err(AwsError::AccessDenied {
action: "autoscaling:DescribeAutoScalingGroups".into(),
message: "denied".into(),
});
}
Ok(self.group.clone())
}
}
fn named(status: Option<&str>, owner: Option<&str>) -> NamedResource {
NamedResource {
status: status.map(String::from),
tags: owner
.map(|o| vec![("triple-c:owner".to_string(), o.to_string())])
.unwrap_or_default(),
}
}
fn record_json(env: &str, tag: &str) -> String {
let r = recorded(env, tag, 2);
json!({
"schema": 1, "instance_type": r.instance_type, "cluster_arn": r.cluster_arn,
"capacity_provider": r.capacity_provider, "user_tag": r.user_tag, "max_instances": 2,
})
.to_string()
}
fn prepare(api: &FakeOnboard, choice: Option<Ec2Choice>) -> Result<Value, AwsError> {
tokio::runtime::Builder::new_current_thread()
.build()
.unwrap()
.block_on(prepare_onboarding(api, "dev", REGION, ACCOUNT, USER_ID, choice))
}
fn alice() -> Option<Ec2Choice> {
Some(Ec2Choice::Ec2(Ec2Compute {
instance_type: "m7i.large".into(),
max_instances: 1,
user_tag: "alice".into(),
}))
}
#[test]
fn onboarding_checks_the_ec2_names_before_it_applies() {
let stem = "triple-c-dev--u-alice";
// Fargate only: no name lookups at all.
let api = FakeOnboard::default();
let vars = prepare(&api, Some(Ec2Choice::None)).unwrap();
assert_eq!(vars["ec2_instance_type"], "");
assert!(api.calls().iter().all(|c| c.starts_with("parameter ")), "{:?}", api.calls());
// Free names: the vars come back.
let api = FakeOnboard::default();
let vars = prepare(&api, alice()).unwrap();
assert_eq!(vars["ec2_user_tag"], "alice");
assert!(api.calls().contains(&format!("cluster {stem}")));
assert!(api.calls().contains(&format!("group {stem}")));
// Another owner's cluster: refused, so nothing is launched.
let api = FakeOnboard {
cluster: Some(named(Some("ACTIVE"), Some("AROAOTHER:bob"))),
..Default::default()
};
let e = prepare(&api, alice()).unwrap_err();
assert!(format!("{e:?}").contains("another onboarding"), "{e:?}");
// My own half-built cluster and group (a failed apply, no record yet).
let api = FakeOnboard {
cluster: Some(named(Some("PROVISIONING"), Some(USER_ID))),
group: Some(named(None, Some(USER_ID))),
..Default::default()
};
assert!(prepare(&api, alice()).is_ok());
// Untagged: refused unless my record names it.
let untagged = || FakeOnboard {
cluster: Some(named(Some("ACTIVE"), None)),
..Default::default()
};
assert!(prepare(&untagged(), alice()).is_err());
let api = FakeOnboard {
record: Some(record_json("dev", "alice")),
..untagged()
};
assert!(prepare(&api, alice()).is_ok());
// A deleted (INACTIVE) cluster doesn't hold the name; every other
// status does.
let api = FakeOnboard {
cluster: Some(named(Some("INACTIVE"), None)),
..Default::default()
};
assert!(prepare(&api, alice()).is_ok());
for status in ["ACTIVE", "PROVISIONING", "DEPROVISIONING", "FAILED"] {
let api = FakeOnboard {
cluster: Some(named(Some(status), Some("AROAOTHER:bob"))),
..Default::default()
};
assert!(prepare(&api, alice()).is_err(), "{status}");
}
// An Auto Scaling group alone is enough to refuse.
let api = FakeOnboard {
group: Some(named(None, Some("AROAOTHER:bob"))),
..Default::default()
};
assert!(prepare(&api, alice()).is_err());
// A denied group lookup is an error, never "free".
let api = FakeOnboard {
group_denied: true,
..Default::default()
};
assert!(prepare(&api, alice()).is_err());
}
#[test]
fn capacity_kept_from_the_record_is_checked_too() {
let stem = "triple-c-dev--u-alice";
// Omitted choice + a record: the record's capacity, and its names are
// checked (here another owner's group appeared under it).
let api = FakeOnboard {
record: Some(record_json("dev", "alice")),
cluster: Some(named(Some("ACTIVE"), Some(USER_ID))),
group: Some(named(None, Some(USER_ID))),
..Default::default()
};
let vars = prepare(&api, None).unwrap();
assert_eq!(vars["ec2_instance_type"], "m7g.xlarge");
assert_eq!(vars["ec2_max_instances"], 2);
assert!(api.calls().contains(&format!("group {stem}")));
// A record that can't be used (unreadable or invalid) and no choice:
// refused before any name lookup or launch.
for api in [
FakeOnboard {
record_denied: true,
..Default::default()
},
FakeOnboard {
record: Some("{\"schema\": 9}".into()),
..Default::default()
},
] {
let e = prepare(&api, None).unwrap_err();
assert!(format!("{e:?}").contains("explicitly"), "{e:?}");
assert!(api.calls().iter().all(|c| c.starts_with("parameter ")));
}
}
#[test]
fn a_bundle_warns_when_its_cluster_name_looks_taken() {
let run = |api: &FakeOnboard| {
let c = Ec2Compute {
instance_type: "m7i.large".into(),
max_instances: 1,
user_tag: "alice".into(),
};
tokio::runtime::Builder::new_current_thread()
.build()
.unwrap()
.block_on(bundle_ec2_warning(api, "dev", REGION, ACCOUNT, USER_ID, &c))
};
assert_eq!(run(&FakeOnboard::default()), None);
let api = FakeOnboard {
cluster: Some(named(Some("ACTIVE"), Some("AROAOTHER:bob"))),
..Default::default()
};
assert!(run(&api).unwrap().contains("another onboarding"));
// Only the cluster half: a workspace user can't read Auto Scaling.
assert!(!api.calls().iter().any(|c| c.starts_with("group ")));
let api = FakeOnboard {
cluster: Some(named(Some("ACTIVE"), Some(USER_ID))),
..Default::default()
};
assert_eq!(run(&api), None);
}
#[test]
fn the_ui_gets_the_apps_instance_table() {
let table = list_ec2_instance_types();
assert_eq!(table.types, crate::models::project::ec2_instance_types());
assert_eq!(table.types.len(), 30);
assert_eq!(table.default_type, "m7i.large");
assert!(table.types.iter().any(|t| t.instance_type == table.default_type));
}
+94 -18
View File
@@ -11,6 +11,7 @@ use crate::aws::permissions::{check_permissions, PermissionReport, SdkPermission
use crate::aws::policies::{policy_files, Tier};
use crate::aws::pricing::{self, HourlyEstimate, RatesCache, SdkPricingApi};
use crate::aws::{caller_identity, sdk_config, sdk_config_for_keys, AwsError, CallerIdentity};
use crate::models::project::{ec2_instance_spec, LaunchType};
use crate::models::{
validate_connection_name, validate_profile_name, validate_supported_region, AppSettings, AwsAuth,
AwsWorkspaceConfig, HostConnection, HostKind,
@@ -322,18 +323,33 @@ fn pricing_cache() -> Option<&'static RatesCache> {
/// error:** a bad region, a missing connection, a denied or slow call (8 s
/// bound) or an unreadable price list is `None`, shown as "Estimate
/// unavailable". Rates are cached a day per region.
///
/// An EC2 project (E10) is priced as `instance_type` (the user's, from their
/// onboarding record, which the UI already holds): its on-demand hourly
/// price, the same cache and back-off rules. No instance type, or one not in
/// the app's table, is `None`.
#[tauri::command]
pub async fn aws_cost_estimate(
connection_id: String,
region: String,
config: AwsWorkspaceConfig,
instance_type: Option<String>,
state: State<'_, AppState>,
) -> Result<Option<HourlyEstimate>, String> {
let Some(cache) = pricing_cache() else {
return Ok(None);
};
let settings = state.settings_store.get();
Ok(cost_estimate(cache, &connection_id, &region, &config, &settings, SystemTime::now()).await)
Ok(cost_estimate(
cache,
&connection_id,
&region,
&config,
instance_type.as_deref(),
&settings,
SystemTime::now(),
)
.await)
}
async fn cost_estimate(
@@ -341,24 +357,41 @@ async fn cost_estimate(
connection_id: &str,
region: &str,
config: &AwsWorkspaceConfig,
instance_type: Option<&str>,
settings: &AppSettings,
now: SystemTime,
) -> Option<HourlyEstimate> {
let region = validate_supported_region(region).ok()?;
// Only now touch the connection (a keys connection reads the keychain),
// and inside the fetch's 8 s bound: reading a profile's credentials can
// hang as well (an SSO or process provider).
let api = || async {
let conn = aws_conn(connection_id, settings).ok()?;
let cfg = sdk_config(&conn, pricing::PRICING_REGION).await.ok()?;
Some(SdkPricingApi::new(&cfg))
};
if config.launch_type == LaunchType::Ec2 {
let instance_type = instance_type.filter(|t| ec2_instance_spec(t).is_some())?;
let instance = async {
match cache.read_ec2(&region, instance_type, now) {
Some((price, true)) => Some(price),
_ => {
cache
.ec2_price_with::<SdkPricingApi, _>(api(), &region, instance_type, now)
.await
}
}
};
// The root volume's price is best effort: without it the estimate
// is the instance's alone (and says so).
let gp3 = cache.gp3_price_with::<SdkPricingApi, _>(api(), &region, now);
let (price, gp3) = tokio::join!(instance, gp3);
return Some(pricing::estimate_ec2(config, price?, gp3));
}
let rates = match cache.read(&region, now) {
Some((rates, true)) => Some(rates),
_ => {
// Only now touch the connection (a keys connection reads the
// keychain); without one, a stale cache still answers.
// Built inside the fetch's 8 s bound: reading a profile's
// credentials can hang as well (an SSO or process provider).
let api = async {
let conn = aws_conn(connection_id, settings).ok()?;
let cfg = sdk_config(&conn, pricing::PRICING_REGION).await.ok()?;
Some(SdkPricingApi::new(&cfg))
};
cache.rates_with::<SdkPricingApi, _>(api, &region, now).await
}
// Without a connection, a stale cache still answers.
_ => cache.rates_with::<SdkPricingApi, _>(api(), &region, now).await,
}?;
Some(pricing::estimate(config, &rates.for_arch(config.arch)?))
}
@@ -604,7 +637,15 @@ mod tests {
fn the_policy_document_is_every_file_of_the_tier() {
let admin = aws_policy_document(Tier::EnvironmentAdmin);
let names: Vec<&str> = admin.iter().map(|f| f.name.as_str()).collect();
assert_eq!(names, ["environment-admin.json", "environment-admin-network.json"]);
assert_eq!(
names,
[
"environment-admin.json",
"environment-admin-network.json",
"environment-admin-compute.json",
"environment-admin-scaling.json"
]
);
for f in &admin {
serde_json::from_str::<serde_json::Value>(&f.json).expect("policy JSON");
}
@@ -646,7 +687,7 @@ mod tests {
spot: true,
..Default::default()
};
let e = cost_estimate(&cache, "gone", "us-west-2", &config, &settings, now)
let e = cost_estimate(&cache, "gone", "us-west-2", &config, None, &settings, now)
.await
.expect("estimate");
assert!((e.usd_per_hour - (0.04 + 2.0 * 0.004)).abs() < 1e-9);
@@ -656,7 +697,7 @@ mod tests {
arch: crate::models::Arch::Arm64,
..config.clone()
};
assert_eq!(cost_estimate(&cache, "gone", "us-west-2", &arm, &settings, now).await, None);
assert_eq!(cost_estimate(&cache, "gone", "us-west-2", &arm, None, &settings, now).await, None);
}
#[tokio::test]
@@ -667,11 +708,46 @@ mod tests {
let config = AwsWorkspaceConfig::default();
let now = SystemTime::now();
// No cache and no such connection.
assert_eq!(cost_estimate(&cache, "gone", "us-west-2", &config, &settings, now).await, None);
assert_eq!(cost_estimate(&cache, "gone", "us-west-2", &config, None, &settings, now).await, None);
// A region that isn't one (it would name the cache file).
cache.write("us-west-2", &sample_rates(), now).unwrap();
for bad in ["../us-west-2", "", "us-west-2/x"] {
assert_eq!(cost_estimate(&cache, "gone", bad, &config, &settings, now).await, None);
assert_eq!(cost_estimate(&cache, "gone", bad, &config, None, &settings, now).await, None);
}
}
/// E10: an EC2 project is priced as the user's instance type, from the
/// same cache; without a known type there is no estimate.
#[tokio::test]
async fn an_ec2_projects_estimate_is_its_instance_types_price() {
let dir = tempfile::tempdir().unwrap();
let cache = RatesCache::new(dir.path().to_path_buf());
let now = SystemTime::now();
cache.write("us-west-2", &sample_rates(), now).unwrap();
cache.write_ec2("us-west-2", "m7i.large", 0.1008, now).unwrap();
let settings = AppSettings::default();
let config = AwsWorkspaceConfig {
launch_type: LaunchType::Ec2,
cpu: 1024,
memory_mib: 2048,
storage: crate::models::project::StorageMode::Ephemeral,
..Default::default()
};
let e = cost_estimate(&cache, "gone", "us-west-2", &config, Some("m7i.large"), &settings, now)
.await
.expect("estimate");
assert!((e.usd_per_hour - 0.1008).abs() < 1e-9);
assert!(e.ec2_instance && !e.note_spot && !e.efs_billed_separately);
// No instance type, one not in the table, or one with no cached price
// (and no connection to ask): no estimate — never Fargate's.
for t in [None, Some("x9.huge"), Some("m7g.large")] {
assert_eq!(cost_estimate(&cache, "gone", "us-west-2", &config, t, &settings, now).await, None, "{t:?}");
}
// A Fargate project ignores an instance type.
let fargate = AwsWorkspaceConfig { launch_type: LaunchType::Fargate, ..config.clone() };
let f = cost_estimate(&cache, "gone", "us-west-2", &fargate, Some("m7i.large"), &settings, now)
.await
.expect("fargate");
assert!(!f.ec2_instance);
}
}
@@ -1225,6 +1225,10 @@ fn validate_gateway_sidecar_update(
let Some(problem) = crate::aws::workspace::gateway_sidecar_problem(project, gateway) else {
return Ok(());
};
// Not on EC2 at all (bridge networking): refused whatever was stored.
if problem == crate::aws::workspace::EC2_NO_SIDECAR {
return Err(problem);
}
let was_on = stored.aws.as_ref().is_some_and(|a| a.gateway_sidecar);
if was_on && project.backend == crate::models::project::Backend::OpenAiCompatible {
return Ok(());
@@ -1248,6 +1252,15 @@ fn validate_aws_update(
crate::models::validate_region(&place.region)?;
crate::aws::tofu_modules::validate_environment_name(&place.environment)?;
crate::models::validate_aws_workspace(cfg)?;
// Docker access and VPN need the user's own instance (E6, E7).
if cfg.launch_type != crate::models::LaunchType::Ec2 {
if project.allow_docker_access {
return Err("Docker access on AWS needs the EC2 launch type; turn it off first".into());
}
if project.vpn_support_enabled {
return Err("VPN support on AWS needs the EC2 launch type; turn it off first".into());
}
}
// Q6: forwarded over loopback, TCP only, never a reserved port
// (rows the save adds or edits; everything when newly on AWS).
crate::host::port_forwards::validate_aws_port_mappings_update(
@@ -1589,6 +1602,20 @@ fn emit_connect_warning(app_handle: &tauri::AppHandle, project_id: &str, name: &
let _ = app_handle.emit("workspace-warning", connect_warning_payload(project_id, name, detail));
}
/// A start that an app restart interrupted (EC2 cold start): topic "start".
fn emit_interrupted_warning(app_handle: &tauri::AppHandle, project_id: &str, name: &str, detail: &str) {
let _ = app_handle.emit("workspace-warning", interrupted_warning_payload(project_id, name, detail));
}
fn interrupted_warning_payload(project_id: &str, name: &str, detail: &str) -> serde_json::Value {
serde_json::json!({
"project_id": project_id,
"message": format!("“{name}”: a start was interrupted"),
"detail": detail,
"topic": "start",
})
}
fn connect_warning_payload(project_id: &str, name: &str, detail: &str) -> serde_json::Value {
serde_json::json!({
"project_id": project_id,
@@ -1646,6 +1673,29 @@ pub async fn aws_environment_max_hours(
Ok(crate::host::aws_host::environment_max_hours(&project, &settings).await)
}
/// The caller's EC2 capacity in an AWS project's environment, for Host
/// settings' launch-type control and reservation limits (phase 7). Never an
/// error: a local project or a failed read is `unknown`.
#[tauri::command]
pub async fn aws_ec2_capacity(
project_id: String,
state: State<'_, AppState>,
) -> Result<crate::host::aws_host::Ec2Capacity, String> {
let Some(project) = state.projects_store.get(&project_id).filter(|p| p.is_aws()) else {
return Ok(crate::host::aws_host::Ec2Capacity::Unknown);
};
let settings = state.settings_store.get();
Ok(crate::host::aws_host::ec2_capacity(&project, &settings).await)
}
/// Why `config`'s CPU/memory reservations don't fit `instance_type`, or
/// null (`models::project::ec2_fit_problem`: the start's own rule, so Host
/// settings never disagree with it).
#[tauri::command]
pub fn ec2_fit_problem(config: AwsWorkspaceConfig, instance_type: String) -> Option<String> {
crate::models::project::ec2_fit_problem(&config, &instance_type)
}
#[tauri::command]
pub async fn stop_project_container(
project_id: String,
@@ -2173,6 +2223,10 @@ pub async fn reconcile_aws_projects(
let on_ports_warning = move |detail: String| {
emit_ports_warning(&ports_handle, &ports_id, &ports_name, &detail)
};
let (int_handle, int_id, int_name) = (handle.clone(), id.clone(), name.clone());
let on_interrupted_warning = move |detail: String| {
emit_interrupted_warning(&int_handle, &int_id, &int_name, &detail)
};
let on_connect_warning =
move |detail: String| emit_connect_warning(&handle, &id, &name, &detail);
if tokio::time::timeout(
@@ -2183,6 +2237,7 @@ pub async fn reconcile_aws_projects(
store,
on_connect_warning,
on_ports_warning,
on_interrupted_warning,
),
)
.await
@@ -2213,6 +2268,16 @@ mod tests {
use super::*;
#[test]
fn the_interrupted_warning_has_its_own_topic_and_headline() {
let v = interrupted_warning_payload("p1", "demo", "retry");
assert_eq!(v["topic"], "start");
assert_eq!(v["detail"], "retry");
let m = v["message"].as_str().unwrap();
assert!(m.contains("demo") && m.contains("a start was interrupted"));
assert!(!m.contains("credentials"));
}
// ── Pending-cleanup aging ────────────────────────────────────────────
#[test]
@@ -2894,6 +2959,43 @@ mod tests {
assert!(validate_gateway_sidecar_update(&on, &on, &half).is_ok());
// Off: nothing to check.
assert!(validate_gateway_sidecar_update(&off, &off, &half).is_ok());
// EC2 (bridge networking): refused even when it was already on.
let mut ec2 = on.clone();
ec2.aws.as_mut().unwrap().launch_type = crate::models::LaunchType::Ec2;
let e = validate_gateway_sidecar_update(&on, &ec2, &with_models).unwrap_err();
assert_eq!(e, crate::aws::workspace::EC2_NO_SIDECAR);
assert!(validate_gateway_sidecar_update(&ec2, &ec2, &with_models).is_err());
ec2.aws.as_mut().unwrap().gateway_sidecar = false;
assert!(validate_gateway_sidecar_update(&on, &ec2, &with_models).is_ok());
}
#[test]
fn docker_and_vpn_on_aws_need_the_ec2_launch_type() {
use crate::models::LaunchType;
let hosts = vec![host("h1")];
let stored = aws_project();
for (docker, vpn) in [(true, false), (false, true)] {
let mut p = stored.clone();
p.allow_docker_access = docker;
p.vpn_support_enabled = vpn;
assert!(validate_aws_update(&stored, &p, &hosts).is_err());
// Already stored that way on Fargate: still refused (no grandfathering).
assert!(validate_aws_update(&p, &p, &hosts).is_err());
// Ec2 -> Fargate with the flags still on: refused.
let mut ec2 = p.clone();
ec2.aws.as_mut().unwrap().launch_type = LaunchType::Ec2;
assert!(validate_aws_update(&ec2, &ec2, &hosts).is_ok());
assert!(validate_aws_update(&ec2, &p, &hosts).is_err());
// Local -> AWS(Fargate) with the flags on: refused.
let mut local = stored.clone();
local.host_id = None;
local.aws_placement = None;
local.aws = None;
local.allow_docker_access = docker;
local.vpn_support_enabled = vpn;
assert!(validate_aws_update(&local, &p, &hosts).is_err());
}
}
#[test]
@@ -2666,3 +2666,91 @@ fn two_connects_at_once_never_mix_their_files() {
assert_eq!(c.run(&["../../etc"]).status.code(), Some(2));
assert_eq!(c.run(&[&a]).status.code(), Some(1));
}
// ------------------------------------------- Docker access and VPN (phase 7, EC2)
/// The `docker_socket` block with the socket at `sock` and the account tools
/// faked (they log to stdout): `getent` knows no group, so one is added.
fn docker_socket_step(sock: &Path, aws: bool) -> Output {
let script = format!(
"getent() {{ return 2; }}\n\
groupadd() {{ echo \"groupadd $*\"; }}\n\
usermod() {{ echo \"usermod $*\"; }}\n\
triplec_aws={}\n{}",
u8::from(aws),
eblock("docker_socket").replace("/var/run/docker.sock", &sock.display().to_string())
);
bash(&script, &[], &[], &[])
}
fn unix_socket(path: &Path) -> std::os::unix::net::UnixListener {
std::os::unix::net::UnixListener::bind(path).expect("bind socket")
}
/// E6: an EC2 project's Docker access is its instance's socket, so the group
/// step runs in aws mode too (it was Docker-only), and as before without one.
#[test]
fn the_docker_socket_group_step_runs_in_aws_mode_when_the_socket_exists() {
let tmp = tempfile::tempdir().unwrap();
let sock = tmp.path().join("docker.sock");
let _listener = unix_socket(&sock);
let gid = fs::metadata(&sock).map(|m| std::os::unix::fs::MetadataExt::gid(&m)).unwrap();
for aws in [true, false] {
let out = docker_socket_step(&sock, aws);
let t = text(&out);
assert!(out.status.success(), "{t}");
assert!(t.contains(&format!("groupadd -g {gid} docker-host")), "aws={aws}: {t}");
assert!(t.contains("usermod -aG"), "aws={aws}: {t}");
assert!(t.trim_end().ends_with(" claude"), "aws={aws}: {t}");
}
// No socket (Fargate, or Docker access off): nothing happens.
let out = docker_socket_step(&tmp.path().join("absent.sock"), true);
assert!(out.status.success());
assert_eq!(text(&out), "");
// A regular file there is not a socket either.
let file = tmp.path().join("file.sock");
fs::write(&file, "").unwrap();
assert_eq!(text(&docker_socket_step(&file, true)), "");
assert!(!eblock("docker_socket").contains("triplec_aws"), "no aws gate");
}
/// The feature-skills block with the skill source and destination in `tmp`
/// and `chown` switched off.
fn feature_skills(tmp: &Path, vpn: &str) -> Output {
let src = tmp.join("opt-skills");
let dest = tmp.join("home-skills");
let script = format!(
"chown() {{ :; }}\ntriplec_aws=1\n{}\n[ -z \"${{VPN_SUPPORT_ENABLED+x}}\" ] && echo unset-ok",
eblock("feature_skills")
.replace("/opt/triple-c-skills", &src.display().to_string())
.replace("/home/claude/.claude/skills", &dest.display().to_string())
);
bash(&script, &[], &[("VPN_SUPPORT_ENABLED", vpn)], &[])
}
/// E7: in aws mode the VPN skill follows `VPN_SUPPORT_ENABLED` as on Docker
/// (installed when 1, removed when 0), and the variable is gone after.
#[test]
fn the_vpn_skill_follows_vpn_support_in_aws_mode() {
let tmp = tempfile::tempdir().unwrap();
let src = tmp.path().join("opt-skills/pia-vpn");
fs::create_dir_all(&src).unwrap();
fs::write(src.join("SKILL.md"), "vpn").unwrap();
let dest = tmp.path().join("home-skills/pia-vpn/SKILL.md");
let out = feature_skills(tmp.path(), "1");
let t = text(&out);
assert!(t.contains("pia-vpn skill installed"), "{t}");
assert!(t.contains("unset-ok"), "{t}");
assert_eq!(fs::read_to_string(&dest).unwrap(), "vpn");
let t = text(&feature_skills(tmp.path(), "0"));
assert!(t.contains("pia-vpn skill removed"), "{t}");
assert!(!dest.exists());
// The call itself is not behind any mode check.
let call = ENTRYPOINT
.lines()
.find(|l| l.starts_with("install_feature_skill pia-vpn"))
.expect("the call");
assert!(!call.contains("triplec_aws"));
}
+31
View File
@@ -1112,6 +1112,37 @@ fn vpn_host_config(enabled: bool) -> VpnHostConfigParts {
)
}
/// What a VPN client needs, as plain data, for hosts that aren't Docker (an
/// AWS EC2 task definition): the capabilities, the devices as
/// `(host path, container path, cgroup permissions)`, and the sysctls,
/// sorted. Derived from [`vpn_host_config`] so the two can't drift.
pub(crate) struct VpnRequirements {
pub capabilities: Vec<String>,
pub devices: Vec<(String, String, String)>,
pub sysctls: Vec<(String, String)>,
}
pub(crate) fn vpn_requirements() -> VpnRequirements {
let (caps, devices, sysctls) = vpn_host_config(true);
let mut sysctls: Vec<(String, String)> = sysctls.unwrap_or_default().into_iter().collect();
sysctls.sort();
VpnRequirements {
capabilities: caps.unwrap_or_default(),
devices: devices
.unwrap_or_default()
.into_iter()
.map(|d| {
(
d.path_on_host.unwrap_or_default(),
d.path_in_container.unwrap_or_default(),
d.cgroup_permissions.unwrap_or_default(),
)
})
.collect(),
sysctls,
}
}
/// Turn the daemon's device-passthrough failure into an explanation.
///
/// **This fires on `start`, not `create`.** Verified against Docker 29.7:
+563 -62
View File
@@ -1,6 +1,9 @@
//! `ProjectHost::Aws`: a project's container is an ECS Fargate task in one
//! of the account's environments, reached through an SSM port-forward to
//! the in-task agent (`ssm.rs`, `agent.rs`).
//! `ProjectHost::Aws`: a project's container is an ECS task in one of the
//! account's environments — Fargate on the environment's cluster, or (phase
//! 7, `LaunchType::Ec2`) on the user's own EC2 capacity, their per-user
//! cluster from their record — reached through an SSM port-forward to the
//! in-task agent (`ssm.rs`, `agent.rs`). A task's cluster is always read
//! back from its ARN, so stop, reconcile and the watch need no record.
//!
//! Start: connection → `SdkConfig` → caller identity → environment record →
//! onboarding → the project's live task, if there is one (its token is read
@@ -27,27 +30,30 @@ use super::tunnels::{self, LiveTunnel, TunnelHandle};
use super::watch::{AwsTaskLookup, TaskGroup, WatchedTask};
use super::ContainerTarget;
use crate::aws::environments::EnvironmentInfo;
use crate::aws::environments::UserRecord;
use crate::aws::runtime::{
check_onboarded, cluster_arn_of_task, cluster_of_task, ensure_agent_token, read_agent_token,
read_environment,
check_onboarded, cluster_arn_of_task, cluster_of_task, ec2_cluster_of_task, project_task_clusters,
ensure_agent_token, read_agent_token, read_environment, read_user_record,
remove_workspace_resources, wait_access_point, PollTiming, RemoveRequest, RuntimeApi,
SdkRuntimeApi,
};
use crate::aws::tasks::{
find_or_run, find_project_task, stopped_among, task_log_stream, wait_ready, SdkTaskApi,
StopKind, TaskApi, TaskInfo, TaskLogs, WaitTiming, LOG_TAIL_LINES, STOP_REASON,
find_or_run, find_project_task, stopped_among, task_log_stream, wait_ready,
waiting_for_instance, WAIT_TIMED_OUT, RunOn, SdkTaskApi, StopKind, TaskApi, TaskInfo, TaskLogs, WaitTiming,
LOG_TAIL_LINES, STOP_REASON,
};
use crate::aws::tofu_modules::{
environment_cluster_name, log_stream_prefix, owner_key, task_family,
};
use crate::aws::workspace::{
gateway_sidecar_problem, new_agent_token, prepare_workspace, GatewayPayload, SdkWorkspaceApi,
WorkspaceApi,
gateway_sidecar_problem, new_agent_token, no_ec2_capacity, prepare_workspace, GatewayPayload,
SdkWorkspaceApi, WorkspaceApi,
};
use crate::aws::{caller_identity, sdk_config, CallerIdentity};
use crate::docker::ca_certs::{container_cert_name, ResolvedCa};
use crate::models::app_settings::ImageSource;
use crate::models::container_config::REGISTRY_IMAGE;
use crate::models::project::LaunchType;
use crate::models::{AppSettings, Project, ProjectStatus};
use crate::storage::projects_store::ProjectsStore;
use crate::storage::settings_store::SettingsStore;
@@ -58,6 +64,8 @@ pub const PROGRESS_READING: &str = "Reading environment…";
pub const PROGRESS_PREPARING: &str = "Preparing workspace…";
pub const PROGRESS_STARTING: &str = "Starting task…";
pub const PROGRESS_WAITING: &str = "Waiting for the task (about 2 minutes)…";
/// An EC2 task waiting for an instance with room (capacity scaled to zero).
pub const PROGRESS_INSTANCE: &str = "Starting an instance (about 3 minutes)…";
pub const PROGRESS_CONNECTING: &str = "Connecting…";
/// A concurrent start won the RunTask race: the secret now holds a token
@@ -137,16 +145,26 @@ fn is_registry_reference(name: &str) -> bool {
/// [`aws_env`]'s answer: the pairs, and the resolved CA whose files ship too.
pub type AwsEnv = (Vec<(String, String)>, Option<ResolvedCa>);
/// Whether Claude Code's sandbox (bubblewrap) may run where `project`'s task
/// runs — the one place this is decided. Not on AWS: Fargate can't create
/// namespaces (P9), and neither can an ECS task on the user's EC2 instance
/// (E8, measured live in phase 7: `bwrap --unshare-user` fails with "No
/// permissions to create new namespace" — Docker's default seccomp profile
/// refuses it without CAP_SYS_ADMIN, though the AMI allows user namespaces).
pub fn sandbox_allowed(_project: &Project) -> bool {
false
}
/// The environment pairs an AWS task gets: [`workspace_env`] with the same
/// inputs Docker resolves, and Claude Code's sandbox forced off (R4: Fargate
/// support for bubblewrap is unverified). Returns the resolved CA too.
/// inputs Docker resolves, and Claude Code's sandbox forced off where
/// [`sandbox_allowed`] says no. Returns the resolved CA too.
///
/// [`workspace_env`]: crate::docker::workspace_env::workspace_env
pub fn aws_env(project: &Project, settings: &AppSettings) -> Result<AwsEnv, String> {
let mut p = project.clone();
if p.sandbox_mode_enabled {
if p.sandbox_mode_enabled && !sandbox_allowed(project) {
log::info!(
"Project {}: Claude Code's sandbox is turned off on AWS (not yet supported on Fargate)",
"Project {}: Claude Code's sandbox is turned off on AWS (bubblewrap can't run in an ECS task)",
project.id
);
p.sandbox_mode_enabled = false;
@@ -410,6 +428,9 @@ pub struct StartInputs<'a> {
/// attaching to a running one never needs the provider key or models.
pub gateway: &'a (dyn Fn() -> Result<Option<SidecarSecrets>, String> + Send + Sync),
pub wait: WaitTiming,
/// The wait for a task on the user's EC2 capacity (an instance may have
/// to launch first: [`WaitTiming::ec2`]).
pub ec2_wait: WaitTiming,
pub access_point_wait: PollTiming,
/// Called as soon as the project has a task (found, or RunTask
/// returned) so the store holds its ARN before the long wait — a crash
@@ -468,6 +489,236 @@ async fn stop_created<T: TaskApi>(api: &T, cluster: &str, arn: &str) -> bool {
}
}
/// Wait for `arn` on `cluster`, telling the user when an EC2 task is waiting
/// for an instance ([`PROGRESS_INSTANCE`], then [`PROGRESS_WAITING`] again
/// once one has room). An EC2 task that never got an instance fails with
/// what to check rather than ECS's last status.
#[allow(clippy::too_many_arguments)]
async fn wait_task<T: TaskApi>(
tasks: &T,
cluster: &str,
arn: &str,
logs: &TaskLogs,
timing: WaitTiming,
ec2: Option<u32>,
progress: &(dyn Fn(&str) + Send + Sync),
) -> Result<TaskInfo, String> {
let waiting = std::sync::atomic::AtomicBool::new(false);
let r = wait_ready(tasks, cluster, arn, logs, timing, |t| {
if ec2.is_none() {
return;
}
let now = waiting_for_instance(t);
if now != waiting.swap(now, std::sync::atomic::Ordering::Relaxed) {
progress(if now { PROGRESS_INSTANCE } else { PROGRESS_WAITING });
}
})
.await;
match r {
Ok(t) => Ok(t),
Err(e)
if waiting.load(std::sync::atomic::Ordering::Relaxed)
&& e.to_string().starts_with(WAIT_TIMED_OUT) =>
{
let max = ec2.unwrap_or(1);
Err(format!(
"No EC2 instance had room for the task within {} minutes. Your capacity may \
already be running its maximum of {max} instance{} (stop another EC2 \
workspace), or its Auto Scaling group could not launch one (an admin can see \
why in the group's activity in the EC2 console).",
timing.timeout.as_secs() / 60,
if max == 1 { "" } else { "s" }
))
}
Err(e) => Err(e.to_string()),
}
}
/// What a start says when ECS ran its task with ECS Exec off even with an
/// instance up (it should not happen; the task can't be reached).
pub const EXEC_OFF: &str = "ECS started the task with ECS Exec turned off, so Triple-C can't \
reach it. Start the project again.";
/// The most a start waits for a task it stopped to reach STOPPED before
/// running another (its own budget, apart from the placement wait).
const STOP_WAIT: Duration = Duration::from_secs(3 * 60);
/// A start found the project's task with ECS Exec off and still waiting for
/// an instance, recently started: another start is bringing one up.
pub const INSTANCE_RACE: &str = "Another start of this workspace is bringing up an EC2 \
instance; try again in a minute (once it is up, Start connects to it), or Stop the \
workspace to cancel it";
/// The stop reason of a task a start replaced (one with ECS Exec off).
pub const REPLACED_REASON: &str = "Replaced by another Triple-C start";
/// How long past the EC2 placement wait an exec-off task still waiting for
/// an instance counts as another start's, before it is taken as orphaned.
/// The task's age is ECS's `createdAt` against this machine's clock, so the
/// margin also absorbs clock skew: a local clock more than this far ahead
/// could take another start's waiting task for an orphan and replace it (the
/// other start then says so: "Another start … replaced this one"); one far
/// behind only delays the replacement. NTP-synced clocks are well inside it.
const ORPHAN_MARGIN: Duration = Duration::from_secs(10 * 60);
/// Seconds since the Unix epoch on this machine's clock (0 if it is before).
fn now_unix() -> i64 {
std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.map_or(0, |d| d.as_secs() as i64)
}
/// Whether an exec-off task still waiting for an instance is older than
/// any start could still be waiting for it (`ec2_wait` + a margin): a
/// crashed start's orphan. No creation time reads as fresh.
fn is_orphaned(t: &TaskInfo, ec2_wait: WaitTiming, now_unix: i64) -> bool {
let limit = (ec2_wait.timeout + ORPHAN_MARGIN).as_secs() as i64;
t.created_unix.is_some_and(|c| now_unix - c > limit)
}
/// Why [`rerun_once_placed`] did not give a new task.
enum RerunError {
/// A concurrent start's live task was found: record it, report
/// [`START_RACE`].
Race(TaskInfo),
Failed(String),
}
impl From<String> for RerunError {
fn from(e: String) -> Self {
RerunError::Failed(e)
}
}
/// Wait (at most `min(timeout, STOP_WAIT)`) until `arn` is STOPPED or gone.
async fn wait_stopped<T: TaskApi>(tasks: &T, cluster: &str, arn: &str, timing: WaitTiming) -> Result<(), String> {
let deadline = tokio::time::Instant::now() + timing.timeout.min(STOP_WAIT);
let wanted = [arn.to_string()];
loop {
let gone = tasks
.describe(cluster, &wanted)
.await
.map_err(|e| e.to_string())?
.into_iter()
.find(|t| t.arn == arn)
.is_none_or(|t| t.last_status == "STOPPED");
if gone {
return Ok(());
}
if tokio::time::Instant::now() + timing.interval > deadline {
return Err(format!(
"The task {arn} did not stop in time; start the project again in a minute."
));
}
tokio::time::sleep(timing.interval).await;
}
}
/// A cold EC2 start's second RunTask. ECS turns `enableExecuteCommand` off
/// on a RunTask that has to wait for an instance (PROVISIONING; verified
/// live, phase 7) and the task keeps it off once placed, so it could never
/// be reached. That first task still makes the capacity provider launch an
/// instance: once an instance has taken it (and it is still meant to run),
/// it is stopped and the task run again — through [`find_or_run`], so a
/// concurrent start's task is never doubled ([`START_RACE`]) — now straight
/// onto that instance with ECS Exec. A first task that stops by itself (a
/// user's Stop, a failed placement) ends the start with its reason; nothing
/// is run again. Returns the new task.
#[allow(clippy::too_many_arguments)]
async fn rerun_once_placed<T: TaskApi>(
tasks: &T,
env: &EnvironmentInfo,
owner: &str,
project_id: &str,
compute: &crate::aws::environments::UserCompute,
first: &str,
task_def_arn: &str,
timing: WaitTiming,
progress: &(dyn Fn(&str) + Send + Sync),
) -> Result<TaskInfo, RerunError> {
let cluster = compute.cluster_arn.as_str();
let deadline = tokio::time::Instant::now() + timing.timeout;
let wanted = [first.to_string()];
progress(PROGRESS_INSTANCE);
// Until an instance has taken the first task.
loop {
let Some(t) = tasks
.describe(cluster, &wanted)
.await
.map_err(|e| e.to_string())?
.into_iter()
.find(|t| t.arn == first)
else {
return Err(format!("The task {first} is no longer known to ECS; start the project again").into());
};
if t.desired_status == "STOPPED" || t.last_status == "STOPPED" {
if t.stopped_reason.as_deref() == Some(REPLACED_REASON) {
return Err(RerunError::Failed(
"Another start of this workspace replaced this one; Start again in a minute \
to connect to it."
.into(),
));
}
return Err(format!(
"The task stopped before an EC2 instance took it ({}); start the project again.",
t.stopped_reason.as_deref().unwrap_or("no reason given")
)
.into());
}
if t.container_instance_arn.is_some() && t.desired_status == "RUNNING" {
break;
}
if tokio::time::Instant::now() + timing.interval > deadline {
let max = compute.max_instances;
return Err(format!(
"No EC2 instance had room for the task within {} minutes. Your capacity may \
already be running its maximum of {max} instance{} (stop another EC2 \
workspace), or its Auto Scaling group could not launch one (an admin can see \
why in the group's activity in the EC2 console).",
timing.timeout.as_secs() / 60,
if max == 1 { "" } else { "s" }
)
.into());
}
tokio::time::sleep(timing.interval).await;
}
// Free its reservation before running again, or the second task would
// wait for an instance too.
tasks
.stop_task(cluster, first, STOP_REASON)
.await
.map_err(|e| e.to_string())?;
wait_stopped(tasks, cluster, first, timing).await?;
progress(PROGRESS_WAITING);
let start = find_or_run(tasks, env, owner, project_id, task_def_arn, RunOn::Ec2(compute))
.await
.map_err(|e| e.to_string())?;
if start.reused {
return Err(RerunError::Race(start.task));
}
if !start.task.exec_enabled {
let _ = stop_created(tasks, cluster, &start.task.arn).await;
return Err(RerunError::Failed(EXEC_OFF.into()));
}
Ok(start.task)
}
/// Where a project's task may be ([`project_task_clusters`]): the
/// environment's cluster, the record's EC2 cluster, the stored task's.
fn task_clusters(env: &EnvironmentInfo, record: Option<&UserRecord>, stored: Option<&str>) -> Vec<String> {
project_task_clusters(
&env.name,
env.cluster_arn.clone(),
record.and_then(UserRecord::valid).map(|c| c.cluster_arn.clone()),
stored,
)
}
/// The cluster (ARN) a task runs on, from its ARN.
fn task_cluster(env: &EnvironmentInfo, task_arn: &str) -> String {
cluster_arn_of_task(task_arn).unwrap_or_else(|| env.cluster_arn.clone())
}
/// The start sequence; see the module docs.
pub async fn start_core<T: TaskApi, W: WorkspaceApi, R: RuntimeApi, C: Connector>(
tasks: &T,
@@ -496,9 +747,56 @@ pub async fn start_core<T: TaskApi, W: WorkspaceApi, R: RuntimeApi, C: Connector
stream_prefix: log_stream_prefix(&ok, &project.id),
};
let existing = find_project_task(tasks, &env.cluster_arn, &family, &inp.who.user_id, &project.id)
.await
.map_err(s)?;
// EC2 (phase 7): the user's record says where new tasks run. Read it
// only for EC2 projects; a Fargate start asks exactly what it did. A
// failed read still lets a live task (found in the environment's
// cluster or the stored task's) be attached; only a new run needs it.
let (record, record_err) = if cfg.launch_type == LaunchType::Ec2 {
match read_user_record(rt, &env.name, &ok, &inp.who.account, region).await {
Ok(r) => (Some(r), None),
Err(e) => {
log::warn!("Project {}: could not read the EC2 record ({e})", project.id);
(None, Some(format!("Could not read your EC2 capacity record: {e}")))
}
}
} else {
(None, None)
};
// The EC2 wait (and its progress) for a task on a per-user cluster.
let ec2_of = |task_arn: &str| {
ec2_cluster_of_task(&env.name, task_arn).map(|_| {
record.as_ref().and_then(UserRecord::valid).map_or(1, |c| c.max_instances)
})
};
let timing_of = |ec2: Option<u32>| if ec2.is_some() { inp.ec2_wait } else { inp.wait };
let stored = cfg.task_arn.as_deref();
let mut existing = None;
for cluster in task_clusters(&env, record.as_ref(), stored) {
existing = find_project_task(tasks, &cluster, &family, &inp.who.user_id, &project.id)
.await
.map_err(s)?;
if existing.is_some() {
break;
}
}
// A task with ECS Exec off (a cold EC2 start that stopped halfway) can
// never be attached to: stop it and start afresh.
// One still waiting for an instance belongs to a start in progress.
// An old one no start can still be waiting for is an orphan (a crashed
// start): it is replaced like a placed one.
if let Some(t) = existing.take_if(|t| !t.exec_enabled) {
if waiting_for_instance(&t) && !is_orphaned(&t, inp.ec2_wait, now_unix()) {
return Err(INSTANCE_RACE.into());
}
log::warn!("Project {}: its task {} has ECS Exec off; stopping it", project.id, t.arn);
let cluster = task_cluster(&env, &t.arn);
tasks
.stop_task(&cluster, &t.arn, REPLACED_REASON)
.await
.map_err(|e| format!("Could not stop the workspace's unreachable task {}: {e}", t.arn))?;
wait_stopped(tasks, &cluster, &t.arn, timing_of(Some(1))).await?;
}
let (task, token, access_point_id, workspace_access_point_id, created) = if let Some(task) = existing {
// Attach: never rotate a running task's token (P2).
@@ -511,12 +809,32 @@ pub async fn start_core<T: TaskApi, W: WorkspaceApi, R: RuntimeApi, C: Connector
}));
let token = read_agent_token(rt, &env.name, &ok, &project.id).await.map_err(s)?;
progress(PROGRESS_WAITING);
let task = match wait_ready(tasks, &env.cluster_arn, &task.arn, &logs, inp.wait, |_| {}).await {
let ec2 = ec2_of(&task.arn);
let cluster = task_cluster(&env, &task.arn);
let task = match wait_task(tasks, &cluster, &task.arn, &logs, timing_of(ec2), ec2, progress).await {
Ok(t) => t,
Err(e) => return Err(with_log_tail(tasks, &logs, &task.arn, e.to_string()).await),
Err(e) => return Err(with_log_tail(tasks, &logs, &task.arn, e).await),
};
(task, token, None, None, false)
} else {
// Where the new task runs; an EC2 project without usable capacity
// (or whose record could not be read) fails here, before anything
// is written.
if let Some(e) = record_err {
return Err(e);
}
let run_on = match &record {
None => RunOn::Fargate { spot: cfg.spot },
Some(UserRecord::Valid(c)) => RunOn::Ec2(c),
Some(UserRecord::Absent) => return Err(no_ec2_capacity(&env.name)),
Some(UserRecord::Unusable(why)) => {
return Err(format!(
"Your EC2 capacity record in {} can't be used ({why}); ask an admin to \
re-apply your onboarding.",
env.name
))
}
};
progress(PROGRESS_PREPARING);
let sidecar = (inp.gateway)()?;
let prepared = prepare_workspace(
@@ -529,6 +847,7 @@ pub async fn start_core<T: TaskApi, W: WorkspaceApi, R: RuntimeApi, C: Connector
inp.custom_env_keys,
inp.image,
sidecar.as_ref().map(SidecarSecrets::payload),
record.as_ref().and_then(UserRecord::valid),
)
.await
.map_err(s)?;
@@ -545,7 +864,7 @@ pub async fn start_core<T: TaskApi, W: WorkspaceApi, R: RuntimeApi, C: Connector
&inp.who.user_id,
&project.id,
&prepared.task_definition_arn,
cfg.spot,
run_on,
)
.await
.map_err(s)?;
@@ -560,19 +879,47 @@ pub async fn start_core<T: TaskApi, W: WorkspaceApi, R: RuntimeApi, C: Connector
}));
return Err(START_RACE.into());
}
let arn = start.task.arn.clone();
(inp.on_task)(Some(&TaskRecord {
task_arn: arn.clone(),
access_point_id: prepared.access_point_id.clone(),
task_definition_arn: Some(prepared.task_definition_arn.clone()),
workspace_access_point_id: prepared.workspace_access_point_id.clone(),
}));
let mut arn = start.task.arn.clone();
let record_task = |arn: &str| {
(inp.on_task)(Some(&TaskRecord {
task_arn: arn.to_string(),
access_point_id: prepared.access_point_id.clone(),
task_definition_arn: Some(prepared.task_definition_arn.clone()),
workspace_access_point_id: prepared.workspace_access_point_id.clone(),
}))
};
record_task(&arn);
progress(PROGRESS_WAITING);
let task = match wait_ready(tasks, &env.cluster_arn, &arn, &logs, inp.wait, |_| {}).await {
let cluster = run_on.cluster(&env).to_string();
if let (RunOn::Ec2(compute), false) = (run_on, start.task.exec_enabled) {
match rerun_once_placed(tasks, &env, &inp.who.user_id, &project.id, compute, &arn, &prepared.task_definition_arn, inp.ec2_wait, progress).await {
Ok(t) => {
arn = t.arn;
record_task(&arn);
}
Err(RerunError::Race(winner)) => {
// As on the first launch: the winner's task is the project's.
record_task(&winner.arn);
return Err(START_RACE.into());
}
Err(RerunError::Failed(e)) => {
let msg = with_log_tail(tasks, &logs, &arn, e).await;
if stop_created(tasks, &cluster, &arn).await {
(inp.on_task)(None);
}
return Err(msg);
}
}
}
let ec2 = match run_on {
RunOn::Ec2(c) => Some(c.max_instances),
RunOn::Fargate { .. } => None,
};
let task = match wait_task(tasks, &cluster, &arn, &logs, timing_of(ec2), ec2, progress).await {
Ok(t) => t,
Err(e) => {
let msg = with_log_tail(tasks, &logs, &arn, e.to_string()).await;
if stop_created(tasks, &env.cluster_arn, &arn).await {
let msg = with_log_tail(tasks, &logs, &arn, e).await;
if stop_created(tasks, &cluster, &arn).await {
(inp.on_task)(None);
}
return Err(msg);
@@ -617,7 +964,7 @@ pub async fn start_core<T: TaskApi, W: WorkspaceApi, R: RuntimeApi, C: Connector
})
}
Err(e) => {
if created && stop_created(tasks, &env.cluster_arn, &task.arn).await {
if created && stop_created(tasks, &task_cluster(&env, &task.arn), &task.arn).await {
(inp.on_task)(None);
}
Err(e)
@@ -654,6 +1001,60 @@ pub async fn environment_max_hours(project: &Project, settings: &AppSettings) ->
Some(info.max_workspace_hours)
}
/// What Host settings know of the user's EC2 capacity in an AWS project's
/// environment (phase 7): the launch-type control offers EC2 only with
/// `Ec2`, and sizes the reservations against `available_*`
/// ([`ec2_available`]; `None` for a type the app's table doesn't know).
#[derive(Debug, Clone, PartialEq, serde::Serialize)]
#[serde(tag = "state", rename_all = "snake_case")]
pub enum Ec2Capacity {
Ec2 {
compute: crate::aws::environments::UserCompute,
available_cpu: Option<u32>,
available_memory_mib: Option<u32>,
},
/// No user record: no EC2 capacity in this environment.
None,
/// A record that can't be used (the reason names a field, never a value).
Unusable { reason: String },
/// It couldn't be read (offline, no connection, slow, a local project).
Unknown,
}
/// [`Ec2Capacity`] from a read of the user record (`None`: no read).
pub fn ec2_capacity_from(read: Option<UserRecord>) -> Ec2Capacity {
match read {
Some(UserRecord::Valid(compute)) => {
let available = crate::models::project::ec2_available(&compute.instance_type);
Ec2Capacity::Ec2 {
available_cpu: available.map(|a| a.0),
available_memory_mib: available.map(|a| a.1),
compute,
}
}
Some(UserRecord::Absent) => Ec2Capacity::None,
Some(UserRecord::Unusable(reason)) => Ec2Capacity::Unusable { reason },
None => Ec2Capacity::Unknown,
}
}
/// The caller's EC2 capacity in `project`'s environment, from their user
/// record. Never an error: anything that stops the read (10 s bound) is
/// [`Ec2Capacity::Unknown`].
pub async fn ec2_capacity(project: &Project, settings: &AppSettings) -> Ec2Capacity {
let read = async {
let placement = project.aws_placement.as_ref()?;
let (_, region, sdk) = project_sdk(project, settings).await.ok()?;
let who = caller_identity(&sdk).await.ok()?;
let ok = owner_key(&region, &who.user_id);
read_user_record(&SdkRuntimeApi::new(&sdk), &placement.environment, &ok, &who.account, &region)
.await
.map_err(|e| log::info!("Project {}: could not read the EC2 record ({e})", project.id))
.ok()
};
ec2_capacity_from(tokio::time::timeout(std::time::Duration::from_secs(10), read).await.ok().flatten())
}
/// `ProjectHost::Aws::ensure_started`.
pub async fn ensure_started(
project: &Project,
@@ -730,6 +1131,7 @@ pub async fn ensure_started_with(
image: &image,
gateway: &sidecar,
wait: WaitTiming::default(),
ec2_wait: WaitTiming::ec2(),
access_point_wait: PollTiming::default(),
on_task: &on_task,
},
@@ -756,7 +1158,7 @@ pub async fn ensure_started_with(
LiveTunnel {
task: ecs,
task_arn: started.task.arn.clone(),
cluster_arn: started.env.cluster_arn.clone(),
cluster_arn: task_cluster(&started.env, &started.task.arn),
connection_id,
region,
sdk,
@@ -903,22 +1305,28 @@ pub async fn stop_endpoint(ep: &AgentEndpoint) -> Option<Result<(), String>> {
/// Where to look for a project's live task when no ARN is stored (a crash
/// before the store write, or a store from an older build).
pub struct FindTask {
pub cluster: String,
/// The environment's cluster, then the user's EC2 cluster when known.
pub clusters: Vec<String>,
pub family: String,
pub owner_user_id: String,
pub project_id: String,
}
impl FindTask {
fn new(project: &Project, region: &str, owner_user_id: &str) -> Self {
/// `ec2_cluster`: the user's EC2 cluster from their record, if they
/// have one ([`user_ec2_cluster`]); the stored task's per-user cluster
/// is added either way.
fn new(project: &Project, region: &str, owner_user_id: &str, ec2_cluster: Option<String>) -> Self {
let env = project
.aws_placement
.as_ref()
.map(|p| p.environment.clone())
.unwrap_or_default();
let ok = owner_key(region, owner_user_id);
let stored = project.aws.as_ref().and_then(|a| a.task_arn.as_deref());
let clusters = project_task_clusters(&env, environment_cluster_name(&env), ec2_cluster, stored);
FindTask {
cluster: environment_cluster_name(&env),
clusters,
family: task_family(&env, &ok, &project.id),
owner_user_id: owner_user_id.to_string(),
project_id: project.id.clone(),
@@ -926,6 +1334,43 @@ impl FindTask {
}
}
/// The project's live task in any of `find`'s clusters.
async fn find_in_clusters<T: TaskApi>(api: &T, find: &FindTask) -> Result<Option<TaskInfo>, String> {
for cluster in &find.clusters {
if let Some(t) =
find_project_task(api, cluster, &find.family, &find.owner_user_id, &find.project_id)
.await
.map_err(s)?
{
return Ok(Some(t));
}
}
Ok(None)
}
/// The user's EC2 cluster (ARN) for an EC2 project, from their record; `None`
/// for a Fargate project, or when there is no usable record (best effort:
/// only used to look for a task no ARN was stored for).
pub async fn user_ec2_cluster<R: RuntimeApi>(
rt: &R,
project: &Project,
who: &CallerIdentity,
region: &str,
) -> Option<String> {
if !project.aws.as_ref().is_some_and(|a| a.launch_type == LaunchType::Ec2) {
return None;
}
let env = &project.aws_placement.as_ref()?.environment;
match read_user_record(rt, env, &owner_key(region, &who.user_id), &who.account, region).await {
Ok(UserRecord::Valid(c)) => Some(c.cluster_arn),
Ok(_) => None,
Err(e) => {
log::warn!("Project {}: could not read the EC2 record ({e})", project.id);
None
}
}
}
/// The project's task ARN: the stored one, else its live task's.
async fn task_arn_or_find<T: TaskApi>(
api: &T,
@@ -935,12 +1380,7 @@ async fn task_arn_or_find<T: TaskApi>(
if let Some(a) = stored {
return Ok(Some(a.to_string()));
}
Ok(
find_project_task(api, &find.cluster, &find.family, &find.owner_user_id, &find.project_id)
.await
.map_err(s)?
.map(|t| t.arn),
)
Ok(find_in_clusters(api, find).await?.map(|t| t.arn))
}
/// Stop an AWS project's task: through its tunnel when there is one, else
@@ -960,7 +1400,8 @@ pub async fn stop_project(project: &Project, settings: &AppSettings) -> Result<(
Some(a) => Some(a),
None => {
let who = caller_identity(&sdk).await.map_err(s)?;
task_arn_or_find(&api, None, &FindTask::new(project, &region, &who.user_id)).await?
let ec2 = user_ec2_cluster(&SdkRuntimeApi::new(&sdk), project, &who, &region).await;
task_arn_or_find(&api, None, &FindTask::new(project, &region, &who.user_id, ec2)).await?
}
};
if let Some(arn) = arn {
@@ -987,10 +1428,12 @@ pub async fn remove_resources(
let ok = owner_key(&region, &who.user_id);
let env = project.aws_placement.as_ref().map(|p| p.environment.clone()).unwrap_or_default();
let aws = project.aws.as_ref();
let rt = SdkRuntimeApi::new(&sdk);
let ec2_cluster = user_ec2_cluster(&rt, project, &who, &region).await;
remove_workspace_resources(
&SdkTaskApi::new(&sdk),
&SdkWorkspaceApi::new(&sdk),
&SdkRuntimeApi::new(&sdk),
&rt,
&RemoveRequest {
env: &env,
region: &region,
@@ -1001,6 +1444,7 @@ pub async fn remove_resources(
stored_access_point_id: aws.and_then(|a| a.access_point_id.as_deref()),
stored_workspace_access_point_id: aws
.and_then(|a| a.workspace_access_point_id.as_deref()),
ec2_cluster: ec2_cluster.as_deref(),
discard_data,
},
)
@@ -1018,9 +1462,20 @@ pub enum Reconciled {
/// The task is up (or on its way): re-open the tunnel.
Live(Box<TaskInfo>),
/// The task stopped, or ECS no longer knows it: the project is Stopped.
/// Also an EC2 task with ECS Exec off that could never be attached to
/// (placed, or orphaned): reconcile stopped it.
Gone,
/// An EC2 task with ECS Exec off, still waiting for an instance and
/// recent: a start (perhaps this app's, before a restart) was
/// interrupted mid cold start. The project is Error, its ARN kept so
/// Stop still reaches the task; [`INTERRUPTED`] says what to do.
Interrupted,
}
/// What reconcile says about [`Reconciled::Interrupted`].
pub const INTERRUPTED: &str = "A start of this workspace was interrupted while its EC2 \
instance came up; Start again (or Stop it).";
/// Ask ECS about a project's task: the stored ARN, else (none stored) its
/// live task found by family and tags.
pub async fn reconcile_task<T: TaskApi>(
@@ -1041,11 +1496,9 @@ pub async fn reconcile_task<T: TaskApi>(
// or a crash before the store write) may have left the project a newer
// one: look before calling it Gone (m3).
if task_arn.is_some() {
if let Some(t) =
find_project_task(api, &find.cluster, &find.family, &find.owner_user_id, &find.project_id)
.await
.map_err(s)?
.filter(|t| t.arn != arn && is_live(t))
if let Some(t) = find_in_clusters(api, find)
.await?
.filter(|t| t.arn != arn && is_live(t))
{
return Ok(Reconciled::Live(Box::new(t)));
}
@@ -1100,7 +1553,33 @@ pub async fn reconcile_project<T: TaskApi>(
find: &FindTask,
) -> Result<Reconciled, String> {
let arn = project.aws.as_ref().and_then(|a| a.task_arn.clone());
let r = reconcile_task(api, arn.as_deref(), find).await?;
let mut r = reconcile_task(api, arn.as_deref(), find).await?;
// A task with ECS Exec off can never be attached to (a cold EC2 start
// the app was closed during): never Running. Placed, or too old for any
// start to still wait for it → stop it (then it's Gone); still waiting
// and recent → Interrupted, left for a Start or Stop.
if let Reconciled::Live(t) = &r {
if !t.exec_enabled {
if waiting_for_instance(t) && !is_orphaned(t, WaitTiming::ec2(), now_unix()) {
if arn.as_deref() != Some(t.arn.as_str()) {
record_task(store, &project.id, Some(&live_record(t)));
}
store.update_status(&project.id, ProjectStatus::Error)?;
return Ok(Reconciled::Interrupted);
}
log::warn!("Project {}: its task {} has ECS Exec off; stopping it", project.id, t.arn);
let cluster = cluster_of_task(&t.arn).ok_or_else(|| format!("{} is not a task ARN with a cluster", t.arn))?;
if let Err(e) = bounded_stop(api.stop_task(cluster, &t.arn, REPLACED_REASON)).await {
// It may still run: keep (or record) its ARN so Stop finds it.
if arn.as_deref() != Some(t.arn.as_str()) {
record_task(store, &project.id, Some(&live_record(t)));
}
store.update_status(&project.id, ProjectStatus::Error)?;
return Err(format!("Could not stop the workspace's unreachable task {}: {e}", t.arn));
}
r = Reconciled::Gone;
}
}
match &r {
Reconciled::Gone => {
store.update_status(&project.id, ProjectStatus::Stopped)?;
@@ -1108,37 +1587,42 @@ pub async fn reconcile_project<T: TaskApi>(
}
Reconciled::Live(t) => {
if arn.as_deref() != Some(t.arn.as_str()) {
record_task(
store,
&project.id,
Some(&TaskRecord {
task_arn: t.arn.clone(),
access_point_id: None,
task_definition_arn: Some(t.task_definition_arn.clone()),
workspace_access_point_id: None,
}),
);
record_task(store, &project.id, Some(&live_record(t)));
}
if project.status != ProjectStatus::Running {
store.update_status(&project.id, ProjectStatus::Running)?;
}
}
Reconciled::Interrupted => {}
}
Ok(r)
}
/// The store record of a live task reconcile found (no access points: the
/// stored ones stand).
fn live_record(t: &TaskInfo) -> TaskRecord {
TaskRecord {
task_arn: t.arn.clone(),
access_point_id: None,
task_definition_arn: Some(t.task_definition_arn.clone()),
workspace_access_point_id: None,
}
}
/// `reconcile_project_statuses` for an AWS project: decide, and when the
/// task is live re-open its tunnel in the background (terminals wait for
/// it through [`tunnels::wait_endpoint`]), then — copy-at-connect — copy the
/// credentials in, calling `on_connect_warning` if that fails — after
/// opening its port mappings (`on_ports_warning` for any that fail). Never
/// blocks on the tunnel.
/// blocks on the tunnel. `on_interrupted_warning` reports a start that an
/// app restart interrupted ([`INTERRUPTED`]), on its own toast topic.
pub async fn reconcile(
project: &Project,
settings: &AppSettings,
store: Arc<ProjectsStore>,
on_connect_warning: impl FnOnce(String) + Send + 'static,
on_ports_warning: impl FnOnce(String) + Send + 'static,
on_interrupted_warning: impl FnOnce(String) + Send + 'static,
) {
let (connection_id, region, sdk) = match project_sdk(project, settings).await {
Ok(v) => v,
@@ -1155,7 +1639,8 @@ pub async fn reconcile(
}
};
let api = SdkTaskApi::new(&sdk);
let find = FindTask::new(project, &region, &who.user_id);
let ec2 = user_ec2_cluster(&SdkRuntimeApi::new(&sdk), project, &who, &region).await;
let find = FindTask::new(project, &region, &who.user_id, ec2);
match reconcile_project(&api, &store, project, &find).await {
Ok(Reconciled::Live(task)) => {
let generation = tunnels::begin_reattach(&project.id);
@@ -1188,6 +1673,10 @@ pub async fn reconcile(
Ok(Reconciled::Gone) => {
log::info!("Project '{}' ({}): its task is gone — set to Stopped", project.name, project.id)
}
Ok(Reconciled::Interrupted) => {
log::info!("Project '{}' ({}): {INTERRUPTED}", project.name, project.id);
on_interrupted_warning(INTERRUPTED.to_string());
}
Err(e) => log::warn!("Project {}: could not check its task ({e}); status left as is", project.id),
}
}
@@ -1227,14 +1716,26 @@ async fn reattach(
.map_err(s)?;
let token = read_agent_token(&rt, &env.name, &ok, &project.id).await.map_err(s)?;
let logs = TaskLogs { group: env.log_group.clone(), stream_prefix: log_stream_prefix(&ok, &project.id) };
let task = wait_ready(&api, &env.cluster_arn, &task.arn, &logs, WaitTiming::default(), |_| {})
// An EC2 task may still be waiting for its instance after a restart.
let cluster = task_cluster(&env, &task.arn);
let timing = if ec2_cluster_of_task(&env.name, &task.arn).is_some() {
WaitTiming::ec2()
} else {
WaitTiming::default()
};
let task = wait_ready(&api, &cluster, &task.arn, &logs, timing, |_| {})
.await
.map_err(s)?;
// Reconcile never hands over an exec-off task; never wait on a tunnel
// ECS Exec can't open either.
if !task.exec_enabled {
return Err(format!("The task {} has ECS Exec off; Start the project again", task.arn));
}
let (forward, endpoint) = SsmConnector { sdk: ctx.sdk.clone() }.connect(&task, &token).await?;
let tunnel = LiveTunnel {
task: ecs_task(&task)?,
task_arn: task.arn.clone(),
cluster_arn: env.cluster_arn.clone(),
cluster_arn: cluster,
connection_id: ctx.connection_id,
region: ctx.region,
sdk: ctx.sdk,
+676 -6
View File
@@ -56,6 +56,9 @@ fn task(p: &Project, id: &str, last: &str, ready: bool) -> TaskInfo {
stopped_reason: (last == "STOPPED").then(|| "Essential container exited".into()),
task_definition_arn: format!("arn:aws:ecs:{REGION}:{ACCOUNT}:task-definition/{}:1", family(p)),
runtime_id: ready.then(|| "rt-1".into()),
container_instance_arn: None,
exec_enabled: true,
created_unix: None,
exec_agent_running: ready,
exit_code: None,
owner: Some(USER.into()),
@@ -85,6 +88,8 @@ struct Fake {
tasks: Mutex<HashMap<String, TaskInfo>>,
/// What RunTask starts.
run: Option<TaskInfo>,
/// When non-empty, what successive RunTasks start (before `run`).
runs: Mutex<Vec<TaskInfo>>,
logs: Vec<String>,
role: Option<Knowable>,
connect_fails: bool,
@@ -94,10 +99,27 @@ struct Fake {
rotated_by_loser: bool,
/// StopTask is refused (expired credentials, say).
stop_fails: bool,
/// StopTask is accepted but the task never reaches STOPPED.
stop_noop: bool,
/// What `on_task` was told, in order.
recorded: Mutex<Vec<Option<String>>>,
/// `group stream limit` of each log read.
log_reads: Mutex<Vec<String>>,
/// The caller's EC2 record (absent by default).
user_record: Option<crate::aws::environments::Param>,
/// Reading the EC2 record fails (throttled, say).
user_record_fails: bool,
/// RunTask requests, in order.
run_requests: Mutex<Vec<RunTaskInput>>,
/// The clusters ListTasks / StopTask / DescribeTasks were asked about.
list_clusters: Mutex<Vec<String>>,
stop_clusters: Mutex<Vec<String>>,
describe_clusters: Mutex<Vec<String>>,
/// When set, ListTasks answers per cluster (empty for any other).
listed_by_cluster: Option<HashMap<String, Vec<String>>>,
/// Successive states of the run task, one per DescribeTasks (the last
/// repeats); overrides `tasks` for that ARN.
script: Mutex<Vec<TaskInfo>>,
}
impl Fake {
@@ -117,26 +139,54 @@ impl Fake {
}
impl TaskApi for Fake {
async fn run_task(&self, _req: RunTaskInput) -> Result<TaskInfo, AwsError> {
async fn run_task(&self, req: RunTaskInput) -> Result<TaskInfo, AwsError> {
self.log("run");
let t = self.run.clone().expect("a run was expected");
self.run_requests.lock().unwrap().push(req);
let queued = {
let mut q = self.runs.lock().unwrap();
(!q.is_empty()).then(|| q.remove(0))
};
let t = queued.or_else(|| self.run.clone()).expect("a run was expected");
self.tasks.lock().unwrap().insert(t.arn.clone(), t.clone());
Ok(t)
}
async fn describe(&self, _c: &str, arns: &[String]) -> Result<Vec<TaskInfo>, AwsError> {
async fn describe(&self, c: &str, arns: &[String]) -> Result<Vec<TaskInfo>, AwsError> {
self.describe_clusters.lock().unwrap().push(c.into());
{
let mut script = self.script.lock().unwrap();
if let Some(first) = script.first().cloned().filter(|t| arns.contains(&t.arn)) {
if script.len() > 1 {
script.remove(0);
}
return Ok(vec![first]);
}
}
let t = self.tasks.lock().unwrap();
Ok(arns.iter().filter_map(|a| t.get(a).cloned()).collect())
}
async fn list_project_tasks(&self, _c: &str, _f: &str) -> Result<Vec<String>, AwsError> {
async fn list_project_tasks(&self, c: &str, _f: &str) -> Result<Vec<String>, AwsError> {
self.log("list");
self.list_clusters.lock().unwrap().push(c.into());
if let Some(by) = &self.listed_by_cluster {
return Ok(by.get(c).cloned().unwrap_or_default());
}
let mut l = self.listed.lock().unwrap();
Ok(if l.len() > 1 { l.remove(0) } else { l.first().cloned().unwrap_or_default() })
}
async fn stop_task(&self, _c: &str, arn: &str, reason: &str) -> Result<(), AwsError> {
async fn stop_task(&self, c: &str, arn: &str, reason: &str) -> Result<(), AwsError> {
self.log(format!("stop {arn} {reason}"));
self.stop_clusters.lock().unwrap().push(c.into());
if self.stop_fails {
return Err(AwsError::Other { message: "ExpiredToken".into() });
}
// As ECS does: a stopped task is no longer live.
if self.stop_noop {
return Ok(());
}
if let Some(t) = self.tasks.lock().unwrap().get_mut(arn) {
t.desired_status = "STOPPED".into();
t.last_status = "STOPPED".into();
}
Ok(())
}
async fn log_tail(&self, g: &str, st: &str, l: i32) -> Result<Vec<String>, AwsError> {
@@ -213,6 +263,13 @@ impl RuntimeApi for Fake {
self.log(format!("record {env}"));
Ok(Some(record()))
}
async fn user_record(&self, env: &str, _owner_key: &str) -> Result<crate::aws::environments::Param, AwsError> {
self.log(format!("user-record {env}"));
if self.user_record_fails {
return Err(AwsError::Other { message: "Throttling".into() });
}
Ok(self.user_record.clone().unwrap_or(crate::aws::environments::Param::Missing))
}
async fn role_exists(&self, _name: &str) -> Result<Knowable, AwsError> {
self.log("role");
Ok(self.role.clone().unwrap_or(Knowable::Yes))
@@ -284,6 +341,10 @@ fn inputs<'a>(
interval: Duration::from_millis(1),
timeout: Duration::from_millis(30),
},
ec2_wait: WaitTiming {
interval: Duration::from_millis(1),
timeout: Duration::from_millis(30),
},
access_point_wait: PollTiming {
interval: Duration::from_millis(1),
timeout: Duration::from_millis(30),
@@ -561,6 +622,21 @@ fn the_sandbox_is_forced_off_on_aws() {
let settings = &pairs.iter().find(|(k, _)| k == "CLAUDE_CODE_SETTINGS_JSON").unwrap().1;
let v: serde_json::Value = serde_json::from_str(settings).unwrap();
assert_eq!(v["sandbox"]["enabled"], false, "{settings}");
assert!(!sandbox_allowed(&p));
}
/// E8, measured live (phase 7): bubblewrap can't create a user namespace
/// in an ECS task on the ECS-optimized AMI either (Docker's default seccomp
/// profile refuses it without CAP_SYS_ADMIN), so EC2 keeps it off too.
#[test]
fn the_sandbox_is_forced_off_on_ec2_too() {
let mut p = ec2_project();
p.sandbox_mode_enabled = true;
assert!(!sandbox_allowed(&p));
let (pairs, _) = aws_env(&p, &AppSettings::default()).unwrap();
let settings = &pairs.iter().find(|(k, _)| k == "CLAUDE_CODE_SETTINGS_JSON").unwrap().1;
let v: serde_json::Value = serde_json::from_str(settings).unwrap();
assert_eq!(v["sandbox"]["enabled"], false, "{settings}");
}
#[test]
@@ -636,7 +712,7 @@ async fn an_aws_projects_target_is_its_registered_tunnel() {
}
fn find(p: &Project) -> FindTask {
FindTask::new(p, REGION, USER)
FindTask::new(p, REGION, USER, None)
}
fn store_with(p: &Project) -> (tempfile::TempDir, ProjectsStore) {
@@ -1241,3 +1317,597 @@ mod gateway_sidecar_inputs {
}
}
// ------------------------------------------------------------- EC2 (phase 7)
const USER_CLUSTER: &str = "triple-c-dev--u-alice";
fn user_cluster_arn() -> String {
format!("arn:aws:ecs:{REGION}:{ACCOUNT}:cluster/{USER_CLUSTER}")
}
fn ec2_task_arn(id: &str) -> String {
format!("arn:aws:ecs:{REGION}:{ACCOUNT}:task/{USER_CLUSTER}/{id}")
}
fn ec2_project() -> Project {
let mut p = aws_project(StorageMode::Ephemeral);
let a = p.aws.as_mut().unwrap();
a.launch_type = LaunchType::Ec2;
a.cpu = 1024;
a.memory_mib = 4096;
p
}
fn ec2_task(p: &Project, id: &str, last: &str, ready: bool, placed: bool) -> TaskInfo {
TaskInfo {
arn: ec2_task_arn(id),
container_instance_arn: placed.then(|| {
format!("arn:aws:ecs:{REGION}:{ACCOUNT}:container-instance/{USER_CLUSTER}/ci-1")
}),
..task(p, id, last, ready)
}
}
fn user_record_param() -> crate::aws::environments::Param {
crate::aws::environments::Param::Value(
serde_json::json!({
"schema": 1, "instance_type": "m7i.large", "max_instances": 2, "user_tag": "alice",
"cluster_arn": user_cluster_arn(),
"capacity_provider": USER_CLUSTER,
})
.to_string(),
)
}
#[tokio::test]
async fn an_ec2_start_runs_on_the_users_capacity_and_says_when_it_waits_for_an_instance() {
let p = ec2_project();
let fake = Fake {
user_record: Some(user_record_param()),
run: Some(ec2_task(&p, "e1", "PROVISIONING", false, false)),
..Default::default()
};
// Cold start: no instance, then one is placed, then the task is up.
*fake.script.lock().unwrap() = vec![
ec2_task(&p, "e1", "PROVISIONING", false, false),
ec2_task(&p, "e1", "PROVISIONING", false, false),
ec2_task(&p, "e1", "PENDING", false, true),
ec2_task(&p, "e1", "RUNNING", true, true),
];
let (r, steps) = start(&fake, &p).await;
let started = r.unwrap();
assert!(started.created);
assert_eq!(started.task.arn, ec2_task_arn("e1"));
assert_eq!(
steps,
vec![
PROGRESS_READING,
PROGRESS_PREPARING,
PROGRESS_STARTING,
PROGRESS_WAITING,
PROGRESS_INSTANCE,
PROGRESS_WAITING,
PROGRESS_CONNECTING
]
);
assert_eq!(PROGRESS_INSTANCE, "Starting an instance (about 3 minutes)…");
// The record is read before anything is written; the project's task is
// looked for in both clusters, then once more where it will run.
let calls = fake.calls();
let pos = |c: &str| calls.iter().position(|x| x == c).unwrap();
assert!(pos("user-record dev") < pos("put-secret"), "{calls:?}");
assert_eq!(
*fake.list_clusters.lock().unwrap(),
[cluster_arn(), user_cluster_arn(), user_cluster_arn()]
);
let req = &fake.run_requests.lock().unwrap()[0];
assert_eq!(req.cluster(), Some(user_cluster_arn().as_str()));
assert_eq!(req.capacity_provider_strategy()[0].capacity_provider(), USER_CLUSTER);
assert!(req.network_configuration().is_none() && req.launch_type().is_none());
assert!(fake.describe_clusters.lock().unwrap().iter().all(|c| *c == user_cluster_arn()));
}
/// ECS turns ECS Exec off on a RunTask that has to wait for an instance
/// (verified live): that task only brings the instance up; once placed it
/// is stopped and the task run again, onto the instance, with ECS Exec.
#[tokio::test]
async fn a_cold_ec2_start_runs_the_task_again_once_an_instance_is_up() {
let p = ec2_project();
let execless = |last: &str, placed: bool| TaskInfo { exec_enabled: false, ..ec2_task(&p, "e1", last, false, placed) };
let fake = Fake {
user_record: Some(user_record_param()),
runs: Mutex::new(vec![execless("PROVISIONING", false), ec2_task(&p, "e9", "PENDING", false, true)]),
..Default::default()
};
*fake.script.lock().unwrap() = vec![
execless("PROVISIONING", false),
execless("PENDING", true),
TaskInfo { desired_status: "STOPPED".into(), ..execless("PENDING", true) },
execless("STOPPED", true),
ec2_task(&p, "e9", "RUNNING", true, true),
];
let (r, steps) = start(&fake, &p).await;
let started = r.unwrap();
assert_eq!(started.task.arn, ec2_task_arn("e9"));
let calls = fake.calls();
assert_eq!(calls.iter().filter(|c| *c == "run").count(), 2, "{calls:?}");
let stop = format!("stop {} {STOP_REASON}", ec2_task_arn("e1"));
let pos = |c: &str| calls.iter().rposition(|x| x == c).unwrap();
assert!(calls.iter().position(|c| *c == stop).unwrap() < pos("run"), "{calls:?}");
assert_eq!(*fake.stop_clusters.lock().unwrap(), [user_cluster_arn()]);
// Both requests ask for ECS Exec on the user's capacity.
for req in fake.run_requests.lock().unwrap().iter() {
assert_eq!(req.enable_execute_command(), Some(true));
assert_eq!(req.capacity_provider_strategy()[0].capacity_provider(), USER_CLUSTER);
}
// The second task is the one recorded.
assert_eq!(*fake.recorded.lock().unwrap(), vec![Some(ec2_task_arn("e1")), Some(ec2_task_arn("e9"))]);
assert_eq!(
steps,
vec![
PROGRESS_READING,
PROGRESS_PREPARING,
PROGRESS_STARTING,
PROGRESS_WAITING,
PROGRESS_INSTANCE,
PROGRESS_WAITING,
PROGRESS_CONNECTING
]
);
}
#[tokio::test]
async fn a_second_task_without_ecs_exec_is_stopped_and_reported() {
let p = ec2_project();
let execless = |id: &str, last: &str, placed: bool| TaskInfo { exec_enabled: false, ..ec2_task(&p, id, last, false, placed) };
let fake = Fake {
user_record: Some(user_record_param()),
runs: Mutex::new(vec![execless("e1", "PROVISIONING", false), execless("e9", "PENDING", true)]),
..Default::default()
};
*fake.script.lock().unwrap() = vec![execless("e1", "PENDING", true), execless("e1", "STOPPED", true)];
let (r, _) = start(&fake, &p).await;
assert_eq!(r.err().unwrap().lines().next(), Some(EXEC_OFF));
let calls = fake.calls();
assert!(calls.contains(&format!("stop {} {STOP_REASON}", ec2_task_arn("e9"))), "{calls:?}");
}
#[tokio::test]
async fn a_cold_ec2_task_stopped_before_placement_ends_the_start_with_its_reason() {
let p = ec2_project();
let execless = |last: &str| TaskInfo { exec_enabled: false, ..ec2_task(&p, "e1", last, false, false) };
let fake = Fake {
user_record: Some(user_record_param()),
runs: Mutex::new(vec![execless("PROVISIONING")]),
..Default::default()
};
*fake.script.lock().unwrap() = vec![
execless("PROVISIONING"),
TaskInfo {
desired_status: "STOPPED".into(),
stopped_reason: Some("Task stopped by user".into()),
..execless("PROVISIONING")
},
];
let (r, _) = start(&fake, &p).await;
let e = r.err().unwrap();
assert!(e.contains("Task stopped by user"), "{e}");
assert_eq!(fake.calls().iter().filter(|c| *c == "run").count(), 1, "no second RunTask");
}
#[tokio::test]
async fn an_execless_task_still_waiting_for_an_instance_is_another_starts() {
let mut p = ec2_project();
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("other"));
let other = TaskInfo { exec_enabled: false, ..ec2_task(&p, "other", "PROVISIONING", false, false) };
let fake = Fake {
user_record: Some(user_record_param()),
listed_by_cluster: Some(HashMap::from([(user_cluster_arn(), vec![ec2_task_arn("other")])])),
..Default::default()
}
.with_task(other);
let (r, _) = start(&fake, &p).await;
assert_eq!(r.err().unwrap(), INSTANCE_RACE);
assert!(!fake.called("run") && !fake.called("stop"), "{:?}", fake.calls());
}
#[tokio::test]
async fn the_cold_reruns_launch_never_doubles_another_starts_task() {
let p = ec2_project();
let execless = |last: &str, placed: bool| TaskInfo { exec_enabled: false, ..ec2_task(&p, "e1", last, false, placed) };
let winner = ec2_task(&p, "w1", "PENDING", false, true);
let fake = Fake {
user_record: Some(user_record_param()),
runs: Mutex::new(vec![execless("PROVISIONING", false)]),
// Nothing for the start's own finds; the winner's task by the rerun.
listed: Mutex::new(vec![vec![], vec![], vec![], vec![ec2_task_arn("w1")]]),
..Default::default()
}
.with_task(winner);
*fake.script.lock().unwrap() = vec![execless("PENDING", true), execless("STOPPED", true)];
let (r, _) = start(&fake, &p).await;
assert_eq!(r.err().unwrap(), START_RACE);
assert_eq!(fake.calls().iter().filter(|c| *c == "run").count(), 1, "{:?}", fake.calls());
// The winner's task is recorded, as on the first launch's race.
assert_eq!(fake.recorded.lock().unwrap().last().cloned().flatten(), Some(ec2_task_arn("w1")));
}
#[tokio::test]
async fn an_execless_task_that_cannot_be_stopped_fails_the_start() {
let mut p = ec2_project();
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("old"));
let old = TaskInfo { exec_enabled: false, ..ec2_task(&p, "old", "RUNNING", false, true) };
let fake = Fake {
user_record: Some(user_record_param()),
listed_by_cluster: Some(HashMap::from([(user_cluster_arn(), vec![ec2_task_arn("old")])])),
stop_fails: true,
..Default::default()
}
.with_task(old);
let (r, _) = start(&fake, &p).await;
let e = r.err().unwrap();
assert!(e.contains("Could not stop") && e.contains("ExpiredToken"), "{e}");
assert!(!fake.called("run"));
}
fn now_unix() -> i64 {
std::time::SystemTime::now().duration_since(std::time::UNIX_EPOCH).unwrap().as_secs() as i64
}
/// An exec-off task still waiting for an instance, older than any start
/// could still be waiting (a crashed start's orphan), is replaced; a fresh
/// one is another start's.
#[tokio::test]
async fn an_orphaned_waiting_execless_task_is_replaced_and_a_fresh_one_is_left_alone() {
let mut p = ec2_project();
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("old"));
let waiting = |created: i64| TaskInfo {
exec_enabled: false,
created_unix: Some(created),
..ec2_task(&p, "old", "PROVISIONING", false, false)
};
let fake_with = |t: TaskInfo| {
Fake {
user_record: Some(user_record_param()),
listed_by_cluster: Some(HashMap::from([(user_cluster_arn(), vec![ec2_task_arn("old")])])),
run: Some(ec2_task(&p, "new", "RUNNING", true, true)),
..Default::default()
}
.with_task(t)
};
let fake = fake_with(waiting(now_unix() - 3600));
let (r, _) = start(&fake, &p).await;
assert_eq!(r.unwrap().task.arn, ec2_task_arn("new"));
assert!(fake.calls().contains(&format!("stop {} {REPLACED_REASON}", ec2_task_arn("old"))), "{:?}", fake.calls());
let fake = fake_with(waiting(now_unix()));
let (r, _) = start(&fake, &p).await;
let e = r.err().unwrap();
assert_eq!(e, INSTANCE_RACE);
assert!(e.contains("Stop"), "{e}");
assert!(!fake.called("run") && !fake.called("stop"));
}
#[tokio::test]
async fn a_cold_start_whose_task_another_start_replaced_says_so() {
let p = ec2_project();
let execless = |last: &str| TaskInfo { exec_enabled: false, ..ec2_task(&p, "e1", last, false, false) };
let fake = Fake {
user_record: Some(user_record_param()),
runs: Mutex::new(vec![execless("PROVISIONING")]),
..Default::default()
};
*fake.script.lock().unwrap() = vec![TaskInfo {
desired_status: "STOPPED".into(),
stopped_reason: Some(REPLACED_REASON.into()),
..execless("PROVISIONING")
}];
let (r, _) = start(&fake, &p).await;
let e = r.err().unwrap();
assert!(e.starts_with("Another start of this workspace replaced this one"), "{e}");
assert_eq!(fake.calls().iter().filter(|c| *c == "run").count(), 1);
}
#[tokio::test]
async fn an_execless_task_that_never_stops_fails_the_start_without_a_new_run() {
let mut p = ec2_project();
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("old"));
let old = TaskInfo { exec_enabled: false, ..ec2_task(&p, "old", "RUNNING", false, true) };
let fake = Fake {
user_record: Some(user_record_param()),
listed_by_cluster: Some(HashMap::from([(user_cluster_arn(), vec![ec2_task_arn("old")])])),
stop_noop: true,
..Default::default()
}
.with_task(old);
let (r, _) = start(&fake, &p).await;
let e = r.err().unwrap();
assert!(e.contains("did not stop in time"), "{e}");
assert!(fake.called("stop") && !fake.called("run"));
}
#[tokio::test]
async fn a_live_task_with_ecs_exec_off_is_stopped_and_replaced() {
let mut p = ec2_project();
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("old"));
let old = TaskInfo { exec_enabled: false, ..ec2_task(&p, "old", "RUNNING", false, true) };
let fake = Fake {
user_record: Some(user_record_param()),
listed_by_cluster: Some(HashMap::from([(user_cluster_arn(), vec![ec2_task_arn("old")])])),
run: Some(ec2_task(&p, "new", "RUNNING", true, true)),
..Default::default()
}
.with_task(old);
let (r, _) = start(&fake, &p).await;
assert_eq!(r.unwrap().task.arn, ec2_task_arn("new"));
let calls = fake.calls();
let stop = calls.iter().position(|c| *c == format!("stop {} {REPLACED_REASON}", ec2_task_arn("old"))).expect("the old task is stopped");
assert!(stop < calls.iter().position(|c| c == "run").unwrap(), "{calls:?}");
}
#[tokio::test]
async fn a_fargate_start_never_reads_the_ec2_record() {
let p = aws_project(StorageMode::Ephemeral);
let fake = Fake {
user_record: Some(user_record_param()),
run: Some(task(&p, "f1", "RUNNING", true)),
..Default::default()
};
let (r, _) = start(&fake, &p).await;
r.unwrap();
assert!(!fake.called("user-record"));
assert_eq!(*fake.list_clusters.lock().unwrap(), [cluster_arn(), cluster_arn()]);
}
#[tokio::test]
async fn an_ec2_start_without_capacity_says_so_before_anything_is_written() {
let p = ec2_project();
let fake = Fake::default();
let (r, _) = start(&fake, &p).await;
let e = r.err().unwrap();
assert_eq!(
e,
"You have no EC2 capacity in dev; ask an admin to add it to your onboarding (or run the \
project on Fargate)."
);
assert!(!fake.called("put-secret") && !fake.called("register") && !fake.called("run"));
// A record that can't be used is not mistaken for none.
let fake = Fake {
user_record: Some(crate::aws::environments::Param::Value("{}".into())),
..Default::default()
};
let (r, _) = start(&fake, &p).await;
let e = r.err().unwrap();
assert!(e.contains("can't be used") && e.contains("re-apply your onboarding"), "{e}");
assert!(!fake.called("put-secret"));
}
#[tokio::test]
async fn an_ec2_task_that_never_gets_an_instance_is_stopped_with_what_to_check() {
let p = ec2_project();
let fake = Fake {
user_record: Some(user_record_param()),
run: Some(ec2_task(&p, "e2", "PROVISIONING", false, false)),
..Default::default()
};
let (r, steps) = start(&fake, &p).await;
let e = r.err().unwrap();
assert!(e.contains("No EC2 instance had room"), "{e}");
assert!(e.contains("maximum of 2 instances"), "{e}");
assert!(steps.contains(&PROGRESS_INSTANCE.to_string()));
assert!(fake.calls().contains(&format!("stop {} {STOP_REASON}", ec2_task_arn("e2"))));
assert_eq!(*fake.stop_clusters.lock().unwrap(), [user_cluster_arn()]);
assert_eq!(*fake.recorded.lock().unwrap(), vec![Some(ec2_task_arn("e2")), None]);
}
/// The record is gone (capacity removed) while the task still runs: the
/// stored ARN leads to it, and attaching needs no record.
#[tokio::test]
async fn a_running_ec2_task_is_attached_to_even_after_the_record_is_gone() {
let mut p = ec2_project();
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("e3"));
let live = ec2_task(&p, "e3", "RUNNING", true, true);
let fake = Fake {
listed_by_cluster: Some(HashMap::from([(user_cluster_arn(), vec![live.arn.clone()])])),
..Default::default()
}
.with_task(live);
let (r, _) = start(&fake, &p).await;
let started = r.unwrap();
assert!(!started.created);
assert_eq!(started.task.arn, ec2_task_arn("e3"));
assert!(!fake.called("run") && !fake.called("put-secret"));
assert_eq!(*fake.list_clusters.lock().unwrap(), [cluster_arn(), user_cluster_arn()]);
assert!(fake.describe_clusters.lock().unwrap().iter().all(|c| *c == user_cluster_arn()));
}
#[tokio::test]
async fn an_ec2_projects_task_is_found_and_watched_on_the_users_cluster() {
// FindTask: the environment's cluster, the record's, and the stored
// task's per-user cluster, without repeats.
let mut p = ec2_project();
let f = FindTask::new(&p, REGION, USER, Some(user_cluster_arn()));
assert_eq!(f.clusters, ["triple-c-dev".to_string(), user_cluster_arn()]);
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("e4"));
assert_eq!(find(&p).clusters, ["triple-c-dev".to_string(), user_cluster_arn()]);
let f = FindTask::new(&p, REGION, USER, Some(user_cluster_arn()));
assert_eq!(f.clusters.len(), 2);
// Reconcile with no ARN stored finds the task on the user's cluster.
let mut orphan = ec2_project();
orphan.status = ProjectStatus::Running;
let live = ec2_task(&orphan, "e5", "RUNNING", true, true);
let fake = Fake {
listed_by_cluster: Some(HashMap::from([(user_cluster_arn(), vec![live.arn.clone()])])),
..Default::default()
}
.with_task(live);
let (_d, store) = store_with(&orphan);
let find = FindTask::new(&orphan, REGION, USER, Some(user_cluster_arn()));
assert!(matches!(
reconcile_project(&fake, &store, &orphan, &find).await.unwrap(),
Reconciled::Live(_)
));
assert_eq!(store.get(&orphan.id).unwrap().aws.unwrap().task_arn, Some(ec2_task_arn("e5")));
// The watch groups by the task's own cluster.
let mut running = ec2_project();
running.status = ProjectStatus::Running;
running.container_id = Some("http://127.0.0.1:61301/".into());
running.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("e6"));
let w = watched_task(&running).expect("watched");
assert_eq!(w.group.cluster_arn, user_cluster_arn());
let mut fargate = aws_project(StorageMode::Ephemeral);
fargate.status = ProjectStatus::Running;
fargate.container_id = Some("http://127.0.0.1:61302/".into());
fargate.aws.as_mut().unwrap().task_arn = Some(task_arn("f6"));
assert_ne!(watched_task(&fargate).unwrap().group, w.group);
}
#[tokio::test]
async fn stopping_an_ec2_task_by_its_arn_uses_its_cluster() {
let fake = Fake::default();
stop_task_arn(&fake, &ec2_task_arn("e7")).await.unwrap();
assert_eq!(*fake.stop_clusters.lock().unwrap(), [USER_CLUSTER]);
}
/// E7: the entrypoint installs the VPN skill from `VPN_SUPPORT_ENABLED`, as
/// on Docker; an EC2 project with VPN support gets `1`, Fargate `0`.
#[test]
fn vpn_support_reaches_an_ec2_workspace_as_on_docker() {
let vpn = |p: &Project| {
let (pairs, _) = aws_env(p, &AppSettings::default()).unwrap();
pairs.into_iter().find(|(k, _)| k == "VPN_SUPPORT_ENABLED").map(|(_, v)| v)
};
let mut p = ec2_project();
p.vpn_support_enabled = true;
assert_eq!(vpn(&p).as_deref(), Some("1"));
p.vpn_support_enabled = false;
assert_eq!(vpn(&p).as_deref(), Some("0"));
assert_eq!(vpn(&aws_project(StorageMode::Ephemeral)).as_deref(), Some("0"));
}
/// Host settings' view of the user's EC2 capacity: the record's compute with
/// the fit limits, none, unusable (with the reason) or unknown.
#[test]
fn ec2_capacity_reads_as_host_settings_need_it() {
use crate::aws::environments::{UserCompute, UserRecord};
let compute = UserCompute {
instance_type: "m7i.large".into(),
max_instances: 1,
user_tag: "alice".into(),
cluster_arn: "arn:aws:ecs:us-west-2:123456789012:cluster/triple-c-dev--u-alice".into(),
capacity_provider: "triple-c-dev--u-alice".into(),
};
let c = ec2_capacity_from(Some(UserRecord::Valid(compute.clone())));
assert_eq!(
c,
Ec2Capacity::Ec2 { compute: compute.clone(), available_cpu: Some(1920), available_memory_mib: Some(7373) }
);
let v = serde_json::to_value(&c).unwrap();
assert_eq!(v["state"], "ec2");
assert_eq!(v["compute"]["instance_type"], "m7i.large");
assert_eq!(v["available_memory_mib"], 7373);
let unknown_type = UserCompute { instance_type: "x9.huge".into(), ..compute };
assert!(matches!(
ec2_capacity_from(Some(UserRecord::Valid(unknown_type))),
Ec2Capacity::Ec2 { available_cpu: None, available_memory_mib: None, .. }
));
assert_eq!(serde_json::to_value(ec2_capacity_from(Some(UserRecord::Absent))).unwrap()["state"], "none");
let v = serde_json::to_value(ec2_capacity_from(Some(UserRecord::Unusable("schema is not 1".into())))).unwrap();
assert_eq!(v, serde_json::json!({ "state": "unusable", "reason": "schema is not 1" }));
assert_eq!(serde_json::to_value(ec2_capacity_from(None)).unwrap(), serde_json::json!({ "state": "unknown" }));
}
/// Reconcile never marks an exec-off task Running (an app restart during a
/// cold EC2 start): a placed or orphaned one is stopped (its ARN cleared
/// only once the stop succeeded); a recent one still waiting is Error with
/// its ARN kept.
#[tokio::test]
async fn reconcile_never_attaches_to_a_task_with_ecs_exec_off() {
let mut p = ec2_project();
p.status = ProjectStatus::Running;
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("x"));
let execless = |last: &str, placed: bool, created: i64| TaskInfo {
exec_enabled: false,
created_unix: Some(created),
..ec2_task(&p, "x", last, false, placed)
};
let find = FindTask::new(&p, REGION, USER, Some(user_cluster_arn()));
let stopped_with_replaced = |f: &Fake| f.calls().contains(&format!("stop {} {REPLACED_REASON}", ec2_task_arn("x")));
// Placed: stopped, Stopped, ARN cleared.
let fake = Fake::default().with_task(execless("PENDING", true, now_unix()));
let (_d, store) = store_with(&p);
assert_eq!(reconcile_project(&fake, &store, &p, &find).await.unwrap(), Reconciled::Gone);
assert!(stopped_with_replaced(&fake), "{:?}", fake.calls());
let after = store.get(&p.id).unwrap();
assert_eq!(after.status, ProjectStatus::Stopped);
assert_eq!(after.aws.unwrap().task_arn, None);
// Orphaned (still waiting, but older than any start waits): the same.
let fake = Fake::default().with_task(execless("PROVISIONING", false, now_unix() - 3600));
let (_d, store) = store_with(&p);
assert_eq!(reconcile_project(&fake, &store, &p, &find).await.unwrap(), Reconciled::Gone);
assert!(stopped_with_replaced(&fake));
assert_eq!(store.get(&p.id).unwrap().status, ProjectStatus::Stopped);
// The stop fails: Error, the ARN kept.
let fake = Fake { stop_fails: true, ..Default::default() }.with_task(execless("RUNNING", true, now_unix()));
let (_d, store) = store_with(&p);
let e = reconcile_project(&fake, &store, &p, &find).await.unwrap_err();
assert!(e.contains("Could not stop") && e.contains("ExpiredToken"), "{e}");
let after = store.get(&p.id).unwrap();
assert_eq!(after.status, ProjectStatus::Error);
assert_eq!(after.aws.unwrap().task_arn, Some(ec2_task_arn("x")));
// Recent and still waiting: Interrupted (Error), ARN kept, nothing stopped.
let fake = Fake::default().with_task(execless("PROVISIONING", false, now_unix() - 60));
let (_d, store) = store_with(&p);
assert_eq!(reconcile_project(&fake, &store, &p, &find).await.unwrap(), Reconciled::Interrupted);
assert!(!fake.called("stop"));
let after = store.get(&p.id).unwrap();
assert_eq!(after.status, ProjectStatus::Error);
assert_eq!(after.aws.unwrap().task_arn, Some(ec2_task_arn("x")));
assert!(INTERRUPTED.contains("Start again"));
}
#[test]
fn an_orphan_is_older_than_the_ec2_wait_plus_ten_minutes() {
assert_eq!(ORPHAN_MARGIN, Duration::from_secs(10 * 60));
let p = ec2_project();
let at = |age: i64| TaskInfo { created_unix: Some(1_000_000 - age), ..ec2_task(&p, "o", "PROVISIONING", false, false) };
let wait = WaitTiming::ec2();
assert!(!is_orphaned(&at(19 * 60), wait, 1_000_000));
assert!(is_orphaned(&at(21 * 60), wait, 1_000_000));
assert!(!is_orphaned(&TaskInfo { created_unix: None, ..at(0) }, wait, 1_000_000));
}
/// A failed read of the EC2 record does not stop an attach: the stored
/// task's cluster is still searched. Only a new run needs the record.
#[tokio::test]
async fn an_ec2_record_read_error_still_attaches_and_fails_only_a_new_run() {
let mut p = ec2_project();
p.aws.as_mut().unwrap().task_arn = Some(ec2_task_arn("e8"));
let live = ec2_task(&p, "e8", "RUNNING", true, true);
let fake = Fake {
user_record_fails: true,
listed_by_cluster: Some(HashMap::from([(user_cluster_arn(), vec![live.arn.clone()])])),
..Default::default()
}
.with_task(live);
let (r, _) = start(&fake, &p).await;
let started = r.unwrap();
assert!(!started.created);
assert_eq!(started.task.arn, ec2_task_arn("e8"));
assert_eq!(*fake.list_clusters.lock().unwrap(), [cluster_arn(), user_cluster_arn()]);
// Nothing live: the start fails with the read error, before any write.
let fake = Fake { user_record_fails: true, ..Default::default() };
let (r, _) = start(&fake, &p).await;
let e = r.err().unwrap();
assert!(e.contains("Could not read your EC2 capacity record") && e.contains("Throttling"), "{e}");
assert!(!fake.called("put-secret") && !fake.called("run"));
}
+3
View File
@@ -555,6 +555,8 @@ pub fn run() {
commands::project_commands::aws_task_log,
commands::project_commands::aws_workspace_status,
commands::project_commands::aws_environment_max_hours,
commands::project_commands::aws_ec2_capacity,
commands::project_commands::ec2_fit_problem,
commands::project_commands::remove_aws_resources,
commands::project_commands::rebuild_project_container,
commands::project_commands::reconcile_project_statuses,
@@ -644,6 +646,7 @@ pub fn run() {
commands::environment_commands::cancel_aws_deployment,
commands::environment_commands::push_aws_saved_state,
commands::environment_commands::onboard_aws_user,
commands::environment_commands::list_ec2_instance_types,
commands::environment_commands::deploy_aws_permissions_boundary,
commands::environment_commands::save_aws_admin_bundle,
commands::settings_commands::detect_host_timezone,
+305 -16
View File
@@ -521,8 +521,20 @@ fn default_aws_ephemeral_gib() -> u32 {
21
}
/// Where an AWS workspace's task runs.
#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
#[serde(rename_all = "snake_case")]
pub enum LaunchType {
#[default]
Fargate,
Ec2,
}
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
pub struct AwsWorkspaceConfig {
/// Fargate (default; absent in old JSON) or the user's own EC2 instances.
#[serde(default)]
pub launch_type: LaunchType,
#[serde(default = "default_aws_cpu")]
pub cpu: u32,
#[serde(default = "default_aws_memory_mib")]
@@ -568,6 +580,7 @@ pub const MAX_SETUP_SCRIPT_BYTES: usize = 32_768;
impl Default for AwsWorkspaceConfig {
fn default() -> Self {
Self {
launch_type: LaunchType::default(),
cpu: default_aws_cpu(),
memory_mib: default_aws_memory_mib(),
arch: Arch::default(),
@@ -700,23 +713,179 @@ fn describe_sizes(sizes: &[u32]) -> String {
format!("{} to {} MiB in steps of {}", first, last, sizes[1] - sizes[0])
}
pub fn validate_aws_workspace(cfg: &AwsWorkspaceConfig) -> Result<(), String> {
let sizes = fargate_memory_sizes(cfg.cpu).ok_or_else(|| {
format!(
"{} is not a Fargate CPU size (256, 512, 1024, 2048, 4096, 8192 or 16384)",
cfg.cpu
)
})?;
if !sizes.contains(&cfg.memory_mib) {
return Err(format!(
"{} MiB of memory is not valid with {} CPU units: use {}",
cfg.memory_mib,
cfg.cpu,
describe_sizes(&sizes)
));
/// CPU units and memory (MiB) held back on an EC2 instance for the ECS agent
/// and the OS.
pub const EC2_RESERVED_CPU_UNITS: u32 = 128;
pub const EC2_RESERVED_MEMORY_MIB: u32 = 256;
/// The instance families the built-in table knows, in the order the app
/// lists them; a family ending in `g` is Graviton (ARM).
const EC2_FAMILIES: [&str; 8] = ["m7i", "c7i", "r7i", "t3", "m7g", "c7g", "r7g", "t4g"];
const EC2_SIZES: [&str; 4] = ["large", "xlarge", "2xlarge", "4xlarge"];
/// The instance type onboarding offers first (x86; `m7g.large` is its ARM
/// counterpart).
pub const DEFAULT_EC2_INSTANCE_TYPE: &str = "m7i.large";
/// One row of the built-in instance table, for the onboarding UI.
#[derive(Debug, Clone, Serialize, PartialEq, Eq)]
pub struct Ec2InstanceType {
pub instance_type: String,
pub arch: Arch,
pub vcpu: u32,
pub memory_mib: u32,
}
/// Every type [`ec2_instance_spec`] knows (x86 families first), so the UI's
/// list and the app's validation share one table.
pub fn ec2_instance_types() -> Vec<Ec2InstanceType> {
EC2_FAMILIES
.iter()
.flat_map(|family| EC2_SIZES.iter().map(move |size| format!("{family}.{size}")))
.filter_map(|instance_type| {
let (cpu, memory_mib) = ec2_instance_spec(&instance_type)?;
let arch = if instance_type.split('.').next()?.ends_with('g') {
Arch::Arm64
} else {
Arch::X86_64
};
Some(Ec2InstanceType {
instance_type,
arch,
vcpu: cpu / 1024,
memory_mib,
})
})
.collect()
}
/// `(vcpu * 1024, memory MiB)` of a common EC2 instance type, or None when the
/// type is not in the built-in table (m7i, m7g, c7i, c7g, r7i, r7g, t3, t4g).
pub fn ec2_instance_spec(instance_type: &str) -> Option<(u32, u32)> {
let (family, size) = instance_type.split_once('.')?;
let vcpu: u32 = match size {
"large" => 2,
"xlarge" => 4,
"2xlarge" => 8,
"4xlarge" => 16,
_ => return None,
};
// GiB per vCPU: general purpose 4, compute 2, memory 8.
let gib_per_vcpu: u32 = match family {
"m7i" | "m7g" => 4,
"c7i" | "c7g" => 2,
"r7i" | "r7g" => 8,
"t3" | "t4g" => {
// Burstable families: memory is 4 GiB per vCPU at xlarge and above,
// 8 GiB at large; AWS offers no 4xlarge in these families.
return match size {
"large" => Some((2048, 8192)),
"xlarge" => Some((4096, 16384)),
"2xlarge" => Some((8192, 32768)),
_ => None,
};
}
_ => return None,
};
Some((vcpu * 1024, vcpu * gib_per_vcpu * 1024))
}
/// Whether an EC2 workspace's reservations fit an instance type.
#[derive(Debug, Clone, PartialEq, Eq, Serialize)]
#[serde(tag = "kind", rename_all = "snake_case")]
pub enum Fit {
Ok,
/// The reservations exceed what the instance leaves for tasks.
TooBig { available_cpu: u32, available_memory_mib: u32 },
/// The type is not in the table: the caller shows a "must fit" warning.
UnknownType,
}
/// Memory (MiB) an instance of `memory_mib` nominal memory keeps from tasks
/// in [`fits_instance`]: a tenth of it (what ECS doesn't register, with
/// room to spare), and never less than [`EC2_RESERVED_MEMORY_MIB`].
pub fn ec2_memory_reserve_mib(memory_mib: u32) -> u32 {
(memory_mib / 10).max(EC2_RESERVED_MEMORY_MIB)
}
/// Check an EC2 config's reservations against `instance_type`, less the agent
/// and OS reserve. A Fargate config always fits (no instance involved).
///
/// Nominal sizes only, so memory keeps a margin: ECS registers less memory
/// than the instance's nominal (the ECS-optimized AMI and kernel keep some —
/// about 7.5 GiB of an 8 GiB type is registered), and a task asking for more
/// than the registered amount stays PROVISIONING with nothing able to place
/// it. The registered figure can't be read before an instance exists (scale
/// to zero), so the reserve is [`ec2_memory_reserve_mib`]: 10 % of the
/// nominal memory, at least [`EC2_RESERVED_MEMORY_MIB`].
pub fn fits_instance(cfg: &AwsWorkspaceConfig, instance_type: &str) -> Fit {
if cfg.launch_type != LaunchType::Ec2 {
return Fit::Ok;
}
if !(21..=200).contains(&cfg.ephemeral_gib) {
return Err("Ephemeral storage must be 21 to 200 GiB".into());
let Some((available_cpu, available_memory_mib)) = ec2_available(instance_type) else {
return Fit::UnknownType;
};
if cfg.cpu > available_cpu || cfg.memory_mib > available_memory_mib {
Fit::TooBig { available_cpu, available_memory_mib }
} else {
Fit::Ok
}
}
/// What `instance_type` leaves for one workspace's reservations, as
/// `(CPU units, MiB)`: [`fits_instance`]'s limits (the agent's CPU and the
/// memory margin taken off). `None` for a type not in the table.
pub fn ec2_available(instance_type: &str) -> Option<(u32, u32)> {
let (cpu, mem) = ec2_instance_spec(instance_type)?;
Some((cpu - EC2_RESERVED_CPU_UNITS, mem - ec2_memory_reserve_mib(mem)))
}
/// Why `cfg`'s reservations don't fit `instance_type`, or `None` when they
/// do (or the type is unknown, or `cfg` is Fargate). The one wording of the
/// fit rule: the start's refusal and Host settings both use it.
pub fn ec2_fit_problem(cfg: &AwsWorkspaceConfig, instance_type: &str) -> Option<String> {
match fits_instance(cfg, instance_type) {
Fit::TooBig { available_cpu, available_memory_mib } => Some(format!(
"This project reserves {} CPU units and {} MiB of memory, more than your EC2 \
instance type {instance_type} has for a workspace ({available_cpu} CPU units, \
{available_memory_mib} MiB).",
cfg.cpu, cfg.memory_mib
)),
Fit::Ok | Fit::UnknownType => None,
}
}
pub fn validate_aws_workspace(cfg: &AwsWorkspaceConfig) -> Result<(), String> {
match cfg.launch_type {
LaunchType::Fargate => {
let sizes = fargate_memory_sizes(cfg.cpu).ok_or_else(|| {
format!(
"{} is not a Fargate CPU size (256, 512, 1024, 2048, 4096, 8192 or 16384)",
cfg.cpu
)
})?;
if !sizes.contains(&cfg.memory_mib) {
return Err(format!(
"{} MiB of memory is not valid with {} CPU units: use {}",
cfg.memory_mib,
cfg.cpu,
describe_sizes(&sizes)
));
}
if !(21..=200).contains(&cfg.ephemeral_gib) {
return Err("Ephemeral storage must be 21 to 200 GiB".into());
}
}
LaunchType::Ec2 => {
// Reservations on the user's instance; the instance type is not on
// the project, so `fits_instance` does the fit check later.
if cfg.cpu == 0 || cfg.memory_mib == 0 {
return Err("An EC2 workspace needs a positive CPU and memory reservation".into());
}
if cfg.spot {
return Err("Spot is not available on EC2 yet".into());
}
}
}
if let Some(m) = cfg.idle_stop_minutes {
if !(5..=1440).contains(&m) {
@@ -1436,6 +1605,126 @@ mod tests {
assert!(p.is_aws());
}
fn ec2(cpu: u32, mem: u32) -> AwsWorkspaceConfig {
AwsWorkspaceConfig { launch_type: LaunchType::Ec2, cpu, memory_mib: mem, ..Default::default() }
}
#[test]
fn old_json_is_fargate_and_ec2_round_trips() {
let c: AwsWorkspaceConfig = serde_json::from_str("{}").unwrap();
assert_eq!(c.launch_type, LaunchType::Fargate);
let e: AwsWorkspaceConfig = serde_json::from_str(r#"{"launch_type":"ec2"}"#).unwrap();
assert_eq!(e.launch_type, LaunchType::Ec2);
assert!(serde_json::to_string(&e).unwrap().contains(r#""launch_type":"ec2""#));
}
#[test]
fn ec2_validation_shape() {
// Reservations need not be a Fargate pair.
assert!(validate_aws_workspace(&ec2(1500, 3000)).is_ok());
assert!(validate_aws_workspace(&ec2(0, 3000)).is_err());
assert!(validate_aws_workspace(&ec2(1024, 0)).is_err());
let mut s = ec2(1024, 2048);
s.spot = true;
assert!(validate_aws_workspace(&s).unwrap_err().contains("Spot"));
// Ephemeral size is ignored on EC2.
let mut e = ec2(1024, 2048);
e.ephemeral_gib = 5;
assert!(validate_aws_workspace(&e).is_ok());
// Fargate still rejects the non-pair.
assert!(validate_aws_workspace(&cfg(1500, 3000)).is_err());
}
#[test]
fn the_instance_type_list_is_the_table() {
let list = ec2_instance_types();
assert_eq!(list.len(), 30);
for t in &list {
assert_eq!(
ec2_instance_spec(&t.instance_type),
Some((t.vcpu * 1024, t.memory_mib)),
"{}",
t.instance_type
);
}
// And the other way: every type the spec table sizes is listed (a
// family added to one but not the other fails here).
let families = [
"m7i", "m7g", "m7a", "m6i", "c7i", "c7g", "c7a", "r7i", "r7g", "r7a", "t3", "t3a",
"t4g", "p5", "g6",
];
for family in families {
for size in ["medium", "large", "xlarge", "2xlarge", "4xlarge", "8xlarge", "metal"] {
let t = format!("{family}.{size}");
if ec2_instance_spec(&t).is_some() {
assert!(list.iter().any(|row| row.instance_type == t), "{t} missing");
}
}
}
let arch = |name: &str| list.iter().find(|t| t.instance_type == name).unwrap().arch;
assert_eq!(arch("m7i.large"), Arch::X86_64);
assert_eq!(arch("t3.xlarge"), Arch::X86_64);
assert_eq!(arch("m7g.large"), Arch::Arm64);
assert_eq!(arch("t4g.2xlarge"), Arch::Arm64);
assert!(list.iter().any(|t| t.instance_type == DEFAULT_EC2_INSTANCE_TYPE));
assert_eq!(arch(DEFAULT_EC2_INSTANCE_TYPE), Arch::X86_64);
let first = &list[0];
assert_eq!(
serde_json::to_value(first).unwrap(),
serde_json::json!({"instance_type": "m7i.large", "arch": "x86_64", "vcpu": 2, "memory_mib": 8192})
);
}
#[test]
fn instance_table_values() {
assert_eq!(ec2_instance_spec("m7i.large"), Some((2048, 8192)));
assert_eq!(ec2_instance_spec("m7g.4xlarge"), Some((16384, 65536)));
assert_eq!(ec2_instance_spec("c7i.xlarge"), Some((4096, 8192)));
assert_eq!(ec2_instance_spec("r7g.2xlarge"), Some((8192, 65536)));
assert_eq!(ec2_instance_spec("t3.large"), Some((2048, 8192)));
assert_eq!(ec2_instance_spec("t4g.2xlarge"), Some((8192, 32768)));
assert_eq!(ec2_instance_spec("t3.4xlarge"), None);
assert_eq!(ec2_instance_spec("p5.48xlarge"), None);
assert_eq!(ec2_instance_spec("m7i"), None);
}
#[test]
fn fit_checks_reserve_for_the_agent() {
// m7i.large: 2048 / 8192 less 128 CPU units and 10 % of the memory
// (819 MiB: ECS registers less than the nominal 8 GiB).
assert_eq!(fits_instance(&ec2(1920, 7373), "m7i.large"), Fit::Ok);
assert_eq!(
fits_instance(&ec2(1921, 4096), "m7i.large"),
Fit::TooBig { available_cpu: 1920, available_memory_mib: 7373 }
);
assert!(matches!(fits_instance(&ec2(1024, 7374), "m7i.large"), Fit::TooBig { .. }));
// The reserve is never under the agent's 256 MiB, and grows with
// the instance (r7i.4xlarge: 128 GiB nominal).
assert_eq!(ec2_memory_reserve_mib(2048), 256);
assert_eq!(ec2_memory_reserve_mib(131072), 13107);
assert_eq!(fits_instance(&ec2(1024, 2048), "z9.huge"), Fit::UnknownType);
// Fargate configs have no instance to fit.
assert_eq!(fits_instance(&cfg(16384, 122880), "m7i.large"), Fit::Ok);
}
#[test]
fn the_fit_problem_words_the_fit_rule_once() {
assert_eq!(ec2_available("m7i.large"), Some((1920, 7373)));
assert_eq!(ec2_available("t4g.large"), Some((1920, 7373)));
assert_eq!(ec2_available("z9.huge"), None);
assert_eq!(ec2_fit_problem(&ec2(1920, 7373), "m7i.large"), None);
assert_eq!(
ec2_fit_problem(&ec2(1024, 7374), "m7i.large").as_deref(),
Some(
"This project reserves 1024 CPU units and 7374 MiB of memory, more than your EC2 \
instance type m7i.large has for a workspace (1920 CPU units, 7373 MiB)."
)
);
// Unknown types and Fargate configs: no problem to report here.
assert_eq!(ec2_fit_problem(&ec2(99999, 99999), "z9.huge"), None);
assert_eq!(ec2_fit_problem(&cfg(16384, 122880), "m7i.large"), None);
}
#[test]
fn cpu_and_memory_must_be_a_fargate_pair() {
// Shared with the frontend's copy (`lib/awsWorkspace.test.ts`).
@@ -149,4 +149,11 @@ describe("PermissionModeControl", () => {
/Caution/i,
);
});
it("shows the sandbox OFF on an AWS project whatever the setting says", () => {
const aws = { ...baseProject, host_id: "aws-1", sandbox_mode_enabled: true, permission_mode: "bypass" as const };
render(<PermissionModeControl project={aws} onChange={vi.fn()} />);
expect(screen.getByTestId("sandbox-state")).toHaveTextContent(/Sandbox OFF/);
expect(screen.getByTestId("permission-mode-hint")).toHaveTextContent(/Caution/i);
});
});
@@ -1,5 +1,6 @@
import type { PermissionMode, Project } from "../../lib/types";
import SegmentedControl, { type Segment } from "../ui/SegmentedControl";
import { sandboxEffective } from "../../lib/awsWorkspace";
export const PERMISSION_MODES: Segment<PermissionMode>[] = [
{ value: "plan", label: "Plan", hint: "Claude proposes a plan and makes no changes." },
@@ -56,7 +57,7 @@ export default function PermissionModeControl({
disabledReason,
}: Props) {
const mode = effectivePermissionMode(project);
const sandboxOn = project.sandbox_mode_enabled;
const sandboxOn = sandboxEffective(project);
const uncontainedBypass = mode === "bypass" && !sandboxOn;
const segments = PERMISSION_MODES.map((segment) =>
@@ -4,10 +4,12 @@ import type { AwsWorkspaceConfig, HourlyEstimate } from "../../../lib/types";
import { defaultAwsWorkspace } from "../../../lib/awsWorkspace";
const awsCostEstimate =
vi.fn<(id: string, region: string, config: AwsWorkspaceConfig) => Promise<HourlyEstimate | null>>();
vi.fn<
(id: string, region: string, config: AwsWorkspaceConfig, instanceType: string | null) => Promise<HourlyEstimate | null>
>();
vi.mock("../../../lib/tauri-commands", () => ({
awsCostEstimate: (id: string, region: string, config: AwsWorkspaceConfig) =>
awsCostEstimate(id, region, config),
awsCostEstimate: (id: string, region: string, config: AwsWorkspaceConfig, instanceType: string | null) =>
awsCostEstimate(id, region, config, instanceType),
}));
import AwsCostEstimate, { ESTIMATE_DEBOUNCE_MS, formatHourly } from "./AwsCostEstimate";
@@ -43,12 +45,14 @@ describe("AwsCostEstimate", () => {
usd_per_hour: 0.05826,
note_spot: false,
efs_billed_separately: false,
ec2_instance: false,
root_volume_included: false,
});
const c = config();
render(<AwsCostEstimate connectionId="c1" region="us-west-2" config={c} />);
expect(screen.getByText("Estimating…")).toBeInTheDocument();
await settle();
expect(awsCostEstimate).toHaveBeenCalledWith("c1", "us-west-2", c);
expect(awsCostEstimate).toHaveBeenCalledWith("c1", "us-west-2", c, null);
expect(screen.getByText("≈ $0.058/hour")).toBeInTheDocument();
expect(screen.queryByText(/Spot/)).toBeNull();
expect(screen.queryByText(/EFS/)).toBeNull();
@@ -59,6 +63,8 @@ describe("AwsCostEstimate", () => {
usd_per_hour: 0.05,
note_spot: true,
efs_billed_separately: true,
ec2_instance: false,
root_volume_included: false,
});
render(
<AwsCostEstimate
@@ -96,7 +102,7 @@ describe("AwsCostEstimate", () => {
});
it("asks once per settled change of a cost field, not for unrelated edits", async () => {
awsCostEstimate.mockResolvedValue({ usd_per_hour: 0.1, note_spot: false, efs_billed_separately: false });
awsCostEstimate.mockResolvedValue({ usd_per_hour: 0.1, note_spot: false, efs_billed_separately: false, ec2_instance: false, root_volume_included: false });
const { rerender } = render(<AwsCostEstimate connectionId="c1" region="us-west-2" config={config()} />);
await settle();
expect(awsCostEstimate).toHaveBeenCalledTimes(1);
@@ -113,4 +119,41 @@ describe("AwsCostEstimate", () => {
expect(awsCostEstimate).toHaveBeenCalledTimes(2);
expect(awsCostEstimate.mock.calls[1][2].cpu).toBe(4096);
});
it("says when an EC2 estimate lacks the root volume's price", async () => {
awsCostEstimate.mockResolvedValue({ usd_per_hour: 0.1008, note_spot: false, efs_billed_separately: false, ec2_instance: true, root_volume_included: false });
render(<AwsCostEstimate connectionId="c1" region="us-west-2" config={config({ launch_type: "ec2" })} instanceType="m7i.large" />);
await settle();
expect(screen.getByText(/Its 50 GiB gp3 root volume \(about \$4 a month\) is not included/)).toBeInTheDocument();
expect(screen.getByText(/public IPv4 address adds/)).toBeInTheDocument();
});
it("prices an EC2 project as its instance type, and says instances scale to zero", async () => {
awsCostEstimate.mockResolvedValue({ usd_per_hour: 0.1008, note_spot: false, efs_billed_separately: true, ec2_instance: true, root_volume_included: true });
const c = config({ launch_type: "ec2", storage: "hybrid" });
const { rerender } = render(
<AwsCostEstimate connectionId="c1" region="us-west-2" config={c} instanceType="m7i.large" />,
);
await settle();
expect(awsCostEstimate).toHaveBeenCalledWith("c1", "us-west-2", c, "m7i.large");
expect(screen.getByText("≈ $0.10/hour")).toBeInTheDocument();
expect(screen.getByText(/Instances scale to zero when no workspace is running/)).toBeInTheDocument();
expect(screen.getByText(/EFS storage is billed separately/)).toBeInTheDocument();
expect(screen.getByText(/Includes its 50 GiB gp3 root volume/)).toBeInTheDocument();
expect(screen.getByText(/public IPv4 address adds \$0.005\/hour/)).toBeInTheDocument();
expect(screen.queryByText(/While the task runs/)).toBeNull();
// The reservations don't change an instance's price: no new call.
rerender(
<AwsCostEstimate connectionId="c1" region="us-west-2" config={{ ...c, cpu: 512, memory_mib: 1024 }} instanceType="m7i.large" />,
);
await settle();
expect(awsCostEstimate).toHaveBeenCalledTimes(1);
});
it("does not ask for an EC2 project whose instance type is unknown", async () => {
render(<AwsCostEstimate connectionId="c1" region="us-west-2" config={config({ launch_type: "ec2" })} instanceType={null} />);
await settle();
expect(awsCostEstimate).not.toHaveBeenCalled();
expect(screen.getByText("Estimate unavailable")).toBeInTheDocument();
});
});
@@ -15,6 +15,8 @@ interface Props {
connectionId: string | null;
region: string | null | undefined;
config: AwsWorkspaceConfig;
/** An EC2 project's instance type (the user's); null when unknown. */
instanceType?: string | null;
}
type State = { kind: "loading" } | { kind: "done"; estimate: HourlyEstimate | null };
@@ -25,21 +27,27 @@ type State = { kind: "loading" } | { kind: "done"; estimate: HourlyEstimate | nu
* network or no connection reads "Estimate unavailable" — never an error toast, and nothing waits
* on it. Re-asked only when a field that changes the price changes, after edits settle.
*/
export default function AwsCostEstimate({ connectionId, region, config }: Props) {
export default function AwsCostEstimate({ connectionId, region, config, instanceType = null }: Props) {
const [state, setState] = useState<State>({ kind: "loading" });
const { cpu, memory_mib, arch, ephemeral_gib, spot, storage } = config;
const { launch_type, cpu, memory_mib, arch, ephemeral_gib, spot, storage } = config;
const ec2 = launch_type === "ec2";
// Only what the backend prices or notes; the rest of the config is sent but changes nothing.
const key = JSON.stringify([connectionId, region, cpu, memory_mib, arch, ephemeral_gib, spot, storage]);
// An EC2 project is priced as its instance (E10): only the type and storage matter.
const key = JSON.stringify(
ec2
? [connectionId, region, launch_type, instanceType, storage]
: [connectionId, region, launch_type, cpu, memory_mib, arch, ephemeral_gib, spot, storage],
);
useEffect(() => {
if (!connectionId || !region) {
if (!connectionId || !region || (ec2 && !instanceType)) {
setState({ kind: "done", estimate: null });
return;
}
let cancelled = false;
setState({ kind: "loading" });
const timer = setTimeout(() => {
awsCostEstimate(connectionId, region, config)
awsCostEstimate(connectionId, region, config, ec2 ? instanceType : null)
.catch(() => null)
.then((estimate) => {
if (!cancelled) setState({ kind: "done", estimate });
@@ -65,7 +73,17 @@ export default function AwsCostEstimate({ connectionId, region, config }: Props)
}`
: "Estimate unavailable"}
</p>
{estimate && (
{estimate?.ec2_instance && (
<p className="text-[var(--text-secondary)]">
While your instance runs. Instances scale to zero when no workspace is running.
{estimate.root_volume_included
? " Includes its 50 GiB gp3 root volume."
: " Its 50 GiB gp3 root volume (about $4 a month) is not included."}
{" In an environment with public networking, the instance's public IPv4 address adds $0.005/hour."}
{estimate.efs_billed_separately && " EFS storage is billed separately."}
</p>
)}
{estimate && !estimate.ec2_instance && (
<p className="text-[var(--text-secondary)]">
While the task runs; nothing while it is stopped.
{estimate.efs_billed_separately && " EFS storage is billed separately."}
@@ -15,6 +15,7 @@ import ContainerMigrationBanner from "./ContainerMigrationBanner";
import AwsWorkspaceStatusCard from "./AwsWorkspaceStatusCard";
import AwsCostEstimate from "./AwsCostEstimate";
import { isAwsProject } from "../../../lib/awsWorkspace";
import { useEc2Capacity } from "../../../lib/ec2Capacity";
import type { ContainerMigration } from "../../../hooks/useContainerMigration";
import SaveIndicator from "../../ui/SaveIndicator";
import Button from "../../ui/Button";
@@ -53,6 +54,11 @@ export default function OverviewTab({
onOpenMigration,
}: Props) {
const [sessions, setSessions] = useState<ClaudeSession[]>([]);
// An EC2 project's estimate is its instance type's price (E10).
const ec2Capacity = useEc2Capacity(
project,
isAwsProject(project) && project.status === "running" && project.aws?.launch_type === "ec2",
);
const [tasks, setTasks] = useState<ScheduledTask[]>([]);
const [notificationCount, setNotificationCount] = useState(0);
const running = project.status === "running";
@@ -152,6 +158,7 @@ export default function OverviewTab({
connectionId={project.host_id}
region={project.aws_placement?.region}
config={project.aws}
instanceType={ec2Capacity?.state === "ec2" ? ec2Capacity.compute.instance_type : null}
/>
</section>
)}
@@ -9,11 +9,16 @@ import { awsProject, localProject } from "../../../../test/awsFixtures";
import type { AppSettings, AuthBridgeStatus, Project } from "../../../../lib/types";
import { useAppState } from "../../../../store/appState";
import { clearEnvironmentMaxHoursCache } from "../../../../lib/environmentMaxHours";
import { clearEc2CapacityCache } from "../../../../lib/ec2Capacity";
import type { Ec2Capacity } from "../../../../lib/types";
const OFF_BRIDGE: AuthBridgeStatus = { enabled: false, active_ports: [], conflicts: [] };
const removeAwsResources = vi.fn();
const awsCostEstimate = vi.fn(async (..._args: unknown[]) => null as unknown);
const awsEnvironmentMaxHours = vi.fn(async (_id: string) => 0 as number | null);
const NO_EC2: Ec2Capacity = { state: "none" };
const awsEc2Capacity = vi.fn(async (_id: string): Promise<Ec2Capacity> => NO_EC2);
const ec2FitProblem = vi.fn(async (..._args: unknown[]) => null as string | null);
vi.mock("../../../../lib/tauri-commands", () => ({
getAuthBridgeStatus: vi.fn(async () => OFF_BRIDGE),
@@ -22,6 +27,8 @@ vi.mock("../../../../lib/tauri-commands", () => ({
removeAwsResources: (id: string, discard: boolean) => removeAwsResources(id, discard),
awsCostEstimate: (...args: unknown[]) => awsCostEstimate(...args),
awsEnvironmentMaxHours: (id: string) => awsEnvironmentMaxHours(id),
awsEc2Capacity: (id: string) => awsEc2Capacity(id),
ec2FitProblem: (...args: unknown[]) => ec2FitProblem(...args),
}));
vi.mock("@tauri-apps/api/event", () => ({ listen: vi.fn(async () => () => {}) }));
vi.mock("@tauri-apps/plugin-dialog", () => ({ open: vi.fn(async () => null) }));
@@ -31,6 +38,9 @@ const save = vi.fn().mockResolvedValue(true);
beforeEach(() => {
vi.clearAllMocks();
clearEnvironmentMaxHoursCache();
clearEc2CapacityCache();
awsEc2Capacity.mockImplementation(async () => NO_EC2);
ec2FitProblem.mockImplementation(async () => null);
});
function toggle(name: string) {
@@ -244,7 +254,7 @@ describe("HostSection", () => {
await waitFor(() =>
expect(within(card).getByText(/≈ \$0\.050\/hour \(on-demand; Spot is usually up to 70% cheaper\)/)).toBeInTheDocument(),
);
expect(awsCostEstimate).toHaveBeenCalledWith("c1", "eu-west-1", p.aws);
expect(awsCostEstimate).toHaveBeenCalledWith("c1", "eu-west-1", p.aws, null);
expect(within(card).getByText(/EFS storage is billed separately/)).toBeInTheDocument();
useAppState.setState({ appSettings: null });
});
@@ -595,3 +605,272 @@ describe("HostSection", () => {
expect(screen.queryByLabelText("Repository 1 URL")).toBeNull();
});
});
// ---- phase 7: EC2 projects ----
const M7I: Ec2Capacity = {
state: "ec2",
compute: {
instance_type: "m7i.large",
max_instances: 1,
user_tag: "alice",
cluster_arn: "arn:aws:ecs:eu-west-1:123456789012:cluster/triple-c-dev--u-alice",
capacity_provider: "triple-c-dev--u-alice",
},
available_cpu: 1920,
available_memory_mib: 7373,
};
const ec2Project = (over: Partial<Project> = {}): Project => {
const p = awsProject(over);
return { ...p, aws: { ...p.aws!, launch_type: "ec2", cpu: 1024, memory_mib: 4096 } };
};
describe("HostSection — launch type (EC2)", () => {
const renderHost = (p: Project = awsProject(), disabled = false) =>
render(<HostSection project={p} save={save} disabled={disabled} />);
const ec2Segment = () => screen.getByRole("radio", { name: /^EC2/ });
it("offers EC2 with the user's instance type when their onboarding has EC2 capacity", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
renderHost();
await waitFor(() => expect(ec2Segment()).toHaveTextContent("EC2 (your instance: m7i.large)"));
expect(ec2Segment()).toBeEnabled();
expect(screen.queryByText(/Ask an admin/)).toBeNull();
});
it("disables EC2 with the reason when the user has no EC2 capacity, or it can't be read", async () => {
renderHost();
await waitFor(() =>
expect(screen.getByText(/Ask an admin to add EC2 capacity to your onboarding/)).toBeInTheDocument(),
);
expect(ec2Segment()).toBeDisabled();
expect(screen.getByRole("radio", { name: "Fargate" })).toBeEnabled();
cleanup();
clearEc2CapacityCache();
awsEc2Capacity.mockImplementation(async () => ({ state: "unusable", reason: "schema is not 1" }));
renderHost();
await waitFor(() => expect(screen.getByText(/can't be used \(schema is not 1\)/)).toBeInTheDocument());
expect(ec2Segment()).toBeDisabled();
cleanup();
clearEc2CapacityCache();
awsEc2Capacity.mockImplementation(async () => ({ state: "unknown" }));
renderHost();
await waitFor(() => expect(screen.getByText(/Could not read your EC2 capacity/)).toBeInTheDocument());
expect(ec2Segment()).toBeDisabled();
});
it("switches to EC2 through launchTypePatch: sidecar and Spot off, sizes fitted", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
const p = awsProject();
p.aws = { ...p.aws!, cpu: 4096, memory_mib: 16384, spot: true, gateway_sidecar: true };
renderHost(p);
await waitFor(() => expect(ec2Segment()).toBeEnabled());
fireEvent.click(ec2Segment());
await waitFor(() => expect(save).toHaveBeenCalledTimes(1));
expect(save.mock.calls[0][0].aws).toMatchObject({
launch_type: "ec2",
gateway_sidecar: false,
spot: false,
cpu: 1920,
memory_mib: 7373,
});
expect(save.mock.calls[0][0].allow_docker_access).toBeUndefined();
});
it("switches back to Fargate turning Docker access and VPN off, with a Fargate size", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
const p = ec2Project({ allow_docker_access: true, vpn_support_enabled: true });
p.aws = { ...p.aws!, cpu: 1920, memory_mib: 7373 };
renderHost(p);
fireEvent.click(screen.getByRole("radio", { name: "Fargate" }));
await waitFor(() => expect(save).toHaveBeenCalledTimes(1));
const patch = save.mock.calls[0][0];
expect(patch).toMatchObject({ allow_docker_access: false, vpn_support_enabled: false });
expect(patch.aws).toMatchObject({ launch_type: "fargate", cpu: 1024, memory_mib: 7168 });
});
it("an EC2 project keeps its launch type selectable back to Fargate even without capacity, and says why it can't start", async () => {
renderHost(ec2Project());
await waitFor(() => expect(screen.getByText(/can't start on EC2 now/)).toBeInTheDocument());
expect(ec2Segment()).toHaveAttribute("aria-checked", "true");
expect(screen.getByRole("radio", { name: "Fargate" })).toBeEnabled();
});
it("on EC2: reservation inputs with the instance's limits, no ephemeral size, Spot or architecture, and the notes", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
renderHost(ec2Project());
await waitFor(() => expect(screen.getByText(/Up to 1920 on your m7i.large/)).toBeInTheDocument());
expect(screen.getByText(/Up to 7373 MiB on your m7i.large/)).toBeInTheDocument();
expect(screen.getByLabelText("CPU reservation (units)")).toHaveValue("1024");
expect(screen.getByLabelText("Memory reservation (MiB)")).toHaveValue("4096");
expect(screen.queryByLabelText("Ephemeral storage (GiB)")).toBeNull();
expect(screen.queryByRole("switch", { name: "Spot" })).toBeNull();
expect(screen.queryByRole("radiogroup", { name: "Architecture" })).toBeNull();
const notes = screen.getByRole("note", { name: "On EC2" });
expect(notes).toHaveTextContent(/first start after idle boots an instance — about 3 minutes/);
expect(notes).toHaveTextContent(/root on your own instance/);
expect(notes).toHaveTextContent(/gateway sidecar is not available on EC2/);
// Fargate shows none of this.
cleanup();
renderHost(awsProject());
expect(screen.queryByRole("note", { name: "On EC2" })).toBeNull();
expect(screen.getByLabelText("Ephemeral storage (GiB)")).toBeInTheDocument();
expect(screen.getByRole("switch", { name: "Spot" })).toBeInTheDocument();
});
it("checks the reservations with the backend's fit rule before saving", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
ec2FitProblem.mockImplementation(async (config) =>
(config as { memory_mib: number }).memory_mib === 8000
? "This project reserves 1024 CPU units and 8000 MiB of memory, more than …"
: null,
);
renderHost(ec2Project());
await waitFor(() => expect(screen.getByText(/Up to 1920/)).toBeInTheDocument());
// The load-time check found the stored 4096 MiB fits: no alert.
await waitFor(() => expect(ec2FitProblem).toHaveBeenCalledTimes(1));
expect(screen.queryByRole("alert")).toBeNull();
const mem = screen.getByLabelText("Memory reservation (MiB)");
fireEvent.change(mem, { target: { value: "8000" } });
fireEvent.blur(mem);
await waitFor(() => expect(screen.getByRole("alert")).toHaveTextContent(/more than/));
expect(ec2FitProblem).toHaveBeenLastCalledWith(expect.objectContaining({ cpu: 1024, memory_mib: 8000 }), "m7i.large");
expect(save).not.toHaveBeenCalled();
fireEvent.change(mem, { target: { value: "6000" } });
fireEvent.blur(mem);
await waitFor(() => expect(save).toHaveBeenCalledTimes(1));
expect(save.mock.calls[0][0].aws).toMatchObject({ cpu: 1024, memory_mib: 6000 });
const cpu = screen.getByLabelText("CPU reservation (units)");
fireEvent.change(cpu, { target: { value: "0" } });
fireEvent.blur(cpu);
await waitFor(() => expect(screen.getByRole("alert")).toHaveTextContent(/positive whole numbers/));
});
it("warns on load when the saved reservations no longer fit the instance type, once", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
ec2FitProblem.mockImplementation(async () => "This project reserves 4096 CPU units …, more than an m7i.large");
const p = ec2Project();
p.aws = { ...p.aws!, cpu: 4096 };
const { rerender } = renderHost(p);
await waitFor(() =>
expect(screen.getByRole("alert")).toHaveTextContent(/saved reservations no longer fit: .*more than an m7i.large/),
);
expect(ec2FitProblem).toHaveBeenCalledWith(expect.objectContaining({ cpu: 4096 }), "m7i.large");
rerender(<HostSection project={{ ...p }} save={save} disabled={false} />);
await act(async () => {});
expect(ec2FitProblem).toHaveBeenCalledTimes(1);
expect(save).not.toHaveBeenCalled();
cleanup();
// Fargate projects, and EC2 without a known instance type, aren't checked.
ec2FitProblem.mockClear();
renderHost(awsProject());
await waitFor(() => expect(ec2Segment()).toBeEnabled());
cleanup();
clearEc2CapacityCache();
awsEc2Capacity.mockImplementation(async () => NO_EC2);
renderHost(ec2Project());
await waitFor(() => expect(screen.getByText(/can't start on EC2 now/)).toBeInTheDocument());
expect(ec2FitProblem).not.toHaveBeenCalled();
});
it("says what a launch-type switch changed besides the type", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
const p = awsProject();
p.aws = { ...p.aws!, cpu: 4096, memory_mib: 16384, spot: true, gateway_sidecar: true };
renderHost(p);
await waitFor(() => expect(ec2Segment()).toBeEnabled());
fireEvent.click(ec2Segment());
expect(await screen.findByRole("status")).toHaveTextContent(
"Switched to EC2: CPU 4096 → 1920 units, memory 16384 → 7373 MiB, Spot turned off, LLM gateway sidecar turned off.",
);
cleanup();
const back = ec2Project({ allow_docker_access: true, vpn_support_enabled: true });
back.aws = { ...back.aws!, cpu: 1920, memory_mib: 7373 };
renderHost(back);
fireEvent.click(screen.getByRole("radio", { name: "Fargate" }));
expect(await screen.findByRole("status")).toHaveTextContent(
"Switched to Fargate: CPU 1920 → 1024 units, memory 7373 → 7168 MiB, Docker access turned off, VPN support turned off.",
);
});
const switchedProject = () => {
const p = awsProject();
p.aws = { ...p.aws!, cpu: 4096, memory_mib: 16384, spot: true, gateway_sidecar: true };
return p;
};
it("shows no switch note when the save was refused", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
renderHost(switchedProject());
await waitFor(() => expect(ec2Segment()).toBeEnabled());
save.mockResolvedValueOnce(false);
fireEvent.click(ec2Segment());
await waitFor(() => expect(save).toHaveBeenCalled());
await new Promise((r) => setTimeout(r, 0));
expect(screen.queryByText(/Switched to/)).toBeNull();
});
it("shows no switch note when the save fails", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
renderHost(switchedProject());
await waitFor(() => expect(ec2Segment()).toBeEnabled());
const unhandled = vi.fn();
process.on("unhandledRejection", unhandled);
save.mockRejectedValueOnce(new Error("boom"));
fireEvent.click(ec2Segment());
await waitFor(() => expect(save).toHaveBeenCalled());
await new Promise((r) => setTimeout(r, 0));
process.off("unhandledRejection", unhandled);
expect(screen.queryByText(/Switched to/)).toBeNull();
});
it("drops the switch note when another setting is saved afterwards", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
renderHost(switchedProject());
await waitFor(() => expect(ec2Segment()).toBeEnabled());
fireEvent.click(ec2Segment());
expect(await screen.findByText(/Switched to EC2/)).toBeInTheDocument();
fireEvent.click(screen.getByRole("radio", { name: "Fargate" }));
await waitFor(() => expect(screen.queryByText(/Switched to EC2/)).toBeNull());
});
it("prices an EC2 project as the user's instance type", async () => {
awsEc2Capacity.mockImplementation(async () => M7I);
useAppState.setState({
appSettings: { hosts: [{ id: "c1", name: "lab" }] } as unknown as AppSettings,
});
const p = ec2Project();
renderHost(p);
await waitFor(() => expect(awsCostEstimate).toHaveBeenCalledWith("c1", "eu-west-1", p.aws, "m7i.large"));
expect(screen.getByText(/price in this region of your EC2 instance \(m7i.large\)/)).toBeInTheDocument();
useAppState.setState({ appSettings: null });
});
});
describe("RuntimeSection on EC2", () => {
it("enables Docker access and VPN, keeps the sandbox off, and says Docker access is root on the instance", () => {
render(<RuntimeSection project={ec2Project({ sandbox_mode_enabled: true })} save={save} disabled={false} />);
for (const name of ["Allow container spawning", "VPN support"]) {
expect(toggle(name)).toBeEnabled();
}
// E8, measured live: bubblewrap can't run in an ECS task on EC2 either.
expect(toggle("Sandbox mode")).toBeDisabled();
expect(toggle("Sandbox mode")).toHaveAttribute("aria-checked", "false");
expect(screen.getByText(/root on your own instance/)).toBeInTheDocument();
expect(screen.getByText(/Off on AWS for now/)).toBeInTheDocument();
// The VPN hint speaks of an ECS task on the user's instance, not of
// recreating a container or its volumes.
const vpn = screen.getByText(/VPN client \(PIA, WireGuard, OpenVPN\)/);
expect(vpn).toHaveTextContent(/an ECS task on your own EC2 instance/);
expect(vpn).toHaveTextContent(/next start/);
expect(vpn).not.toHaveTextContent(/recreates the container|volumes/);
});
});
@@ -1,7 +1,15 @@
import { useEffect, useRef, useState } from "react";
import type { AwsWorkspaceConfig, HostConnection, Project, RepoSource } from "../../../../lib/types";
import type {
AwsWorkspaceConfig,
Ec2Capacity,
HostConnection,
LaunchType,
Project,
RepoSource,
} from "../../../../lib/types";
import * as commands from "../../../../lib/tauri-commands";
import { environmentMaxHours } from "../../../../lib/environmentMaxHours";
import { useEc2Capacity } from "../../../../lib/ec2Capacity";
import { useAppState } from "../../../../store/appState";
import {
COPY_AT_CONNECT_GATEWAY_NOTE,
@@ -16,6 +24,13 @@ import {
setupScriptProblem,
EFS_HINT,
EPHEMERAL_WARNING,
EC2_DOCKER_ACCESS_NOTE,
EC2_NO_CAPACITY,
EC2_NO_SIDECAR,
EC2_SCALE_TO_ZERO_NOTE,
launchTypeChanges,
launchTypePatch,
type Ec2Limits,
FARGATE_CPUS,
defaultAwsWorkspace,
formatCpu,
@@ -43,6 +58,21 @@ interface Props {
const DEFAULT_IDLE_MINUTES = 60;
const STILL_RUNNING = "The task is still running — stop it first.";
/** Why EC2 can't be chosen, from what the user's record says (null: it can). */
function ec2Unavailable(capacity: Ec2Capacity | null): string | null {
if (capacity === null) return "Checking your EC2 capacity in this environment…";
switch (capacity.state) {
case "ec2":
return null;
case "none":
return `You have no EC2 capacity in this environment. ${EC2_NO_CAPACITY}.`;
case "unusable":
return `Your EC2 capacity record can't be used (${capacity.reason}). ${EC2_NO_CAPACITY}.`;
case "unknown":
return "Could not read your EC2 capacity in this environment; EC2 is offered once it can be read.";
}
}
/**
* Where an AWS project runs and what its Fargate task looks like. Only shown
* for AWS projects; a local project's host is chosen when it is added.
@@ -55,6 +85,7 @@ export default function HostSection({ project, save, disabled }: Props) {
const hosts = useAppState((s) => s.appSettings?.hosts ?? NO_HOSTS);
const connection = hosts.find((h) => h.id === project.host_id);
const [cpuText, setCpuText] = useState(String(aws.cpu));
const [memory, setMemory] = useState(String(aws.memory_mib));
const [memoryProblem, setMemoryProblem] = useState<string | null>(null);
const [ephemeral, setEphemeral] = useState(String(aws.ephemeral_gib));
@@ -66,12 +97,47 @@ export default function HostSection({ project, save, disabled }: Props) {
const [setupProblem, setSetupProblem] = useState<string | null>(null);
// The environment's advisory lifetime cap (Q8), from its record.
const [maxHours, setMaxHours] = useState<number | null>(null);
// What the last launch-type switch changed besides the type (M6).
const [switchNote, setSwitchNote] = useState<string | null>(null);
useEffect(() => {
setCpuText(String(aws.cpu));
setMemory(String(aws.memory_mib));
setEphemeral(String(aws.ephemeral_gib));
setIdle(String(aws.idle_stop_minutes ?? DEFAULT_IDLE_MINUTES));
}, [aws.memory_mib, aws.ephemeral_gib, aws.idle_stop_minutes]);
}, [aws.cpu, aws.memory_mib, aws.ephemeral_gib, aws.idle_stop_minutes]);
// Phase 7: the user's own EC2 capacity in this environment (their record),
// which decides whether EC2 can be chosen and what it can reserve.
const capacity = useEc2Capacity(project);
const ec2 = aws.launch_type === "ec2";
const compute = capacity?.state === "ec2" ? capacity.compute : null;
const limits: Ec2Limits | null =
capacity?.state === "ec2" && capacity.available_cpu !== null && capacity.available_memory_mib !== null
? { cpu: capacity.available_cpu, memory_mib: capacity.available_memory_mib }
: null;
const ec2Reason = ec2Unavailable(capacity);
// Once the instance type is known, check the stored reservations against
// it (Rust's fit rule): an instance type changed at onboarding, or a
// record that came back after a save, can leave them too big.
const checkedFit = useRef<string | null>(null);
useEffect(() => {
if (!ec2 || !compute) return;
const key = `${project.id}:${compute.instance_type}`;
if (checkedFit.current === key) return;
checkedFit.current = key;
let cancelled = false;
commands
.ec2FitProblem(latest.current.value, compute.instance_type)
.then((problem) => {
if (!cancelled && problem) setMemoryProblem(`The saved reservations no longer fit: ${problem}`);
})
.catch(() => {}); // The start checks again.
return () => {
cancelled = true;
};
}, [ec2, compute, project.id]);
useEffect(() => {
setSetupScript(aws.setup_script ?? "");
@@ -104,7 +170,11 @@ export default function HostSection({ project, save, disabled }: Props) {
if (latest.current.base !== project.aws && latest.current.inFlight === 0) {
latest.current = { base: project.aws, value: aws, inFlight: 0 };
}
const saveAws = async (patch: Partial<AwsWorkspaceConfig>) => {
// Bumped by every save: a launch-type note only shows if no later save began.
const saveSeq = useRef(0);
const saveAws = async (patch: Partial<AwsWorkspaceConfig>, also: Partial<Project> = {}) => {
saveSeq.current += 1;
setSwitchNote(null);
const l = latest.current;
const next = { ...l.value, ...patch };
l.value = next;
@@ -114,7 +184,7 @@ export default function HostSection({ project, save, disabled }: Props) {
if (l.value === next) l.value = l.base ?? defaultAwsWorkspace();
};
try {
const ok = await save({ aws: next });
const ok = await save({ ...also, aws: next });
if (!ok) refused();
return ok;
} catch (e) {
@@ -141,6 +211,50 @@ export default function HostSection({ project, save, disabled }: Props) {
// Not `disabled` (which also allows Error): on AWS, Error can mean a live task.
const stopped = project.status === "stopped";
// The launch-type switch: `launchTypePatch` keeps each side to what Rust
// accepts there (Docker/VPN, the sidecar, Spot, the sizes).
const switchLaunchType = (launch_type: LaunchType) => {
if (launch_type === latest.current.value.launch_type) return;
setMemoryProblem(null);
const before = { ...project, aws: latest.current.value };
const patch = launchTypePatch(before, launch_type, limits);
const note = launchTypeChanges(before, patch);
const { aws: next, ...also } = patch;
// Shown only once the save went through, and dropped by any later save.
const pending = saveAws(next!, also);
const seq = saveSeq.current;
void pending.then((ok) => {
if (ok && saveSeq.current === seq) setSwitchNote(note);
});
};
// EC2 reservations: positive whole numbers that fit the user's instance
// type — checked by Rust's own fit rule (`ec2_fit_problem`), so this never
// disagrees with the start.
const commitEc2Sizing = async () => {
const cpu = Number(cpuText);
const memory_mib = Number(memory);
const whole = (n: number) => Number.isInteger(n) && n > 0;
if (!whole(cpu) || !whole(memory_mib)) {
setMemoryProblem("The CPU and memory reservations must be positive whole numbers");
return;
}
if (compute) {
let problem: string | null = null;
try {
problem = await commands.ec2FitProblem({ ...aws, cpu, memory_mib }, compute.instance_type);
} catch {
problem = null; // The start checks again.
}
if (problem) {
setMemoryProblem(problem);
return;
}
}
setMemoryProblem(null);
if (cpu !== aws.cpu || memory_mib !== aws.memory_mib) void saveAws({ cpu, memory_mib });
};
const commitMemory = () => {
const value = Number(memory);
const problem = memoryError(aws.cpu, value);
@@ -176,7 +290,11 @@ export default function HostSection({ project, save, disabled }: Props) {
return (
<ConfigGroup
title="Host"
description="This project runs as an AWS Fargate task. Changes apply on its next start."
description={
ec2
? "This project runs on your own EC2 instances, as an ECS task. Changes apply on its next start."
: "This project runs as an AWS Fargate task. Changes apply on its next start."
}
>
<dl className="grid grid-cols-[auto_1fr] gap-x-4 gap-y-1 text-[13px]">
<dt className="text-[var(--text-secondary)]">Connection</dt>
@@ -200,79 +318,173 @@ export default function HostSection({ project, save, disabled }: Props) {
</p>
)}
<div className="flex flex-wrap gap-4">
<Field label="CPU">
{(id) => (
<select
id={id}
value={aws.cpu}
disabled={disabled}
className={selectClass}
onChange={(e) => {
const cpu = Number(e.target.value);
const memory_mib = memoryError(cpu, aws.memory_mib)
? nearestMemory(cpu, aws.memory_mib)
: aws.memory_mib;
setMemoryProblem(null);
void saveAws({ cpu, memory_mib });
}}
>
{FARGATE_CPUS.filter(
// The gateway sidecar needs 1 vCPU / 2 GiB (Rust validates it too).
(c) => !aws.gateway_sidecar || c >= GATEWAY_SIDECAR_MIN_CPU,
).map((c) => (
<option key={c} value={c}>
{formatCpu(c)}
</option>
))}
</select>
)}
</Field>
<Field
label="Memory (MiB)"
hint={(() => {
const opts = memoryOptions(aws.cpu);
return `${formatMemory(opts[0])} to ${formatMemory(opts[opts.length - 1])} with ${formatCpu(aws.cpu)}.`;
})()}
>
{(id) => (
<>
<div>
<span className="block text-[13px] font-medium text-[var(--text-primary)] mb-1">
Launch type
</span>
<SegmentedControl
label="Launch type"
value={aws.launch_type}
disabled={disabled}
onChange={switchLaunchType}
segments={[
{ value: "fargate", label: "Fargate" },
{
value: "ec2",
label: compute ? `EC2 (your instance: ${compute.instance_type})` : "EC2",
disabled: !compute && !ec2,
},
]}
/>
{ec2Reason && (
<p
className={`mt-1 text-xs ${ec2 && capacity !== null && capacity.state !== "unknown" ? "text-[var(--warning)]" : "text-[var(--text-secondary)]"}`}
>
{ec2 && capacity !== null && capacity.state !== "unknown"
? `This project can't start on EC2 now: ${ec2Reason} Or switch it to Fargate.`
: ec2Reason}
</p>
)}
{switchNote && (
<p role="status" className="mt-1 text-xs text-[var(--text-secondary)]">
{switchNote}
</p>
)}
{ec2 && (
<ul
role="note"
aria-label="On EC2"
className="mt-2 px-2 py-1.5 text-xs text-[var(--text-secondary)] bg-[var(--bg-secondary)] border border-[var(--border-color)] rounded-[var(--radius-control)] list-disc list-inside space-y-0.5"
>
<li>{EC2_SCALE_TO_ZERO_NOTE}</li>
<li>{EC2_DOCKER_ACCESS_NOTE}</li>
<li>{EC2_NO_SIDECAR}.</li>
</ul>
)}
</div>
{ec2 ? (
<div className="flex flex-wrap gap-4">
<Field
label="CPU reservation (units)"
hint={
limits && compute
? `1024 units = 1 vCPU. Up to ${limits.cpu} on your ${compute.instance_type}.`
: "1024 units = 1 vCPU. It must fit your instance type."
}
>
{(id) => (
<input
id={id}
inputMode="numeric"
value={cpuText}
disabled={disabled}
aria-invalid={memoryProblem ? true : undefined}
onChange={(e) => setCpuText(e.target.value)}
onBlur={() => void commitEc2Sizing()}
onKeyDown={(e) => e.key === "Enter" && void commitEc2Sizing()}
className={`w-32 ${inputClass}`}
/>
)}
</Field>
<Field
label="Memory reservation (MiB)"
hint={
limits && compute
? `Up to ${limits.memory_mib} MiB on your ${compute.instance_type} (a tenth is kept for the instance).`
: "It must fit your instance type."
}
>
{(id) => (
<input
id={id}
inputMode="numeric"
value={memory}
disabled={disabled}
list={`${id}-options`}
aria-invalid={memoryProblem ? true : undefined}
onChange={(e) => setMemory(e.target.value)}
onBlur={commitMemory}
onKeyDown={(e) => e.key === "Enter" && commitMemory()}
onBlur={() => void commitEc2Sizing()}
onKeyDown={(e) => e.key === "Enter" && void commitEc2Sizing()}
className={`w-32 ${inputClass}`}
/>
<datalist id={`${id}-options`}>
{memoryOptions(aws.cpu).map((m) => (
<option key={m} value={m} />
))}
</datalist>
</>
)}
</Field>
<div>
<span className="block text-[13px] font-medium text-[var(--text-primary)] mb-1">
Architecture
</span>
<SegmentedControl
label="Architecture"
value={aws.arch}
disabled={disabled}
onChange={(arch) => void saveAws({ arch })}
segments={[
{ value: "x86_64", label: "x86_64" },
{ value: "arm64", label: "ARM64 (Graviton)" },
]}
/>
)}
</Field>
</div>
</div>
) : (
<div className="flex flex-wrap gap-4">
<Field label="CPU">
{(id) => (
<select
id={id}
value={aws.cpu}
disabled={disabled}
className={selectClass}
onChange={(e) => {
const cpu = Number(e.target.value);
const memory_mib = memoryError(cpu, aws.memory_mib)
? nearestMemory(cpu, aws.memory_mib)
: aws.memory_mib;
setMemoryProblem(null);
void saveAws({ cpu, memory_mib });
}}
>
{FARGATE_CPUS.filter(
// The gateway sidecar needs 1 vCPU / 2 GiB (Rust validates it too).
(c) => !aws.gateway_sidecar || c >= GATEWAY_SIDECAR_MIN_CPU,
).map((c) => (
<option key={c} value={c}>
{formatCpu(c)}
</option>
))}
</select>
)}
</Field>
<Field
label="Memory (MiB)"
hint={(() => {
const opts = memoryOptions(aws.cpu);
return `${formatMemory(opts[0])} to ${formatMemory(opts[opts.length - 1])} with ${formatCpu(aws.cpu)}.`;
})()}
>
{(id) => (
<>
<input
id={id}
inputMode="numeric"
value={memory}
disabled={disabled}
list={`${id}-options`}
aria-invalid={memoryProblem ? true : undefined}
onChange={(e) => setMemory(e.target.value)}
onBlur={commitMemory}
onKeyDown={(e) => e.key === "Enter" && commitMemory()}
className={`w-32 ${inputClass}`}
/>
<datalist id={`${id}-options`}>
{memoryOptions(aws.cpu).map((m) => (
<option key={m} value={m} />
))}
</datalist>
</>
)}
</Field>
<div>
<span className="block text-[13px] font-medium text-[var(--text-primary)] mb-1">
Architecture
</span>
<SegmentedControl
label="Architecture"
value={aws.arch}
disabled={disabled}
onChange={(arch) => void saveAws({ arch })}
segments={[
{ value: "x86_64", label: "x86_64" },
{ value: "arm64", label: "ARM64 (Graviton)" },
]}
/>
</div>
</div>
)}
{memoryProblem && (
<p role="alert" className="text-xs text-[var(--error)]">
{memoryProblem}
@@ -328,60 +540,66 @@ export default function HostSection({ project, save, disabled }: Props) {
)}
</div>
<Field
label="Ephemeral storage (GiB)"
hint={
aws.storage === "efs"
? "The task's local disk: 21 to 200 GiB. With EFS storage, /workspace is on EFS instead."
: "The task's local disk: 21 to 200 GiB. Repositories and builds live here."
}
>
{(id) => (
<input
id={id}
inputMode="numeric"
value={ephemeral}
disabled={disabled}
aria-invalid={ephemeralProblem ? true : undefined}
onChange={(e) => setEphemeral(e.target.value)}
onBlur={commitEphemeral}
className={`w-24 ${inputClass}`}
/>
)}
</Field>
{ephemeralProblem && (
<p role="alert" className="text-xs text-[var(--error)]">
{ephemeralProblem}
</p>
)}
{!ec2 && (
<>
<Field
label="Ephemeral storage (GiB)"
hint={
aws.storage === "efs"
? "The task's local disk: 21 to 200 GiB. With EFS storage, /workspace is on EFS instead."
: "The task's local disk: 21 to 200 GiB. Repositories and builds live here."
}
>
{(id) => (
<input
id={id}
inputMode="numeric"
value={ephemeral}
disabled={disabled}
aria-invalid={ephemeralProblem ? true : undefined}
onChange={(e) => setEphemeral(e.target.value)}
onBlur={commitEphemeral}
className={`w-24 ${inputClass}`}
/>
)}
</Field>
{ephemeralProblem && (
<p role="alert" className="text-xs text-[var(--error)]">
{ephemeralProblem}
</p>
)}
<SwitchRow
label="Spot"
hint={
aws.wip_push_on_stop
? "Runs on spare capacity at up to 70% off. AWS may reclaim it with two minutes' notice; the work-in-progress push below still runs."
: "Runs on spare capacity at up to 70% off. AWS may reclaim it with two minutes' notice, and with the work-in-progress push below off, unpushed work is lost."
}
control={
<Toggle
<SwitchRow
label="Spot"
checked={aws.spot}
disabled={disabled}
onChange={(spot) => void saveAws({ spot })}
hint={
aws.wip_push_on_stop
? "Runs on spare capacity at up to 70% off. AWS may reclaim it with two minutes' notice; the work-in-progress push below still runs."
: "Runs on spare capacity at up to 70% off. AWS may reclaim it with two minutes' notice, and with the work-in-progress push below off, unpushed work is lost."
}
control={
<Toggle
label="Spot"
checked={aws.spot}
disabled={disabled}
onChange={(spot) => void saveAws({ spot })}
/>
}
/>
}
/>
</>
)}
<section aria-label="Estimated cost">
<h3 className="text-[13px] font-medium text-[var(--text-primary)]">Estimated cost</h3>
<p className="mt-0.5 mb-1 text-xs text-[var(--text-secondary)] leading-snug">
The on-demand Fargate price in this region for the CPU, memory, architecture and disk
above.
{ec2
? `The on-demand price in this region of your EC2 instance${compute ? ` (${compute.instance_type})` : ""}, whatever this project reserves of it.`
: "The on-demand Fargate price in this region for the CPU, memory, architecture and disk above."}
</p>
<AwsCostEstimate
connectionId={connection ? connection.id : null}
region={project.aws_placement?.region}
config={aws}
instanceType={compute?.instance_type ?? null}
/>
</section>
@@ -2,6 +2,8 @@ import { describe, it, expect, vi, beforeEach } from "vitest";
import { render, screen, fireEvent, waitFor } from "@testing-library/react";
import RuntimeSection from "./RuntimeSection";
import type { AuthBridgeStatus, Project } from "../../../../lib/types";
import { launchTypePatch } from "../../../../lib/awsWorkspace";
import { awsProject } from "../../../../test/awsFixtures";
// The auth-bridge row owns its own IPC — see `AuthBridgeRow.tsx` for why it
// does not go through `save`.
@@ -181,3 +183,37 @@ describe("RuntimeSection — auth bridge toggle", () => {
expect(save).not.toHaveBeenCalled();
});
});
describe("RuntimeSection — Docker and VPN on AWS follow the launch type", () => {
beforeEach(() => vi.clearAllMocks());
it("are off and disabled on Fargate, switchable on EC2", () => {
const fargate = awsProject();
const { unmount } = renderSection(fargate);
expect(screen.getByRole("switch", { name: VPN })).toBeDisabled();
expect(screen.getByRole("switch", { name: "Allow container spawning" })).toBeDisabled();
unmount();
renderSection({ ...fargate, aws: { ...fargate.aws!, launch_type: "ec2" } });
expect(screen.getByRole("switch", { name: VPN })).toBeEnabled();
expect(screen.getByRole("switch", { name: "Allow container spawning" })).toBeEnabled();
});
it("saves the flags when toggled on EC2", () => {
const p = awsProject({ vpn_support_enabled: false, allow_docker_access: false });
renderSection({ ...p, aws: { ...p.aws!, launch_type: "ec2" } });
fireEvent.click(screen.getByRole("switch", { name: VPN }));
expect(save).toHaveBeenCalledWith({ vpn_support_enabled: true });
fireEvent.click(screen.getByRole("switch", { name: "Allow container spawning" }));
expect(save).toHaveBeenCalledWith({ allow_docker_access: true });
});
it("launchTypePatch clears Docker and VPN when switching to Fargate", () => {
const p = awsProject({ allow_docker_access: true, vpn_support_enabled: true });
const toFargate = launchTypePatch(p, "fargate");
expect(toFargate).toMatchObject({ allow_docker_access: false, vpn_support_enabled: false });
expect(toFargate.aws?.launch_type).toBe("fargate");
const toEc2 = launchTypePatch(p, "ec2");
expect(toEc2.allow_docker_access).toBeUndefined();
expect(toEc2.aws?.launch_type).toBe("ec2");
});
});
@@ -5,7 +5,14 @@ import PermissionModeControl, { permissionModePatch } from "../../PermissionMode
import ClaudeInstructionsEditor from "../../ClaudeInstructionsEditor";
import ClaudeCodeSettingsEditor from "../../ClaudeCodeSettingsEditor";
import AuthBridgeRow from "./AuthBridgeRow";
import { NOT_ON_AWS, SANDBOX_OFF_ON_AWS, isAwsProject } from "../../../../lib/awsWorkspace";
import {
EC2_DOCKER_ACCESS_NOTE,
EC2_VPN_HINT,
NOT_ON_AWS,
SANDBOX_OFF_ON_AWS,
isAwsProject,
sandboxEffective,
} from "../../../../lib/awsWorkspace";
interface Props {
project: Project;
@@ -21,6 +28,12 @@ export default function RuntimeSection({
disabledReason,
}: Props) {
const aws = isAwsProject(project);
// Docker access and VPN need the user's own EC2 instance (not Fargate).
const ec2 = aws && project.aws?.launch_type === "ec2";
const noPrivileged = aws && !ec2;
// No ECS task can run Claude Code's sandbox: not on Fargate (P9), and not
// on the user's own EC2 instance either (E8, measured live).
const sandboxOff = aws;
return (
<>
<ConfigGroup
@@ -37,15 +50,15 @@ export default function RuntimeSection({
<SwitchRow
label="Sandbox mode"
hint={
aws
? `${SANDBOX_OFF_ON_AWS}: Claude Code's bash sandbox has not been verified on Fargate yet, so an AWS workspace runs with it off whatever this says.`
sandboxOff
? `${SANDBOX_OFF_ON_AWS}: Claude Code's bash sandbox (bubblewrap) can't create its namespaces in an ECS task, on Fargate or EC2, so an AWS workspace runs with it off whatever this says.`
: "Claude Code's bash sandbox (bubblewrap filesystem and network isolation). Triple-C is the source of truth: toggling this overrides any manual /sandbox configuration in the container's settings.json on next start."
}
control={
<Toggle
label="Sandbox mode"
checked={aws ? false : project.sandbox_mode_enabled}
disabled={disabled || aws}
checked={sandboxEffective(project)}
disabled={disabled || sandboxOff}
onChange={(v) => save({ sandbox_mode_enabled: v })}
/>
}
@@ -54,15 +67,17 @@ export default function RuntimeSection({
<SwitchRow
label="Allow container spawning"
hint={
aws
noPrivileged
? NOT_ON_AWS
: "Mounts the Docker socket so Claude can build and run Docker containers from inside the sandbox."
: ec2
? `Mounts the Docker socket so Claude can build and run Docker containers from inside the workspace. ${EC2_DOCKER_ACCESS_NOTE}`
: "Mounts the Docker socket so Claude can build and run Docker containers from inside the sandbox."
}
control={
<Toggle
label="Allow container spawning"
checked={aws ? false : project.allow_docker_access}
disabled={disabled || aws}
checked={noPrivileged ? false : project.allow_docker_access}
disabled={disabled || noPrivileged}
onChange={(v) => save({ allow_docker_access: v })}
/>
}
@@ -70,12 +85,12 @@ export default function RuntimeSection({
<SwitchRow
label="VPN support"
hint={aws ? NOT_ON_AWS : "Grants NET_ADMIN and the /dev/net/tun device so a VPN client (PIA, WireGuard, OpenVPN) can build a tunnel inside the container. Without it a client installs and runs but its connection hangs until it times out. Anything in the container can then reconfigure the container's own network stack; the host's is untouched. Changing this recreates the container on its next start — the home and .claude volumes are preserved."}
hint={noPrivileged ? NOT_ON_AWS : ec2 ? EC2_VPN_HINT : "Grants NET_ADMIN and the /dev/net/tun device so a VPN client (PIA, WireGuard, OpenVPN) can build a tunnel inside the container. Without it a client installs and runs but its connection hangs until it times out. Anything in the container can then reconfigure the container's own network stack; the host's is untouched. Changing this recreates the container on its next start — the home and .claude volumes are preserved."}
control={
<Toggle
label="VPN support"
checked={aws ? false : project.vpn_support_enabled}
disabled={disabled || aws}
checked={noPrivileged ? false : project.vpn_support_enabled}
disabled={disabled || noPrivileged}
onChange={(v) => save({ vpn_support_enabled: v })}
/>
}
@@ -1,6 +1,7 @@
import { describe, it, expect, vi, beforeEach } from "vitest";
import { fireEvent, render, screen, waitFor } from "@testing-library/react";
import AdminBundleModal from "./AdminBundleModal";
import { instanceTable } from "./testkit";
const invoke = vi.fn();
vi.mock("@tauri-apps/api/core", () => ({
@@ -11,7 +12,9 @@ describe("AdminBundleModal", () => {
beforeEach(() => vi.clearAllMocks());
it("saves the bundle and shows the folder and the two commands, with Copy", async () => {
invoke.mockResolvedValue("/home/me/triple-c-onboard-dev");
invoke.mockImplementation(async (cmd: string) =>
cmd === "list_ec2_instance_types" ? instanceTable : { folder: "/home/me/triple-c-onboard-dev", warning: null },
);
const writeText = vi.fn().mockResolvedValue(undefined);
Object.assign(navigator, { clipboard: { writeText } });
render(
@@ -28,11 +31,81 @@ describe("AdminBundleModal", () => {
environment: "dev",
ownerId: "AIDAEXAMPLE",
network: null,
ec2: null,
});
const commands = "tofu init -backend-config=backend.hcl\ntofu apply";
expect(screen.getByText((_, el) => el?.tagName === "PRE" && el.textContent === commands)).toBeInTheDocument();
fireEvent.click(screen.getByRole("button", { name: "Copy commands" }));
await waitFor(() => expect(writeText).toHaveBeenCalledWith(commands));
expect(invoke.mock.calls.filter((c) => c[0] === "save_aws_admin_bundle")).toHaveLength(1);
});
it("an onboarding bundle carries the chosen EC2 capacity", async () => {
invoke.mockImplementation(async (cmd: string) =>
cmd === "list_ec2_instance_types" ? instanceTable : { folder: "/tmp/onboard", warning: "An ECS cluster named triple-c-dev--u-bob already belongs to another onboarding." },
);
render(
<AdminBundleModal connectionId="h1" region="eu-west-1" kind="user_onboard" environment="dev" ownerId="AIDAEXAMPLE" onClose={vi.fn()} />,
);
fireEvent.click(screen.getByRole("radio", { name: "My own EC2 instances" }));
await waitFor(() => expect((screen.getByLabelText("Instance type") as HTMLSelectElement).value).toBe("m7i.large"));
const save = screen.getByRole("button", { name: "Save bundle to folder…" });
// The triple-c-user value is required with EC2.
expect(save).toBeDisabled();
fireEvent.change(screen.getByLabelText("Your triple-c-user value"), { target: { value: " bob " } });
fireEvent.change(screen.getByLabelText("Maximum instances"), { target: { value: "4" } });
fireEvent.click(save);
await waitFor(() =>
expect(invoke).toHaveBeenCalledWith("save_aws_admin_bundle", {
connectionId: "h1",
region: "eu-west-1",
kind: "user_onboard",
environment: "dev",
ownerId: "AIDAEXAMPLE",
network: null,
ec2: { instance_type: "m7i.large", max_instances: 4, user_tag: "bob" },
}),
);
// The app's partial check of the name (cluster only) warns, never refuses.
expect(await screen.findByText(/already belongs to another onboarding/)).toBeInTheDocument();
expect(screen.getByText("/tmp/onboard")).toBeInTheDocument();
});
it("keeps the recorded instance type, read-only, when the list can't load", async () => {
invoke.mockImplementation(async (cmd: string) => {
if (cmd === "list_ec2_instance_types") throw new Error("boom");
return null;
});
render(
<AdminBundleModal
connectionId="h1"
region="eu-west-1"
kind="user_onboard"
environment="dev"
ownerId="AIDAEXAMPLE"
recordedEc2={{
instance_type: "r7g.2xlarge",
max_instances: 2,
user_tag: "alice",
cluster_arn: "arn:aws:ecs:eu-west-1:123456789012:cluster/triple-c-dev--u-alice",
capacity_provider: "triple-c-dev--u-alice",
}}
onClose={vi.fn()}
/>,
);
expect(await screen.findByText(/Could not load the instance types/)).toBeInTheDocument();
const type = screen.getByLabelText("Instance type") as HTMLSelectElement;
expect(type).toBeDisabled();
expect(type.value).toBe("r7g.2xlarge");
expect(screen.getByText(/keeping r7g.2xlarge/)).toBeInTheDocument();
// The kept values can still be saved.
expect(screen.getByRole("button", { name: "Save bundle to folder…" })).toBeEnabled();
});
it("an environment bundle has no EC2 capacity", () => {
invoke.mockResolvedValue(instanceTable);
render(<AdminBundleModal connectionId="h1" region="eu-west-1" kind="environment" onClose={vi.fn()} />);
expect(screen.queryByRole("radio", { name: "My own EC2 instances" })).toBeNull();
});
it("shows nothing new when the folder picker is dismissed", async () => {
@@ -44,7 +117,7 @@ describe("AdminBundleModal", () => {
});
it("an environment bundle asks for the name and network", async () => {
invoke.mockResolvedValue("/tmp/env");
invoke.mockResolvedValue({ folder: "/tmp/env", warning: null });
render(<AdminBundleModal connectionId="h1" region="eu-west-1" kind="environment" onClose={vi.fn()} />);
fireEvent.change(screen.getByLabelText("Name"), { target: { value: "team" } });
fireEvent.click(screen.getByRole("radio", { name: "Private + NAT" }));
@@ -63,7 +136,7 @@ describe("AdminBundleModal", () => {
});
it("an environment bundle carries the lifetime cap", async () => {
invoke.mockResolvedValue("/tmp/env");
invoke.mockResolvedValue({ folder: "/tmp/env", warning: null });
render(<AdminBundleModal connectionId="h1" region="eu-west-1" kind="environment" onClose={vi.fn()} />);
fireEvent.change(screen.getByLabelText("Maximum workspace lifetime (hours, 0 = none)"), { target: { value: "24" } });
fireEvent.click(screen.getByRole("button", { name: "Save bundle to folder…" }));
@@ -1,13 +1,22 @@
import { useState } from "react";
import { saveAwsAdminBundle } from "../../../lib/tauri-commands";
import { awsErrorMessage, toAwsError } from "../../../lib/aws-error";
import type { AwsError, ModuleKind, NetworkChoice } from "../../../lib/types";
import type { AwsError, ModuleKind, NetworkChoice, UserCompute } from "../../../lib/types";
import Button from "../../ui/Button";
import Field, { inputClass } from "../../ui/Field";
import Modal from "../../ui/Modal";
import LifetimeCapField, { parseCapHours } from "./LifetimeCapField";
import Ec2CapacityFields from "./Ec2CapacityFields";
import NetworkFields from "./NetworkFields";
import { DEFAULT_NETWORK, environmentNameError, networkFromDraft, type NetworkDraft } from "./environmentFormat";
import {
DEFAULT_NETWORK,
ec2DraftFrom,
ec2FromDraft,
environmentNameError,
networkFromDraft,
type Ec2Draft,
type NetworkDraft,
} from "./environmentFormat";
const COMMANDS = "tofu init -backend-config=backend.hcl\ntofu apply";
@@ -29,6 +38,7 @@ export default function AdminBundleModal({
environment,
ownerId,
network,
recordedEc2,
reason,
onSaved,
onClose,
@@ -40,6 +50,8 @@ export default function AdminBundleModal({
environment?: string;
ownerId?: string;
network?: NetworkChoice;
/** An onboarding bundle's starting EC2 capacity: the user's recorded one, if any. */
recordedEc2?: UserCompute | null;
/** Why the bundle is needed (e.g. the refusal that led here). */
reason?: string;
/** Called with the folder once a bundle was actually written. */
@@ -50,29 +62,39 @@ export default function AdminBundleModal({
const [name, setName] = useState("main");
const [draft, setDraft] = useState<NetworkDraft>(DEFAULT_NETWORK);
const [capText, setCapText] = useState("0");
const [ec2Draft, setEc2Draft] = useState<Ec2Draft>(() => ec2DraftFrom(recordedEc2));
const [saving, setSaving] = useState(false);
const [saved, setSaved] = useState<string | null>(null);
const [warning, setWarning] = useState<string | null>(null);
const [error, setError] = useState<AwsError | null>(null);
const [copyNote, setCopyNote] = useState<string | null>(null);
const nameError = askEnvironment ? environmentNameError(name) : null;
const picked = askEnvironment ? networkFromDraft(draft) : null;
const capHours = parseCapHours(capText);
const ready = !nameError && (!picked || "network" in picked) && (!askEnvironment || capHours !== null);
const onboarding = kind === "user_onboard";
const ec2 = onboarding ? ec2FromDraft(ec2Draft) : null;
const ready =
!nameError &&
(!picked || "network" in picked) &&
(!askEnvironment || capHours !== null) &&
(!ec2 || "compute" in ec2);
const save = async () => {
setSaving(true);
setError(null);
try {
const folder = await saveAwsAdminBundle(connectionId, region, kind, {
const result = await saveAwsAdminBundle(connectionId, region, kind, {
environment: askEnvironment ? name : environment,
ownerId,
network: picked && "network" in picked ? picked.network : network,
...(askEnvironment && capHours !== null ? { maxWorkspaceHours: capHours } : {}),
...(ec2 && "compute" in ec2 ? { ec2: ec2.compute } : {}),
});
if (folder) {
setSaved(folder);
onSaved?.(folder);
if (result) {
setSaved(result.folder);
setWarning(result.warning);
onSaved?.(result.folder);
}
} catch (e) {
setError(toAwsError(e));
@@ -116,6 +138,7 @@ export default function AdminBundleModal({
<p>
Joining {environment} needs your own task role and file system, which an environment admin creates
by applying this bundle. Send them the folder; once it's applied, Detect again shows you as joined.
With EC2 capacity, its README has the admin first check that your EC2 names aren't already taken.
</p>
)}
{kind === "permissions_boundary" && (
@@ -125,6 +148,7 @@ export default function AdminBundleModal({
</p>
)}
<p>The bundle works with OpenTofu or Terraform 1.10 and later. Its README lists every step.</p>
{onboarding && <Ec2CapacityFields value={ec2Draft} onChange={setEc2Draft} />}
{askEnvironment && (
<div className="space-y-3">
<Field label="Name" hint={nameError ?? undefined}>
@@ -144,6 +168,11 @@ export default function AdminBundleModal({
)}
{saved && (
<div className="space-y-1.5">
{warning && (
<p role="alert" className="text-[var(--warning)]">
{warning}
</p>
)}
<p>
Saved to <code className="font-mono text-[var(--text-primary)] break-all">{saved}</code>
</p>
@@ -144,7 +144,7 @@ describe("DeployEnvironmentModal", () => {
it("offers the boundary bundle when the boundary run fails, with its summary as the reason", async () => {
invoke.mockImplementation(async (cmd: string) => {
if (cmd === "deploy_aws_permissions_boundary") return "b1";
if (cmd === "save_aws_admin_bundle") return "/tmp/b";
if (cmd === "save_aws_admin_bundle") return { folder: "/tmp/b", warning: null };
throw new Error(`unexpected ${cmd}`);
});
const props = renderModal({ scan: scan({ boundary: "no" }) });
@@ -166,7 +166,7 @@ describe("DeployEnvironmentModal", () => {
it("opens the boundary bundle when the boundary command itself is denied", async () => {
invoke.mockImplementation(async (cmd: string) => {
if (cmd === "deploy_aws_permissions_boundary") throw { kind: "access_denied", action: "iam:CreatePolicy", message: "denied iam:CreatePolicy" };
if (cmd === "save_aws_admin_bundle") return "/tmp/b";
if (cmd === "save_aws_admin_bundle") return { folder: "/tmp/b", warning: null };
throw new Error(`unexpected ${cmd}`);
});
renderModal({ scan: scan({ boundary: "no" }) });
@@ -0,0 +1,142 @@
import { useEffect, useState } from "react";
import { listEc2InstanceTypes } from "../../../lib/tauri-commands";
import type { Ec2InstanceTable, Ec2InstanceType } from "../../../lib/types";
import Field, { monoInputClass, selectClass } from "../../ui/Field";
import SegmentedControl from "../../ui/SegmentedControl";
import { EC2_MAX_INSTANCES, userTagError, type Ec2Draft } from "./environmentFormat";
const label = (t: Ec2InstanceType) => `${t.instance_type} — ${t.vcpu} vCPU, ${t.memory_mib / 1024} GiB`;
/**
* An onboarding's EC2 capacity (phase 7): none, or the user's own instances
* of one type from the app's table, up to a maximum, on the cluster named
* after their `triple-c-user` sign-in tag. Shared by self-onboarding and the
* onboarding bundle.
*/
export default function Ec2CapacityFields({
value,
onChange,
}: {
value: Ec2Draft;
onChange: (next: Ec2Draft) => void;
}) {
const [table, setTable] = useState<Ec2InstanceTable | null>(null);
const [loadError, setLoadError] = useState(false);
const set = (patch: Partial<Ec2Draft>) => onChange({ ...value, ...patch });
useEffect(() => {
let live = true;
listEc2InstanceTypes()
.then((t) => live && setTable(t))
.catch(() => live && setLoadError(true));
return () => {
live = false;
};
}, []);
// The table's default fills an empty choice once the table is known.
useEffect(() => {
if (table && !value.instanceType) onChange({ ...value, instanceType: table.default_type });
}, [table, value, onChange]);
const tagError = value.enabled && value.userTag ? userTagError(value.userTag.trim()) : null;
const x86 = table?.types.filter((t) => t.arch === "x86_64") ?? [];
const arm = table?.types.filter((t) => t.arch === "arm64") ?? [];
return (
<div className="space-y-2">
<div className="text-[13px] font-medium text-[var(--text-primary)]">EC2 capacity</div>
<SegmentedControl<"none" | "ec2">
label="EC2 capacity"
value={value.enabled ? "ec2" : "none"}
onChange={(v) => set({ enabled: v === "ec2" })}
segments={[
{ value: "none", label: "None (Fargate only)" },
{ value: "ec2", label: "My own EC2 instances" },
]}
/>
<p className="text-xs text-[var(--text-secondary)] leading-snug">
{value.enabled
? "Projects set to EC2 run on instances only you use. Instances scale to zero when none of your workspaces run; a start after idle takes 2–3 minutes longer. You pay the instance's hourly price while it runs."
: "Every project runs on Fargate. You can add EC2 capacity later by applying again."}
</p>
{value.enabled && (
<div className="space-y-2">
<Field label="Instance type">
{(id) => (
<select
id={id}
className={`${selectClass} w-full`}
value={value.instanceType}
disabled={!table}
onChange={(e) => set({ instanceType: e.target.value })}
>
{!table && <option value={value.instanceType}>{value.instanceType || "Loading…"}</option>}
{x86.length > 0 && (
<optgroup label="x86 (Intel)">
{x86.map((t) => (
<option key={t.instance_type} value={t.instance_type}>
{label(t)}
</option>
))}
</optgroup>
)}
{arm.length > 0 && (
<optgroup label="ARM (Graviton)">
{arm.map((t) => (
<option key={t.instance_type} value={t.instance_type}>
{label(t)}
</option>
))}
</optgroup>
)}
</select>
)}
</Field>
{loadError && (
<p role="alert" className="text-xs text-[var(--error)]">
Could not load the instance types
{value.instanceType ? `; keeping ${value.instanceType}.` : "; detect again to choose one."}
</p>
)}
<Field label="Maximum instances" hint="How many of your EC2 workspaces can run at once (one per instance).">
{(id) => (
<select
id={id}
className={`${selectClass} w-full`}
value={value.maxInstances}
onChange={(e) => set({ maxInstances: Number(e.target.value) })}
>
{Array.from({ length: EC2_MAX_INSTANCES }, (_, i) => i + 1).map((n) => (
<option key={n} value={n}>
{n}
</option>
))}
</select>
)}
</Field>
<Field
label="Your triple-c-user value"
hint={tagError ?? "The value an admin set as your triple-c-user sign-in tag. Your EC2 cluster is named after it."}
>
{(id) => (
<input
id={id}
className={monoInputClass}
autoComplete="off"
spellCheck={false}
value={value.userTag}
aria-invalid={!!tagError}
onChange={(e) => set({ userTag: e.target.value })}
/>
)}
</Field>
<p className="text-xs text-[var(--warning)] leading-snug">
Triple-C can't read your sign-in's tags to check this value. If it isn't exactly your triple-c-user tag,
EC2 workspaces fail to start with access denied.
</p>
</div>
)}
</div>
);
}
@@ -1,8 +1,9 @@
import { describe, it, expect, vi, beforeEach } from "vitest";
import { act, fireEvent, render, screen, waitFor, within } from "@testing-library/react";
import EnvironmentsPanel from "./EnvironmentsPanel";
import JoinEnvironmentModal from "./JoinEnvironmentModal";
import type { DeployProgress, EnvironmentScan } from "../../../lib/types";
import { conn, env, info, progress, scan, tofuEvent } from "./testkit";
import { compute, conn, env, info, instanceTable, progress, scan, tofuEvent } from "./testkit";
const invoke = vi.fn();
vi.mock("@tauri-apps/api/core", () => ({
@@ -72,7 +73,7 @@ describe("EnvironmentsPanel", () => {
});
it("deploy is secondary when environments exist", async () => {
backend(() => scan({ environments: [env({ onboarding: { state: "joined" }, actions: [] })] }));
backend(() => scan({ environments: [env({ onboarding: { state: "joined", ec2: null }, actions: [] })] }));
render(<EnvironmentsPanel connection={conn} tier="environment_admin" />);
await open();
const deploy = await screen.findByRole("button", { name: "Deploy a new environment" });
@@ -204,9 +205,10 @@ describe("EnvironmentsPanel", () => {
backend(
() => {
detects += 1;
return scan({ environments: [env({ onboarding: detects > 1 ? { state: "joined" } : { state: "not_joined" } })] });
return scan({ environments: [env({ onboarding: detects > 1 ? { state: "joined", ec2: null } : { state: "not_joined" } })] });
},
async (cmd) => {
if (cmd === "list_ec2_instance_types") return instanceTable;
if (cmd === "onboard_aws_user") {
// The backend can emit before the invoke resolves: an old follower's
// line (another op) and this run's first line.
@@ -220,9 +222,18 @@ describe("EnvironmentsPanel", () => {
render(<EnvironmentsPanel connection={conn} tier="environment_admin" />);
await open();
fireEvent.click(await screen.findByRole("button", { name: "Join dev" }));
// The join dialog asks for EC2 capacity first; none is the default.
const dialog = screen.getByRole("dialog", { name: "Join dev" });
fireEvent.click(within(dialog).getByRole("button", { name: "Join" }));
expect(await screen.findByText("Planning")).toBeInTheDocument();
expect(screen.queryByRole("dialog")).toBeNull();
expect(screen.queryByText("stale line")).toBeNull();
expect(invoke).toHaveBeenCalledWith("onboard_aws_user", { connectionId: "h1", region: "eu-west-1", environment: "dev" });
expect(invoke).toHaveBeenCalledWith("onboard_aws_user", {
connectionId: "h1",
region: "eu-west-1",
environment: "dev",
ec2: { kind: "none" },
});
expect(screen.getByRole("button", { name: "Cancel deployment" })).toBeDisabled();
emit(progress({ op_id: "op1", state_key: key, events: [tofuEvent("Creating...", { resource: "aws_iam_role.task", action: "create" })] }));
@@ -241,7 +252,7 @@ describe("EnvironmentsPanel", () => {
backend(
() => scan({ environments: [env()] }),
(cmd) => {
if (cmd === "save_aws_admin_bundle") return "/home/me/onboard-dev";
if (cmd === "save_aws_admin_bundle") return { folder: "/home/me/onboard-dev", warning: null };
throw new Error(`unexpected ${cmd}`);
},
);
@@ -261,6 +272,7 @@ describe("EnvironmentsPanel", () => {
environment: "dev",
ownerId: "AIDAEXAMPLEEXAMPLE01",
network: null,
ec2: null,
}),
);
expect(invoke.mock.calls.some((c) => c[0] === "onboard_aws_user")).toBe(false);
@@ -268,6 +280,170 @@ describe("EnvironmentsPanel", () => {
expect(within(screen.getByTestId("env-dev")).getByText("Waiting for an admin to onboard you")).toBeInTheDocument();
});
it("an admin joins with EC2 capacity: a type from the table, max instances and their triple-c-user value", async () => {
backend(
() => scan({ environments: [env()] }),
(cmd) => {
if (cmd === "list_ec2_instance_types") return instanceTable;
if (cmd === "onboard_aws_user") return "op1";
throw new Error(`unexpected ${cmd}`);
},
);
render(<EnvironmentsPanel connection={conn} tier="environment_admin" />);
await open();
fireEvent.click(await screen.findByRole("button", { name: "Join dev" }));
const dialog = screen.getByRole("dialog", { name: "Join dev" });
fireEvent.click(within(dialog).getByRole("radio", { name: "My own EC2 instances" }));
const type = within(dialog).getByLabelText("Instance type") as HTMLSelectElement;
// The table's default, grouped by architecture.
await waitFor(() => expect(type.value).toBe("m7i.large"));
const groups = [...type.querySelectorAll("optgroup")].map((g) => g.label);
expect(groups).toEqual(["x86 (Intel)", "ARM (Graviton)"]);
expect(within(dialog).getByRole("option", { name: "m7g.large — 2 vCPU, 8 GiB" })).toBeInTheDocument();
expect(within(dialog).getByText(/admin set as your triple-c-user sign-in tag/)).toBeInTheDocument();
expect(within(dialog).getByText(/fail to start with access denied/)).toBeInTheDocument();
// The value is required and validated.
const joinButton = within(dialog).getByRole("button", { name: "Join" });
expect(joinButton).toBeDisabled();
const tag = within(dialog).getByLabelText("Your triple-c-user value");
fireEvent.change(tag, { target: { value: "al--ice" } });
expect(within(dialog).getByText(/single hyphens/)).toBeInTheDocument();
expect(joinButton).toBeDisabled();
fireEvent.change(tag, { target: { value: "alice" } });
fireEvent.change(type, { target: { value: "m7g.large" } });
fireEvent.change(within(dialog).getByLabelText("Maximum instances"), { target: { value: "2" } });
fireEvent.click(joinButton);
await waitFor(() =>
expect(invoke).toHaveBeenCalledWith("onboard_aws_user", {
connectionId: "h1",
region: "eu-west-1",
environment: "dev",
ec2: { kind: "ec2", instance_type: "m7g.large", max_instances: 2, user_tag: "alice" },
}),
);
});
it("shows a joined user's compute, and a change starts from the recorded values", async () => {
backend(
() =>
scan({
environments: [
env({ onboarding: { state: "joined", ec2: compute() } }),
env({ name: "plain", onboarding: { state: "joined", ec2: null } }),
],
}),
(cmd) => {
if (cmd === "list_ec2_instance_types") return instanceTable;
if (cmd === "onboard_aws_user") return "op1";
throw new Error(`unexpected ${cmd}`);
},
);
render(<EnvironmentsPanel connection={conn} tier="environment_admin" />);
await open();
expect(await screen.findByTestId("compute-dev")).toHaveTextContent(
"EC2 capacity: m7g.xlarge, up to 3 instances (triple-c-user alice)",
);
expect(screen.getByTestId("compute-plain")).toHaveTextContent("EC2 capacity: none (Fargate only)");
// Joined: no Join, but the capacity can be changed (a re-apply).
const row = screen.getByTestId("env-dev");
expect(within(row).queryByRole("button", { name: "Join dev" })).toBeNull();
fireEvent.click(within(row).getByRole("button", { name: "EC2 capacity in dev" }));
const dialog = screen.getByRole("dialog", { name: "EC2 capacity in dev" });
expect((within(dialog).getByLabelText("Instance type") as HTMLSelectElement).value).toBe("m7g.xlarge");
expect((within(dialog).getByLabelText("Maximum instances") as HTMLSelectElement).value).toBe("3");
expect((within(dialog).getByLabelText("Your triple-c-user value") as HTMLInputElement).value).toBe("alice");
fireEvent.click(within(dialog).getByRole("button", { name: "Apply" }));
await waitFor(() =>
expect(invoke).toHaveBeenCalledWith("onboard_aws_user", {
connectionId: "h1",
region: "eu-west-1",
environment: "dev",
ec2: { kind: "ec2", instance_type: "m7g.xlarge", max_instances: 3, user_tag: "alice" },
}),
);
});
it("finishing an incomplete onboarding starts from its record and says when it removes EC2", async () => {
backend(
() => scan({ environments: [env({ onboarding: { state: "incomplete", lock: null, ec2: compute() } })] }),
(cmd) => {
if (cmd === "list_ec2_instance_types") return instanceTable;
throw new Error(`unexpected ${cmd}`);
},
);
render(<EnvironmentsPanel connection={conn} tier="environment_admin" />);
await open();
fireEvent.click(await screen.findByRole("button", { name: "Join dev" }));
const dialog = screen.getByRole("dialog", { name: "Join dev" });
expect((within(dialog).getByLabelText("Your triple-c-user value") as HTMLInputElement).value).toBe("alice");
expect(within(dialog).queryByText(/This removes your EC2 capacity/)).toBeNull();
fireEvent.click(within(dialog).getByRole("radio", { name: "None (Fargate only)" }));
expect(within(dialog).getByText("This removes your EC2 capacity (m7g.xlarge) in dev.")).toBeInTheDocument();
});
it("an incomplete onboarding with no record warns that None removes half-built EC2 capacity", async () => {
backend(
() => scan({ environments: [env({ onboarding: { state: "incomplete", lock: null, ec2: null } })] }),
(cmd) => {
if (cmd === "list_ec2_instance_types") return instanceTable;
throw new Error(`unexpected ${cmd}`);
},
);
render(<EnvironmentsPanel connection={conn} tier="environment_admin" />);
await open();
fireEvent.click(await screen.findByRole("button", { name: "Join dev" }));
const dialog = screen.getByRole("dialog", { name: "Join dev" });
expect(within(dialog).getByRole("alert")).toHaveTextContent(
/didn't finish, and it may have built part of your EC2 capacity. Choosing None removes anything it built/,
);
fireEvent.click(within(dialog).getByRole("radio", { name: "My own EC2 instances" }));
expect(within(dialog).queryByText(/Choosing None removes/)).toBeNull();
});
it("with no record, the join starts from the EC2 capacity last chosen here", async () => {
invoke.mockImplementation(async (cmd: string) => {
if (cmd === "list_ec2_instance_types") return instanceTable;
throw new Error(`unexpected ${cmd}`);
});
render(
<JoinEnvironmentModal
environment="dev"
region="eu-west-1"
recorded={null}
reapply={false}
incomplete
lastChoice={{ instance_type: "m7g.xlarge", max_instances: 2, user_tag: "alice" }}
onJoin={async () => null}
onClose={() => {}}
/>,
);
const dialog = screen.getByRole("dialog", { name: "Join dev" });
expect((within(dialog).getByLabelText("Your triple-c-user value") as HTMLInputElement).value).toBe("alice");
expect(within(dialog).queryByText(/Choosing None removes/)).toBeNull();
});
it("a joined workspace user's capacity change is a bundle starting from their record", async () => {
backend(
() => scan({ environments: [env({ onboarding: { state: "joined", ec2: compute({ max_instances: 1 }) } })] }),
(cmd) => {
if (cmd === "list_ec2_instance_types") return instanceTable;
if (cmd === "save_aws_admin_bundle") return { folder: "/home/me/onboard-dev", warning: null };
throw new Error(`unexpected ${cmd}`);
},
);
render(<EnvironmentsPanel connection={conn} tier="workspace_user" />);
await open();
fireEvent.click(await screen.findByRole("button", { name: "EC2 capacity in dev" }));
const dialog = screen.getByRole("dialog", { name: "Onboarding bundle" });
fireEvent.click(within(dialog).getByRole("button", { name: "Save bundle to folder…" }));
await waitFor(() =>
expect(invoke).toHaveBeenCalledWith(
"save_aws_admin_bundle",
expect.objectContaining({ ec2: { instance_type: "m7g.xlarge", max_instances: 1, user_tag: "alice" } }),
),
);
});
it("follows a deployment running on this machine when opened, and cancels after confirming", async () => {
const lock = { who: "me@here", created: "x", operation: "OperationTypeApply", age_secs: 60, local_process: true };
backend(
@@ -303,7 +479,7 @@ describe("EnvironmentsPanel", () => {
it("Update pre-fills the environment's recorded cap and sends what is shown", async () => {
backend(
() => scan({ environments: [env({ state: { state: "update_available" }, actions: ["join", "update"], info: { ...info("dev"), max_workspace_hours: 24 }, onboarding: { state: "joined" } })] }),
() => scan({ environments: [env({ state: { state: "update_available" }, actions: ["join", "update"], info: { ...info("dev"), max_workspace_hours: 24 }, onboarding: { state: "joined", ec2: null } })] }),
async (cmd) => {
if (cmd === "update_aws_environment") return "u1";
throw new Error(`unexpected ${cmd}`);
@@ -326,7 +502,7 @@ describe("EnvironmentsPanel", () => {
it("shows a failed run's errors in full", async () => {
backend(
() => scan({ environments: [env({ state: { state: "update_available" }, actions: ["join", "update"], onboarding: { state: "joined" } })] }),
() => scan({ environments: [env({ state: { state: "update_available" }, actions: ["join", "update"], onboarding: { state: "joined", ec2: null } })] }),
async (cmd) => {
if (cmd === "update_aws_environment") return "u1";
throw new Error(`unexpected ${cmd}`);
@@ -383,7 +559,7 @@ describe("EnvironmentsPanel", () => {
expect(screen.queryByRole("button", { name: "Destroy dev" })).toBeNull();
expect(screen.queryByRole("button", { name: "Deploy a new environment" })).toBeNull();
await act(async () =>
resolveUs(scan({ region: "us-east-1", environments: [env({ state: { state: "update_available" }, actions: ["update"], onboarding: { state: "joined" } })] })),
resolveUs(scan({ region: "us-east-1", environments: [env({ state: { state: "update_available" }, actions: ["update"], onboarding: { state: "joined", ec2: null } })] })),
);
fireEvent.click(await screen.findByRole("button", { name: "Update dev" }));
fireEvent.click(within(screen.getByRole("dialog", { name: "Update dev" })).getByRole("button", { name: "Update" }));
@@ -399,7 +575,7 @@ describe("EnvironmentsPanel", () => {
invoke.mockImplementation(async (cmd: string) => {
if (cmd !== "detect_aws_environments") throw new Error(`unexpected ${cmd}`);
calls += 1;
const found = scan({ environments: [env({ actions: ["join", "update"], onboarding: { state: "joined" } })] });
const found = scan({ environments: [env({ actions: ["join", "update"], onboarding: { state: "joined", ec2: null } })] });
if (calls === 1) return found;
return new Promise<EnvironmentScan>((r) => (release = () => r(found)));
});
@@ -414,7 +590,7 @@ describe("EnvironmentsPanel", () => {
it("a done that arrives before the invoke resolves is shown and re-detects", async () => {
backend(
() => scan({ environments: [env({ state: { state: "update_available" }, actions: ["update"], onboarding: { state: "joined" } })] }),
() => scan({ environments: [env({ state: { state: "update_available" }, actions: ["update"], onboarding: { state: "joined", ec2: null } })] }),
async (cmd) => {
if (cmd === "update_aws_environment") {
emit(progress({ op_id: "u1", events: [tofuEvent("Refreshing state")] }));
@@ -482,7 +658,7 @@ describe("EnvironmentsPanel", () => {
backend(
() =>
scan({
environments: [env({ onboarding: { state: "incomplete", lock }, onboarding_actions: ["unlock"] })],
environments: [env({ onboarding: { state: "incomplete", lock, ec2: null }, onboarding_actions: ["unlock"] })],
}),
(cmd) => {
if (cmd === "unlock_aws_state") return undefined;
@@ -515,7 +691,7 @@ describe("EnvironmentsPanel", () => {
scan({
environments: [
env({ name: "a", state: { state: "incomplete", lock: null }, actions: ["push_state"], info: null, onboarding: null }),
env({ name: "b", onboarding: { state: "incomplete", lock: null }, onboarding_actions: ["push_state"] }),
env({ name: "b", onboarding: { state: "incomplete", lock: null, ec2: null }, onboarding_actions: ["push_state"] }),
],
}),
(cmd) => {
@@ -11,9 +11,12 @@ import {
updateAwsEnvironment,
} from "../../../lib/tauri-commands";
import { awsErrorMessage, toAwsError } from "../../../lib/aws-error";
import { clearEc2CapacityCache } from "../../../lib/ec2Capacity";
import type {
AwsError,
DetectedEnvironment,
Ec2Choice,
Ec2Compute,
EnvironmentScan,
HostConnection,
LockInfo,
@@ -21,6 +24,7 @@ import type {
NetworkChoice,
PermissionSummary,
PolicyFile,
UserCompute,
} from "../../../lib/types";
import Button from "../../ui/Button";
import Field, { inputClass, selectClass } from "../../ui/Field";
@@ -29,10 +33,12 @@ import StatusIndicator from "../../ui/StatusIndicator";
import AdminBundleModal from "./AdminBundleModal";
import UpdateEnvironmentModal from "./UpdateEnvironmentModal";
import DeployEnvironmentModal from "./DeployEnvironmentModal";
import JoinEnvironmentModal from "./JoinEnvironmentModal";
import DeployProgress, { runKey, useDeployRuns, type DeployRun } from "./DeployProgress";
import PolicyFiles from "./PolicyFiles";
import {
COMMON_REGIONS,
computeLine,
environmentStateKey,
formatAge,
lockLine,
@@ -101,7 +107,27 @@ function TypedConfirmModal({
);
}
type Bundle = { kind: ModuleKind; region: string; environment?: string; ownerId?: string };
type Bundle = {
kind: ModuleKind;
region: string;
environment?: string;
ownerId?: string;
recordedEc2?: UserCompute | null;
};
/** A self-onboarding (or its re-apply) waiting for its EC2 choice. */
type JoinOf = {
region: string;
name: string;
stateKey: string;
recorded: UserCompute | null;
reapply: boolean;
incomplete: boolean;
lastChoice: Ec2Compute | null;
};
/** The caller's recorded EC2 capacity in `e`, if any. */
const recordedEc2 = (e: DetectedEnvironment): UserCompute | null =>
e.onboarding && (e.onboarding.state === "joined" || e.onboarding.state === "incomplete") ? e.onboarding.ec2 : null;
/** A row-level key: the same environment name can exist in every region. */
const rowKey = (region: string, name: string) => `${region}:${name}`;
@@ -145,6 +171,10 @@ export default function EnvironmentsPanel({
onboarding: boolean;
} | null>(null);
const [bundle, setBundle] = useState<Bundle | null>(null);
const [joinOf, setJoinOf] = useState<JoinOf | null>(null);
// The EC2 capacity last chosen per environment in this run of the app: an
// onboarding that didn't finish has no record to start from.
const lastEc2 = useRef(new Map<string, Ec2Compute>());
const [copyNote, setCopyNote] = useState<string | null>(null);
const request = useRef(0);
const followed = useRef(new Set<string>());
@@ -155,6 +185,9 @@ export default function EnvironmentsPanel({
setDetectError(null);
try {
const found = await detectAwsEnvironments(connection.id, region);
// A scan reads every user record afresh (a join may have changed the
// EC2 capacity): Host settings ask again too.
clearEc2CapacityCache();
if (id === request.current) setScan(found);
} catch (e) {
if (id === request.current) {
@@ -241,15 +274,38 @@ export default function EnvironmentsPanel({
updateAwsEnvironment(connection.id, runRegion, name, hours),
);
// Join (or, once joined, a change of EC2 capacity): admins choose the
// capacity and apply it themselves; others get a bundle for an admin.
const join = (found: EnvironmentScan, e: DetectedEnvironment) => {
const recorded = recordedEc2(e);
if (isAdmin) {
void launch(found.region, e.name, onboardingStateKey(e.name, found.owner_key), `Joining ${e.name}`, () =>
onboardAwsUser(connection.id, found.region, e.name),
);
setJoinOf({
region: found.region,
name: e.name,
stateKey: onboardingStateKey(e.name, found.owner_key),
recorded,
reapply: e.onboarding?.state === "joined",
incomplete: e.onboarding?.state === "incomplete",
lastChoice: lastEc2.current.get(rowKey(found.region, e.name)) ?? null,
});
} else {
setBundle({ kind: "user_onboard", region: found.region, environment: e.name, ownerId: found.owner_id });
setBundle({
kind: "user_onboard",
region: found.region,
environment: e.name,
ownerId: found.owner_id,
recordedEc2: recorded,
});
}
};
const runJoin = (j: JoinOf, ec2: Ec2Choice) => {
const key = rowKey(j.region, j.name);
if (ec2.kind === "ec2") {
const { kind: _kind, ...c } = ec2;
lastEc2.current.set(key, c);
} else lastEc2.current.delete(key);
return launch(j.region, j.name, j.stateKey, `Joining ${j.name}`, () => onboardAwsUser(connection.id, j.region, j.name, ec2));
};
const rowRuns = (name: string) =>
Object.values(runs).filter((r) => r.region === current?.region && runEnvironment(r.stateKey) === name);
@@ -343,6 +399,14 @@ export default function EnvironmentsPanel({
: "Your last onboarding didn't finish; Join again to finish it."}
</div>
)}
{(onboarding === "joined" || recordedEc2(e)) && (
<div className="text-[var(--text-secondary)]" data-testid={`compute-${e.name}`}>
EC2 capacity: {(() => {
const c = recordedEc2(e);
return c ? computeLine(c) : "none (Fargate only)";
})()}
</div>
)}
{!isAdmin && isWaiting && onboarding !== "joined" && (
<div className="text-[var(--text-secondary)]">Waiting for an admin to onboard you</div>
)}
@@ -359,6 +423,16 @@ export default function EnvironmentsPanel({
{!isAdmin && isWaiting ? "Show onboarding bundle" : "Join"}
</Button>
)}
{has("join") && onboarding === "joined" && (
<Button
size="sm"
disabled={busy || joinBlocked}
aria-label={`EC2 capacity in ${e.name}`}
onClick={() => join(found, e)}
>
{isAdmin ? "Change EC2 capacity" : "EC2 capacity bundle"}
</Button>
)}
{mine("follow") && (
<Button
size="sm"
@@ -636,6 +710,18 @@ export default function EnvironmentsPanel({
</p>
</TypedConfirmModal>
)}
{joinOf && (
<JoinEnvironmentModal
environment={joinOf.name}
region={joinOf.region}
recorded={joinOf.recorded}
reapply={joinOf.reapply}
incomplete={joinOf.incomplete}
lastChoice={joinOf.lastChoice}
onJoin={(ec2) => runJoin(joinOf, ec2)}
onClose={() => setJoinOf(null)}
/>
)}
{bundle && (
<AdminBundleModal
connectionId={connection.id}
@@ -643,6 +729,7 @@ export default function EnvironmentsPanel({
kind={bundle.kind}
environment={bundle.environment}
ownerId={bundle.ownerId}
recordedEc2={bundle.recordedEc2}
onSaved={() => {
if (bundle.kind === "user_onboard" && bundle.environment) {
const rk = rowKey(bundle.region, bundle.environment);
@@ -0,0 +1,99 @@
import { useState } from "react";
import { awsErrorMessage } from "../../../lib/aws-error";
import type { AwsError, Ec2Choice, Ec2Compute, UserCompute } from "../../../lib/types";
import Button from "../../ui/Button";
import Modal from "../../ui/Modal";
import Ec2CapacityFields from "./Ec2CapacityFields";
import { ec2DraftFrom, ec2FromDraft, type Ec2Draft } from "./environmentFormat";
/**
* Self-onboarding (environment admins): join an environment, or re-apply an
* onboarding to change its EC2 capacity. Starts from the recorded capacity,
* so a re-apply keeps it unless changed.
*/
export default function JoinEnvironmentModal({
environment,
region,
recorded,
reapply,
incomplete = false,
lastChoice = null,
onJoin,
onClose,
}: {
environment: string;
region: string;
/** The caller's recorded EC2 capacity, if any. */
recorded: UserCompute | null;
/** Already joined: this changes the onboarding's EC2 capacity. */
reapply: boolean;
/** The last onboarding didn't finish (it may have built EC2 capacity
* before writing the record). */
incomplete?: boolean;
/** The EC2 capacity last chosen here in this run of the app, if any:
* the starting point when there is no record. */
lastChoice?: Ec2Compute | null;
/** Starts the run; resolves to the command's error, or null once started. */
onJoin: (ec2: Ec2Choice) => Promise<AwsError | null>;
onClose: () => void;
}) {
const [draft, setDraft] = useState<Ec2Draft>(() => ec2DraftFrom(recorded ?? lastChoice));
const [submitting, setSubmitting] = useState(false);
const [error, setError] = useState<AwsError | null>(null);
const picked = ec2FromDraft(draft);
const ready = "compute" in picked && !submitting;
const verb = reapply ? "Apply" : "Join";
const submit = async () => {
if (!("compute" in picked)) return;
setSubmitting(true);
setError(null);
const choice: Ec2Choice = picked.compute ? { kind: "ec2", ...picked.compute } : { kind: "none" };
const err = await onJoin(choice);
setSubmitting(false);
if (err) setError(err);
else onClose();
};
return (
<Modal
title={reapply ? `EC2 capacity in ${environment}` : `Join ${environment}`}
description={`In ${region}.`}
onClose={onClose}
footer={
<>
<Button size="md" onClick={onClose}>
Cancel
</Button>
<Button size="md" variant="primary" onClick={submit} disabled={!ready}>
{submitting ? `${verb}ing…` : verb}
</Button>
</>
}
>
<div className="space-y-3 text-xs text-[var(--text-secondary)] leading-snug">
{!reapply && <p>Joining creates your own task role and file system in {environment}.</p>}
{recorded && (
<p>Stop your EC2 workspaces before removing EC2 capacity or changing your triple-c-user value: that change can't finish while one runs.</p>
)}
<Ec2CapacityFields value={draft} onChange={setDraft} />
{recorded && !draft.enabled && (
<p role="alert" className="text-[var(--warning)]">
This removes your EC2 capacity ({recorded.instance_type}) in {environment}.
</p>
)}
{incomplete && !recorded && !draft.enabled && (
<p role="alert" className="text-[var(--warning)]">
Your last onboarding in {environment} didn't finish, and it may have built part of your EC2 capacity.
Choosing None removes anything it built; to keep it, choose your own EC2 instances with the same values.
</p>
)}
{error && (
<p role="alert" className="text-[var(--error)]">
{awsErrorMessage(error)}
</p>
)}
</div>
</Modal>
);
}
@@ -1,6 +1,6 @@
// Pure helpers shared by the environments UI. The name and id rules mirror
// `aws/tofu_modules.rs`; the server re-validates everything.
import type { EnvironmentState, LockInfo, NetworkChoice } from "../../../lib/types";
import type { Ec2Compute, EnvironmentState, LockInfo, NetworkChoice, UserCompute } from "../../../lib/types";
import type { StatusTone } from "../../ui/StatusIndicator";
export const COMMON_REGIONS = [
@@ -97,3 +97,49 @@ export function stateStatus(s: EnvironmentState): { tone: StatusTone; label: str
export function stateLock(s: EnvironmentState): LockInfo | null {
return s.state === "deploying" || s.state === "incomplete" ? s.lock : null;
}
/** The most instances a user's EC2 capacity may run at once (`EC2_MAX_INSTANCES`). */
export const EC2_MAX_INSTANCES = 4;
/** The EC2 capacity picker's fields. An empty `instanceType` means the table's default. */
export interface Ec2Draft {
enabled: boolean;
instanceType: string;
maxInstances: number;
userTag: string;
}
/** The picker's start: the recorded capacity (a re-apply keeps it), or none. */
export function ec2DraftFrom(
recorded: Pick<UserCompute, "instance_type" | "max_instances" | "user_tag"> | null | undefined,
): Ec2Draft {
return recorded
? { enabled: true, instanceType: recorded.instance_type, maxInstances: recorded.max_instances, userTag: recorded.user_tag }
: { enabled: false, instanceType: "", maxInstances: 1, userTag: "" };
}
/** `validate_ec2_user_tag`: null when valid, else why not. */
export function userTagError(tag: string): string | null {
if (!tag) return "Enter your triple-c-user value.";
const ok = /^[A-Za-z0-9-]{1,32}$/.test(tag) && !tag.startsWith("-") && !tag.endsWith("-") && !tag.includes("--");
return ok ? null : "Use 1 to 32 letters, digits and single hyphens, not starting or ending with a hyphen.";
}
/** The capacity a draft asks for (null = none), or why it can't be used yet. */
export function ec2FromDraft(d: Ec2Draft): { compute: Ec2Compute | null } | { error: string } {
if (!d.enabled) return { compute: null };
if (!d.instanceType) return { error: "Choose an instance type." };
if (!Number.isInteger(d.maxInstances) || d.maxInstances < 1 || d.maxInstances > EC2_MAX_INSTANCES) {
return { error: `Choose 1 to ${EC2_MAX_INSTANCES} instances.` };
}
const tag = d.userTag.trim();
const tagError = userTagError(tag);
if (tagError) return { error: tagError };
return { compute: { instance_type: d.instanceType, max_instances: d.maxInstances, user_tag: tag } };
}
/** "m7g.large, up to 3 instances (triple-c-user alice)". */
export function computeLine(c: UserCompute): string {
const n = c.max_instances === 1 ? "1 instance" : `up to ${c.max_instances} instances`;
return `${c.instance_type}, ${n} (triple-c-user ${c.user_tag})`;
}
@@ -2,10 +2,12 @@
import type {
DeployProgress,
DetectedEnvironment,
Ec2InstanceTable,
EnvironmentInfo,
EnvironmentScan,
HostConnection,
TofuEvent,
UserCompute,
} from "../../../lib/types";
export const conn: HostConnection = {
@@ -65,3 +67,26 @@ export function tofuEvent(message: string, over: Partial<TofuEvent> = {}): TofuE
export function progress(over: Partial<DeployProgress>): DeployProgress {
return { op_id: "op1", state_key: "environments/dev/terraform.tfstate", events: [], done: false, cancellable: true, ...over };
}
/** A slice of `list_ec2_instance_types`'s answer: x86 and ARM rows. */
export const instanceTable: Ec2InstanceTable = {
default_type: "m7i.large",
types: [
{ instance_type: "m7i.large", arch: "x86_64", vcpu: 2, memory_mib: 8192 },
{ instance_type: "m7i.xlarge", arch: "x86_64", vcpu: 4, memory_mib: 16384 },
{ instance_type: "m7g.large", arch: "arm64", vcpu: 2, memory_mib: 8192 },
{ instance_type: "m7g.xlarge", arch: "arm64", vcpu: 4, memory_mib: 16384 },
],
};
export function compute(over: Partial<UserCompute> = {}): UserCompute {
const tag = over.user_tag ?? "alice";
return {
instance_type: "m7g.xlarge",
max_instances: 3,
user_tag: tag,
cluster_arn: `arn:aws:ecs:eu-west-1:123456789012:cluster/triple-c-dev--u-${tag}`,
capacity_provider: `triple-c-dev--u-${tag}`,
...over,
};
}
@@ -0,0 +1,45 @@
import { describe, it, expect, vi } from "vitest";
import { fireEvent, render, screen } from "@testing-library/react";
import SegmentedControl from "./SegmentedControl";
describe("SegmentedControl", () => {
it("a disabled segment can't be clicked and arrow keys skip it", () => {
const onChange = vi.fn();
render(
<SegmentedControl
label="Pick"
value="a"
onChange={onChange}
segments={[
{ value: "a", label: "A" },
{ value: "b", label: "B", disabled: true },
{ value: "c", label: "C" },
]}
/>,
);
expect(screen.getByRole("radio", { name: "B" })).toBeDisabled();
fireEvent.click(screen.getByRole("radio", { name: "B" }));
expect(onChange).not.toHaveBeenCalled();
fireEvent.keyDown(screen.getByRole("radiogroup"), { key: "ArrowRight" });
expect(onChange).toHaveBeenLastCalledWith("c");
fireEvent.keyDown(screen.getByRole("radiogroup"), { key: "ArrowLeft" });
expect(onChange).toHaveBeenLastCalledWith("c");
});
it("with every other segment disabled, the arrows stay put", () => {
const onChange = vi.fn();
render(
<SegmentedControl
label="Pick"
value="a"
onChange={onChange}
segments={[
{ value: "a", label: "A" },
{ value: "b", label: "B", disabled: true },
]}
/>,
);
fireEvent.keyDown(screen.getByRole("radiogroup"), { key: "ArrowRight" });
expect(onChange).not.toHaveBeenCalled();
});
});
+15 -4
View File
@@ -7,6 +7,8 @@ export interface Segment<T extends string> {
hint?: string;
/** Paint this segment with the caution treatment when selected. */
caution?: boolean;
/** This segment can't be chosen (say why next to the control). */
disabled?: boolean;
}
interface Props<T extends string> {
@@ -35,12 +37,21 @@ export default function SegmentedControl<T extends string>({
const move = (delta: number) => {
const index = segments.findIndex((s) => s.value === value);
const next = segments[(index + delta + segments.length) % segments.length];
// The next segment that can be chosen; none other than this one: stay.
let next: Segment<T> | undefined;
for (let step = 1; step < segments.length; step++) {
const s = segments[(index + delta * step + segments.length * step) % segments.length];
if (!s.disabled) {
next = s;
break;
}
}
if (!next) return;
onChange(next.value);
const target = next.value;
onChange(target);
requestAnimationFrame(() => {
groupRef.current
?.querySelector<HTMLElement>(`[data-segment="${next.value}"]`)
?.querySelector<HTMLElement>(`[data-segment="${target}"]`)
?.focus();
});
};
@@ -79,7 +90,7 @@ export default function SegmentedControl<T extends string>({
data-segment={segment.value}
aria-checked={selected}
tabIndex={selected ? 0 : -1}
disabled={disabled}
disabled={disabled || segment.disabled}
onClick={() => onChange(segment.value)}
className={`h-6 px-2.5 text-xs font-medium rounded-[4px] transition-colors disabled:text-[var(--text-disabled)] disabled:hover:bg-transparent ${cls}`}
>
+47 -1
View File
@@ -6,13 +6,15 @@ import {
hasWorkspaceFiles,
environmentSelectable,
gatewaySidecarProblem,
EC2_NO_SIDECAR,
launchTypePatch,
memoryError,
memoryOptions,
nearestMemory,
repoFolderFromUrl,
stopGuardNeeded,
} from "./awsWorkspace";
import { environment } from "../test/awsFixtures";
import { awsProject, environment } from "../test/awsFixtures";
// Shared with the Rust validator (`models/project.rs`).
import fargatePairs from "../test/fixtures/fargate-pairs.json";
@@ -105,3 +107,47 @@ describe("gatewaySidecarProblem", () => {
).toMatch(/1 vCPU and 2 GiB/);
});
});
describe("launchTypePatch (phase 7)", () => {
const base = () => {
const p = awsProject({ allow_docker_access: true, vpn_support_enabled: true });
return { ...p, aws: { ...p.aws!, gateway_sidecar: true, spot: true, cpu: 4096, memory_mib: 16384 } };
};
it("to EC2: keeps Docker/VPN, clears the sidecar and Spot, and fits the reservations to the instance", () => {
const patch = launchTypePatch(base(), "ec2", { cpu: 1920, memory_mib: 7373 });
expect(patch.allow_docker_access).toBeUndefined();
expect(patch.vpn_support_enabled).toBeUndefined();
expect(patch.aws).toMatchObject({ launch_type: "ec2", gateway_sidecar: false, spot: false, cpu: 1920, memory_mib: 7373 });
// Reservations that already fit stay; unknown limits change nothing.
const small = { ...base(), aws: { ...base().aws, cpu: 1024, memory_mib: 4096 } };
expect(launchTypePatch(small, "ec2", { cpu: 1920, memory_mib: 7373 }).aws).toMatchObject({ cpu: 1024, memory_mib: 4096 });
expect(launchTypePatch(base(), "ec2", null).aws).toMatchObject({ cpu: 4096, memory_mib: 16384 });
});
it("to Fargate: clears Docker/VPN and snaps the reservations to a Fargate pair", () => {
const p = base();
const ec2 = { ...p, aws: { ...p.aws, launch_type: "ec2" as const, cpu: 1920, memory_mib: 7373, spot: false } };
const patch = launchTypePatch(ec2, "fargate");
expect(patch).toMatchObject({ allow_docker_access: false, vpn_support_enabled: false });
expect(patch.aws).toMatchObject({ launch_type: "fargate", cpu: 1024, memory_mib: 7168 });
expect(memoryError(patch.aws!.cpu, patch.aws!.memory_mib)).toBeNull();
// Below the smallest Fargate size: the smallest.
const tiny = { ...ec2, aws: { ...ec2.aws, cpu: 100, memory_mib: 300 } };
expect(launchTypePatch(tiny, "fargate").aws).toMatchObject({ cpu: 256, memory_mib: 512 });
// A valid pair is kept.
const ok = { ...ec2, aws: { ...ec2.aws, cpu: 1024, memory_mib: 4096 } };
expect(launchTypePatch(ok, "fargate").aws).toMatchObject({ cpu: 1024, memory_mib: 4096 });
});
it("the sidecar is not available on EC2 (words it as Rust's EC2_NO_SIDECAR)", () => {
const p = { ...base(), backend: "open_ai_compatible" as const };
const ec2 = { ...p, aws: { ...p.aws, launch_type: "ec2" as const } };
const gw = { models: [{ name: "m", model_id: "x" }] } as never;
expect(gatewaySidecarProblem(ec2, gw)).toBe(EC2_NO_SIDECAR);
expect(EC2_NO_SIDECAR).toBe(
"The LLM gateway sidecar is not available on EC2 yet: turn it off, or run the project on Fargate",
);
expect(gatewaySidecarProblem(p, gw)).toBeNull();
});
});
+87 -1
View File
@@ -2,6 +2,7 @@ import type {
AwsWorkspaceConfig,
DetectedEnvironment,
GatewaySettings,
LaunchType,
Project,
RepoSource,
UnsavedWork,
@@ -11,7 +12,7 @@ import type {
export const NOT_ON_AWS = "Not available on AWS (Fargate)";
/** Q6: an AWS project's port mappings bind loopback only (Docker's bind 0.0.0.0). */
export const AWS_PORTS_LOCALHOST_ONLY = "Reachable only from this machine (localhost)";
/** R4: the AWS host forces Claude Code's sandbox off until it is verified on Fargate. */
/** The AWS host forces Claude Code's sandbox off: bubblewrap can't run in an ECS task (Fargate P9, EC2 E8). */
export const SANDBOX_OFF_ON_AWS = "Off on AWS for now";
export const EPHEMERAL_WARNING =
"Claude login and history are lost when the workspace stops.";
@@ -20,6 +21,13 @@ export function isAwsProject(p: Pick<Project, "host_id">): boolean {
return p.host_id != null;
}
/** Whether Claude Code's sandbox actually runs for `p`: never on AWS (the
* host forces it off there; Rust `aws_host::sandbox_allowed`), else the
* project's setting. */
export function sandboxEffective(p: Pick<Project, "host_id" | "sandbox_mode_enabled">): boolean {
return !isAwsProject(p) && p.sandbox_mode_enabled;
}
/** Fargate CPU sizes (units of 1/1024 vCPU). */
export const FARGATE_CPUS = [256, 512, 1024, 2048, 4096, 8192, 16384] as const;
@@ -77,6 +85,7 @@ export function formatMemory(mib: number): string {
/** Mirrors Rust `AwsWorkspaceConfig::default()`. */
export function defaultAwsWorkspace(): AwsWorkspaceConfig {
return {
launch_type: "fargate",
cpu: 1024,
memory_mib: 4096,
arch: "x86_64",
@@ -197,6 +206,7 @@ export function gatewaySidecarProblem(
project: Pick<Project, "backend" | "aws">,
gateway: Pick<GatewaySettings, "models"> | null,
): string | null {
if (project.aws?.launch_type === "ec2") return EC2_NO_SIDECAR;
if (project.backend !== "open_ai_compatible") {
return "The LLM gateway sidecar needs the OpenAI Compatible backend.";
}
@@ -237,3 +247,79 @@ export function localEndpointProblem(url: string, gatewayPort: number | null): s
}
return null;
}
/** What an EC2 instance leaves for a workspace's reservations (Rust `ec2_available`). */
export interface Ec2Limits {
cpu: number;
memory_mib: number;
}
/** The largest Fargate CPU size at most `cpu` (the smallest when below them all). */
function fargateCpuAtMost(cpu: number): number {
return [...FARGATE_CPUS].reverse().find((c) => c <= cpu) ?? FARGATE_CPUS[0];
}
/**
* The save patch for switching an AWS project's launch type — the one place
* the switch is made. Each side keeps only what Rust accepts there:
* - to Fargate: Docker access and VPN off (they need EC2), and the
* reservations snapped to a Fargate CPU/memory pair;
* - to EC2: the gateway sidecar and Spot off (neither is available on EC2),
* and the reservations lowered to what the instance leaves (`limits`, when
* known).
*/
export function launchTypePatch(
project: Project,
launch_type: LaunchType,
limits: Ec2Limits | null = null,
): Partial<Project> {
const aws: AwsWorkspaceConfig = { ...project.aws!, launch_type };
const patch: Partial<Project> = { aws };
if (launch_type === "ec2") {
aws.gateway_sidecar = false;
aws.spot = false;
if (limits) {
aws.cpu = Math.min(aws.cpu, limits.cpu);
aws.memory_mib = Math.min(aws.memory_mib, limits.memory_mib);
}
} else {
patch.allow_docker_access = false;
patch.vpn_support_enabled = false;
aws.cpu = fargateCpuAtMost(aws.cpu);
if (memoryError(aws.cpu, aws.memory_mib)) aws.memory_mib = nearestMemory(aws.cpu, aws.memory_mib);
}
return patch;
}
/**
* What a launch-type switch's [`launchTypePatch`] changed beyond the launch
* type, as one short line for Host settings (null: nothing else changed).
*/
export function launchTypeChanges(before: Project, patch: Partial<Project>): string | null {
const was = before.aws!;
const now = patch.aws!;
const changes: string[] = [];
if (now.cpu !== was.cpu) changes.push(`CPU ${was.cpu} → ${now.cpu} units`);
if (now.memory_mib !== was.memory_mib) changes.push(`memory ${was.memory_mib} → ${now.memory_mib} MiB`);
if (was.spot && !now.spot) changes.push("Spot turned off");
if (was.gateway_sidecar && !now.gateway_sidecar) changes.push("LLM gateway sidecar turned off");
if (before.allow_docker_access && patch.allow_docker_access === false) changes.push("Docker access turned off");
if (before.vpn_support_enabled && patch.vpn_support_enabled === false) changes.push("VPN support turned off");
if (changes.length === 0) return null;
const to = now.launch_type === "ec2" ? "EC2" : "Fargate";
return `Switched to ${to}: ${changes.join(", ")}.`;
}
/** Mirrors Rust `aws::workspace::EC2_NO_SIDECAR`. */
export const EC2_NO_SIDECAR =
"The LLM gateway sidecar is not available on EC2 yet: turn it off, or run the project on Fargate";
/** Host settings' notes for an EC2 project (phase 7). */
export const EC2_SCALE_TO_ZERO_NOTE =
"Your instances scale to zero when no workspace is running, so the first start after idle boots an instance — about 3 minutes.";
export const EC2_DOCKER_ACCESS_NOTE =
"Docker access mounts your instance's Docker socket: that is root on your own instance, which nobody else's workspace runs on.";
/** RuntimeSection's VPN hint for an EC2 project: an ECS task, not a container the app recreates. */
export const EC2_VPN_HINT =
"Grants NET_ADMIN, the /dev/net/tun device and src_valid_mark so a VPN client (PIA, WireGuard, OpenVPN) can build a tunnel inside the workspace, an ECS task on your own EC2 instance. Anything in the workspace can then reconfigure the task's own network stack; the instance's is untouched. Applies on the project's next start.";
export const EC2_NO_CAPACITY = "Ask an admin to add EC2 capacity to your onboarding";
+79
View File
@@ -0,0 +1,79 @@
import { describe, it, expect, vi, beforeEach } from "vitest";
import type { Ec2Capacity } from "./types";
const EC2: Ec2Capacity = {
state: "ec2",
compute: {
instance_type: "m7i.large",
max_instances: 1,
user_tag: "alice",
cluster_arn: "arn:aws:ecs:eu-west-1:1:cluster/triple-c-dev--u-alice",
capacity_provider: "triple-c-dev--u-alice",
},
available_cpu: 1920,
available_memory_mib: 7373,
};
const awsEc2Capacity = vi.fn(async (_id: string): Promise<Ec2Capacity> => EC2);
vi.mock("./tauri-commands", () => ({
awsEc2Capacity: (id: string) => awsEc2Capacity(id),
}));
import {
EC2_CAPACITY_SHORT_TTL_MS,
EC2_CAPACITY_TTL_MS,
clearEc2CapacityCache,
ec2Capacity,
} from "./ec2Capacity";
describe("ec2Capacity", () => {
beforeEach(() => {
awsEc2Capacity.mockClear();
clearEc2CapacityCache();
});
it("asks once per connection, region and environment until the TTL passes", async () => {
expect(await ec2Capacity("p1", "c1", "eu-west-1", "dev", 0)).toEqual(EC2);
expect(await ec2Capacity("p2", "c1", "eu-west-1", "dev", 1000)).toEqual(EC2);
expect(awsEc2Capacity).toHaveBeenCalledTimes(1);
await ec2Capacity("p1", "c1", "eu-west-1", "prod", 1000);
expect(awsEc2Capacity).toHaveBeenCalledTimes(2);
await ec2Capacity("p1", "c1", "eu-west-1", "dev", EC2_CAPACITY_TTL_MS);
expect(awsEc2Capacity).toHaveBeenCalledTimes(3);
});
it("keeps none and unusable for a minute only, ec2 for ten", async () => {
expect(EC2_CAPACITY_SHORT_TTL_MS).toBe(60 * 1000);
expect(EC2_CAPACITY_TTL_MS).toBe(10 * 60 * 1000);
for (const answer of [{ state: "none" }, { state: "unusable", reason: "x" }] as Ec2Capacity[]) {
clearEc2CapacityCache();
awsEc2Capacity.mockClear();
awsEc2Capacity.mockResolvedValueOnce(answer);
expect(await ec2Capacity("p1", "c1", "eu-west-1", "dev", 0)).toEqual(answer);
await ec2Capacity("p1", "c1", "eu-west-1", "dev", EC2_CAPACITY_SHORT_TTL_MS - 1);
expect(awsEc2Capacity).toHaveBeenCalledTimes(1);
expect(await ec2Capacity("p1", "c1", "eu-west-1", "dev", EC2_CAPACITY_SHORT_TTL_MS)).toEqual(EC2);
expect(awsEc2Capacity).toHaveBeenCalledTimes(2);
// The ec2 answer that replaced it is kept for the long TTL.
await ec2Capacity("p1", "c1", "eu-west-1", "dev", EC2_CAPACITY_TTL_MS);
expect(awsEc2Capacity).toHaveBeenCalledTimes(2);
}
});
it("keeps none, but not unknown or a rejection", async () => {
awsEc2Capacity.mockResolvedValueOnce({ state: "none" });
expect(await ec2Capacity("p1", "c1", "eu-west-1", "dev", 0)).toEqual({ state: "none" });
await ec2Capacity("p1", "c1", "eu-west-1", "dev", 1);
expect(awsEc2Capacity).toHaveBeenCalledTimes(1);
clearEc2CapacityCache();
awsEc2Capacity.mockResolvedValueOnce({ state: "unknown" });
expect(await ec2Capacity("p1", "c1", "eu-west-1", "dev", 0)).toEqual({ state: "unknown" });
expect(await ec2Capacity("p1", "c1", "eu-west-1", "dev", 1)).toEqual(EC2);
expect(awsEc2Capacity).toHaveBeenCalledTimes(3);
awsEc2Capacity.mockRejectedValueOnce(new Error("x"));
clearEc2CapacityCache();
await expect(ec2Capacity("p1", "c1", "eu-west-1", "dev", 0)).rejects.toThrow("x");
expect(await ec2Capacity("p1", "c1", "eu-west-1", "dev", 1)).toEqual(EC2);
});
});
+82
View File
@@ -0,0 +1,82 @@
import { useEffect, useState } from "react";
import * as commands from "./tauri-commands";
import type { Ec2Capacity, Project } from "./types";
/** How long a user's EC2 capacity is reused before AWS is asked again. */
export const EC2_CAPACITY_TTL_MS = 10 * 60 * 1000;
/** The same for any other answer (`none`, `unusable`): an admin adding or
* fixing the capacity shows up within a minute. */
export const EC2_CAPACITY_SHORT_TTL_MS = 60 * 1000;
interface Entry {
at: number;
ttl: number;
value: Promise<Ec2Capacity>;
}
/** Keyed by connection, region and environment: the record is the caller's
* in that environment, so every project placed there shares one read. */
const cache = new Map<string, Entry>();
/**
* `aws_ec2_capacity`, cached per (connection, region, environment) for
* [`EC2_CAPACITY_TTL_MS`] when the answer is `ec2`, and
* [`EC2_CAPACITY_SHORT_TTL_MS`] for `none` or `unusable`. A read in flight
* is shared; an `unknown` answer (offline, slow) or a rejection is not
* kept, so the next mount asks again.
*/
export function ec2Capacity(
projectId: string,
connectionId: string,
region: string,
environment: string,
now: number = Date.now(),
): Promise<Ec2Capacity> {
const key = JSON.stringify([connectionId, region, environment]);
const hit = cache.get(key);
if (hit && now - hit.at < hit.ttl) return hit.value;
const value = Promise.resolve().then(() => commands.awsEc2Capacity(projectId));
const entry: Entry = { at: now, ttl: EC2_CAPACITY_TTL_MS, value };
cache.set(key, entry);
const forget = () => {
if (cache.get(key) === entry) cache.delete(key);
};
value.then((c) => {
if (c.state === "unknown") forget();
else if (c.state !== "ec2") entry.ttl = EC2_CAPACITY_SHORT_TTL_MS;
}, forget);
return value;
}
/** After an onboarding apply changed the user's EC2 capacity, and for tests. */
export function clearEc2CapacityCache() {
cache.clear();
}
/** The project's EC2 capacity (null while it is read); only asked when `enabled`. */
export function useEc2Capacity(
project: Pick<Project, "id" | "host_id" | "aws_placement">,
enabled = true,
): Ec2Capacity | null {
const [capacity, setCapacity] = useState<Ec2Capacity | null>(null);
const region = project.aws_placement?.region;
const environment = project.aws_placement?.environment;
useEffect(() => {
setCapacity(null);
if (!enabled) return;
if (!project.host_id || !region || !environment) {
setCapacity({ state: "unknown" });
return;
}
let cancelled = false;
ec2Capacity(project.id, project.host_id, region, environment)
.catch((): Ec2Capacity => ({ state: "unknown" }))
.then((c) => {
if (!cancelled) setCapacity(c);
});
return () => {
cancelled = true;
};
}, [enabled, project.id, project.host_id, region, environment]);
return capacity;
}
+31 -8
View File
@@ -1,5 +1,5 @@
import { invoke } from "@tauri-apps/api/core";
import type { Project, ProjectPath, NewProjectHost, RemoveProjectOptions, WipChoice, UnsavedWork, WorkspaceStatus, ProjectRemovalReport, ProjectResetOutcome, ContainerInfo, AppSettings, SettingsImportPreview, SettingsImportOutcome, UpdateInfo, ImageUpdateInfo, FileEntry, FileContents, WebTerminalInfo, SttStatus, GatewayStatus, InstallOptions, ClaudeSession, ContainerCapabilities, ScheduledTask, ScheduledTaskInput, SchedulerNotification, AuthBridgeStatus, BrowserViewStatus, BrowserViewPopoutState, BrowserPageState, PlaywrightDetection, BrowserSetupOutcome, BrowserInstallTarget, ContainerStaleness, MigrationOptions, MigrationReport, MigrationState, ClearTokenOutcome, CaCertInfo, UploadOutcome, Note, ViewerFile, ViewerPoll, ViewerSaved, ViewerState, FileDiff, InstallScope, ItemUpdate, Marketplace, MarketplaceAccount, MarketplaceItemRef, MarketplaceSnapshot, ProjectSyncResult, SyncReport, HostConnection, CallerIdentity, PermissionReport, PolicyFile, HourlyEstimate, AwsWorkspaceConfig, Tier, EnvironmentScan, NetworkChoice, ModuleKind } from "./types";
import type { Project, ProjectPath, NewProjectHost, RemoveProjectOptions, WipChoice, UnsavedWork, WorkspaceStatus, ProjectRemovalReport, ProjectResetOutcome, ContainerInfo, AppSettings, SettingsImportPreview, SettingsImportOutcome, UpdateInfo, ImageUpdateInfo, FileEntry, FileContents, WebTerminalInfo, SttStatus, GatewayStatus, InstallOptions, ClaudeSession, ContainerCapabilities, ScheduledTask, ScheduledTaskInput, SchedulerNotification, AuthBridgeStatus, BrowserViewStatus, BrowserViewPopoutState, BrowserPageState, PlaywrightDetection, BrowserSetupOutcome, BrowserInstallTarget, ContainerStaleness, MigrationOptions, MigrationReport, MigrationState, ClearTokenOutcome, CaCertInfo, UploadOutcome, Note, ViewerFile, ViewerPoll, ViewerSaved, ViewerState, FileDiff, InstallScope, ItemUpdate, Marketplace, MarketplaceAccount, MarketplaceItemRef, MarketplaceSnapshot, ProjectSyncResult, SyncReport, HostConnection, CallerIdentity, PermissionReport, PolicyFile, HourlyEstimate, AwsWorkspaceConfig, Tier, EnvironmentScan, NetworkChoice, ModuleKind, Ec2Choice, Ec2Compute, Ec2InstanceTable, Ec2Capacity, SavedBundle } from "./types";
// Docker
export const checkDocker = () => invoke<boolean>("check_docker");
@@ -46,6 +46,11 @@ export const awsWorkspaceStatus = (projectId: string) =>
* null when it can't be read. Never rejects for a missing project. */
export const awsEnvironmentMaxHours = (projectId: string) =>
invoke<number | null>("aws_environment_max_hours", { projectId });
export const awsEc2Capacity = (projectId: string) =>
invoke<Ec2Capacity>("aws_ec2_capacity", { projectId });
/** Rust `ec2_fit_problem`: why the reservations don't fit `instanceType`, or null. */
export const ec2FitProblem = (config: AwsWorkspaceConfig, instanceType: string) =>
invoke<string | null>("ec2_fit_problem", { config, instanceType });
export const removeAwsResources = (projectId: string, discardData: boolean) =>
invoke<Project>("remove_aws_resources", { projectId, discardData });
export const rebuildProjectContainer = (projectId: string) =>
@@ -112,8 +117,13 @@ export const awsPolicyDocument = (tier: Tier) =>
/** The estimated hourly cost of `config` on Fargate in `region`, signed with the project's
* connection. Never rejects for AWS reasons: `null` means "Estimate unavailable" (denied, offline,
* no connection). Rates are cached a day on disk. */
export const awsCostEstimate = (connectionId: string, region: string, config: AwsWorkspaceConfig) =>
invoke<HourlyEstimate | null>("aws_cost_estimate", { connectionId, region, config });
/** `instanceType`: an EC2 project's (the user's, from `awsEc2Capacity`); priced as that instance. */
export const awsCostEstimate = (
connectionId: string,
region: string,
config: AwsWorkspaceConfig,
instanceType: string | null = null,
) => invoke<HourlyEstimate | null>("aws_cost_estimate", { connectionId, region, config, instanceType });
// AWS environments (OpenTofu). Reject with an `AwsError`. Starting commands (and follow) resolve
// to an op id; progress arrives on the `AWS_DEPLOY_PROGRESS_EVENT` event (`DeployProgress`, in
@@ -163,18 +173,30 @@ export const cancelAwsDeployment = (connectionId: string, region: string, stateK
/** Push `errored.tfstate` (a failed run's unsaved state) back to the bucket; resolves to an op id like a deploy. */
export const pushAwsSavedState = (connectionId: string, region: string, stateKey: string) =>
invoke<string>("push_aws_saved_state", { connectionId, region, stateKey });
export const onboardAwsUser = (connectionId: string, region: string, environment: string) =>
invoke<string>("onboard_aws_user", { connectionId, region, environment });
/** Omit `ec2` to keep the caller's recorded EC2 capacity (none without a record). */
export const onboardAwsUser = (connectionId: string, region: string, environment: string, ec2?: Ec2Choice) =>
invoke<string>(
"onboard_aws_user",
ec2 === undefined ? { connectionId, region, environment } : { connectionId, region, environment, ec2 },
);
export const listEc2InstanceTypes = () => invoke<Ec2InstanceTable>("list_ec2_instance_types");
export const deployAwsPermissionsBoundary = (connectionId: string, region: string) =>
invoke<string>("deploy_aws_permissions_boundary", { connectionId, region });
/** Opens a folder picker in Rust; resolves to the folder written, or null if dismissed. */
/** Opens a folder picker in Rust; resolves to the folder written (and, for an onboarding with EC2, a warning when its cluster name looks taken), or null if dismissed. */
export const saveAwsAdminBundle = (
connectionId: string,
region: string,
kind: ModuleKind,
options: { environment?: string; ownerId?: string; network?: NetworkChoice; maxWorkspaceHours?: number } = {},
options: {
environment?: string;
ownerId?: string;
network?: NetworkChoice;
maxWorkspaceHours?: number;
/** An onboarding bundle's EC2 capacity; null = none. */
ec2?: Ec2Compute | null;
} = {},
) =>
invoke<string | null>("save_aws_admin_bundle", {
invoke<SavedBundle | null>("save_aws_admin_bundle", {
connectionId,
region,
kind,
@@ -182,6 +204,7 @@ export const saveAwsAdminBundle = (
ownerId: options.ownerId ?? null,
network: options.network ?? null,
...(options.maxWorkspaceHours === undefined ? {} : { maxWorkspaceHours: options.maxWorkspaceHours }),
...(options.ec2 === undefined ? {} : { ec2: options.ec2 }),
});
/** Check a corporate CA path and report what would be installed. Never
* rejects for a bad path — the reason comes back in `error`. */
+69 -3
View File
@@ -92,7 +92,11 @@ export interface RepoSource {
folder: string;
}
export type LaunchType = "fargate" | "ec2";
export interface AwsWorkspaceConfig {
/** Absent in old data = fargate. */
launch_type: LaunchType;
cpu: number;
memory_mib: number;
arch: Arch;
@@ -1276,6 +1280,12 @@ export interface HourlyEstimate {
note_spot: boolean;
/** Hybrid or EFS storage: EFS is billed per GB-month, not in this figure. */
efs_billed_separately: boolean;
/** EC2 (E10): the user's instance's on-demand price, whatever the project
* reserves of it; instances scale to zero when no workspace is running. */
ec2_instance: boolean;
/** EC2: the figure includes the instance's 50 GiB gp3 root volume (its
* price could be read). A public IPv4 address is never in it. */
root_volume_included: boolean;
}
export type Verdict = "allowed" | "denied" | "unverified";
@@ -1352,12 +1362,68 @@ export type EnvironmentState =
/** `push_state`: push the state a failed run on this machine couldn't save (`errored.tfstate`) back to the bucket. */
export type EnvironmentAction = "join" | "update" | "repair" | "destroy" | "follow" | "unlock" | "push_state";
/** `incomplete`: the last onboarding run didn't finish (Join re-applies and finishes it). */
/** A user's EC2 capacity in an environment: their validated user record (`aws::environments::UserCompute`). */
export interface UserCompute {
instance_type: string;
max_instances: number;
/** The `triple-c-user` value the cluster is named after. */
user_tag: string;
cluster_arn: string;
capacity_provider: string;
}
/** `incomplete`: the last onboarding run didn't finish (Join re-applies and finishes it). `ec2`: the caller's recorded EC2 capacity, null for none. */
export type OnboardingState =
| { state: "joined" }
| { state: "joined"; ec2: UserCompute | null }
| { state: "not_joined" }
| { state: "in_progress"; lock: LockInfo | null }
| { state: "incomplete"; lock: LockInfo | null };
| { state: "incomplete"; lock: LockInfo | null; ec2: UserCompute | null };
/** EC2 capacity as the onboarding module takes it (`aws::tofu_modules::Ec2Compute`). */
export interface Ec2Compute {
instance_type: string;
/** 1 to 4. */
max_instances: number;
/** The user's `triple-c-user` sign-in tag value. */
user_tag: string;
}
/** An onboarding's EC2 choice (`environment_commands::Ec2Choice`). */
export type Ec2Choice = { kind: "none" } | ({ kind: "ec2" } & Ec2Compute);
/** `host::aws_host::Ec2Capacity`: the user's EC2 capacity in an AWS project's
* environment, for Host settings (phase 7). `available_*` are what the
* instance leaves for one workspace (null for a type the app doesn't know). */
export type Ec2Capacity =
| {
state: "ec2";
compute: UserCompute;
available_cpu: number | null;
available_memory_mib: number | null;
}
| { state: "none" }
| { state: "unusable"; reason: string }
| { state: "unknown" };
/** One row of the app's built-in instance table (`models::project::Ec2InstanceType`). */
export interface Ec2InstanceType {
instance_type: string;
arch: Arch;
vcpu: number;
memory_mib: number;
}
/** `environment_commands::SavedBundle`. */
export interface SavedBundle {
folder: string;
/** An onboarding bundle's EC2 name looks taken (a warning; the README has the admin check). */
warning: string | null;
}
export interface Ec2InstanceTable {
default_type: string;
types: Ec2InstanceType[];
}
export type Knowable = "yes" | "no" | "unknown";
+1 -1
View File
@@ -69,7 +69,7 @@ export function environment(
state: { state } as DetectedEnvironment["state"],
actions: [],
info: null,
onboarding: joined ? { state: "joined" } : { state: "not_joined" },
onboarding: joined ? { state: "joined", ec2: null } : { state: "not_joined" },
onboarding_actions: [],
onboarded_users: 1,
};
+260 -19
View File
@@ -10,17 +10,63 @@ tier the signed-in identity has under **Settings → Hosts → Check**.
| `policies/workspace-user-tasks.json` | The **workspace user** tier, part 2: register and tag **their own** ECS task definitions, read **their own** secrets, and read Fargate prices for the cost estimate (`workspace-user.json` has no room left) |
| `policies/environment-admin.json` | The **environment admin** tier, part 1: apply and destroy the environment and onboarding modules (`terraform/`) with OpenTofu, and keep their state |
| `policies/environment-admin-network.json` | The **environment admin** tier, part 2: the network those modules create (VPC, subnets, gateways, routes, security groups) |
| `policies/environment-admin-compute.json` | The **environment admin** tier, part 3: per-user EC2 capacity the onboarding module creates (phase 7): capacity providers, launch templates, the launch rules, the instance role passed to EC2, the ECS-optimized AMI parameter |
| `policies/environment-admin-scaling.json` | The **environment admin** tier, part 4: the per-user Auto Scaling groups, instance profiles and the Auto Scaling/ECS service-linked roles |
| `policies/permissions-boundary.json` | The **permissions boundary** every Triple-C role must carry. It is never attached to a person |
The workspace tier and the admin tier are each **two** files, because one managed policy holds
only 6,144 characters; both files of a tier are required, and the app's **Copy** lists every
file of a tier. An environment admin attaches all four tier files (both workspace user files and
both admin files): the admin files add environment management on top of everything a workspace
user can do. Explicit Deny statements in any file apply to anyone
The workspace tier is **two** files and the admin tier **four**, because one managed policy
holds only 6,144 characters; every file of a tier is required, and the app's **Copy** lists every
file of a tier. An environment admin attaches all six tier files (both workspace user files and
the four admin files): the admin files add environment management on top of everything a
workspace user can do. Explicit Deny statements in any file apply to anyone
who has it attached.
## Workspace user
**EC2 users need a `triple-c-user` principal tag.** An admin gives each user who gets EC2
capacity a unique tag value of 1 to 32 letters, digits and single hyphens (not starting or ending
with one; `validate_ec2_user_tag` in `tofu_modules.rs`):
- **IAM Identity Center:** Settings → *Attributes for access control* → add key `triple-c-user`
with a value such as `${path:userName}`, as long as every user's value is valid as above and
unique (sanitise in the identity source if usernames hold `@`, `.` or `_`, e.g. a dedicated
user attribute). The permission set's sessions then carry `aws:PrincipalTag/triple-c-user`.
- **IAM users or roles:** tag the user (or the role they assume) `triple-c-user = <value>`.
The same value goes into the user's onboarding (`ec2_user_tag`), which names their cluster.
A user without the tag can't run anything on a per-user EC2 cluster (Fargate is unaffected), and
a user whose tag differs from their onboarding's can't run on the cluster it created.
**Setting up EC2 capacity (admin, once per user).** Phase 7 is opt-in per onboarding. In order:
1. Re-apply the **boundary** (version 2; see "One-time setup: the boundary policy"). Without it
the instance role's ECS agent calls are denied and no instance registers.
2. Attach all four admin files to whoever onboards (`environment-admin-compute.json` and
`-scaling.json` are the phase 7 ones; measured sizes under "Attaching the tier policies").
3. Give the user the `triple-c-user` principal tag (above). Before an admin-bundle apply, check
that **no cluster or Auto Scaling group named `triple-c-<env>--u-<tag>` exists**: the app's own
pre-apply check refuses a name that belongs to another onboarding (by the resources'
`triple-c:owner` tag, with the user record as fallback). A bundle run by hand has no such check
in the module (it would take over a cluster of that name), so the bundle's `README.txt` has an
"EC2 capacity" step with the three commands; run them right before `tofu apply`.
4. Onboard with `ec2_instance_type` (supported types below), `ec2_max_instances` (1 to 4) and
`ec2_user_tag` (the same value as the tag). Idle capacity scales to zero (min 0), so the
first start after about 17 minutes of idleness takes 3.5 to 4 minutes.
**Caveats of `triple-c-user`.** It is the isolation boundary between EC2 users, so treat it as one.
- **Session tags override role tags.** If a role's trust policy lets the caller pass session tags
(`sts:TagSession`), the caller can set `triple-c-user` to another user's value and run tasks on
that user's cluster. Trust policies of roles that carry the tag must not allow `sts:TagSession`
(or must restrict it with `aws:TagKeys` so `triple-c-user` can't be passed).
- **The identity-source attribute must not be user-editable.** For IAM Identity Center the
attribute behind `triple-c-user` must be one only an admin can change (not a self-service
profile field); otherwise a user can claim another's value. Likewise no workspace user may be
able to `iam:TagUser`/`iam:TagRole` the tagged principal.
- **A role shared by several people is one EC2 identity.** The tag belongs to the role, so
everyone who assumes it has the same cluster and can use each other's instance (and, with
Docker access, its root). Give each person their own tagged role or permission set.
Every per-user resource is tagged `triple-c:owner = ${aws:userid}`. That value is the `UserId`
printed by `aws sts get-caller-identity`. The policy uses this tag so that a user can only do the
following:
@@ -28,6 +74,13 @@ following:
- **Run tasks.** The task definition must be a `triple-c-*` one tagged with the user's own owner
ID, and the cluster must be a `triple-c-*` one. The new task is tagged with the user's owner ID.
The user can pass only `triple-c-*` roles tagged with their owner ID, and only to ECS tasks.
**Per-user EC2 clusters** (phase 7) are `triple-c-<env>--u-<user tag>`, and two Denies in
`workspace-user-tasks.json` keep everyone else off them: `RunTask`/`StartTask` on a
`cluster/triple-c-*--u-*` cluster is denied unless it is
`cluster/triple-c-*--u-${aws:PrincipalTag/triple-c-user}`, and denied outright when the caller
has no `triple-c-user` principal tag (so a missing tag never falls through, however IAM resolves
the variable). Environment names and user tags can't contain `--`, so an environment's (Fargate)
cluster never matches the pattern and one user's pattern never matches another's cluster.
- **Register their own task definitions** (`workspace-user-tasks.json`). `ecs:RegisterTaskDefinition`
on `triple-c-*` families needs `triple-c:owner = ${aws:userid}` in the request and only the three
ordinary tag keys. `ecs:TagResource` on a task definition is allowed only at registration
@@ -137,11 +190,26 @@ and nothing runs with more than the admin holds. It covers:
- the per-account state bucket `triple-c-tfstate-<account>-<region>` (created by the app, tagged
`triple-c:component=state`, then read and written by OpenTofu's S3 backend, lock files included);
- `triple-c-*` IAM roles, and reading the `triple-c-permissions-boundary` policy;
- ECS clusters (Fargate and Fargate Spot only) and task definitions;
- ECS clusters (the environment's: Fargate and Fargate Spot; a user's: their own EC2 capacity
provider) and task definitions;
- the network: VPCs, subnets, an internet gateway, route tables and routes, an Elastic IP and NAT
gateway, security groups and their rules (`environment-admin-network.json`);
- EFS file systems, mount targets and file-system policies, and KMS keys and aliases;
- `/triple-c/*` log groups and `/triple-c/*` SSM parameters.
- `/triple-c/*` log groups and `/triple-c/*` SSM parameters (including each user's compute record
`/triple-c/users/<env>/<owner key>`);
- per-user EC2 capacity (`environment-admin-compute.json` and `-scaling.json`, phase 7): the
user's own `triple-c-<env>--u-<user tag>` cluster, capacity provider (`triple-c-*--u-*`), a
component-tagged launch template and `triple-c-*--u-*` Auto Scaling group (pinned to a template version, no load balancer, supported
instance types only), the `triple-c-*` instance profile, passing `triple-c-*-instance` roles to
EC2, reading the ECS-optimized AMI's public SSM parameter, and the Auto Scaling and ECS
service-linked roles.
**Supported instance types.** Launches and groups are limited to the types the app sizes
(`ec2_instance_spec` in `models/project.rs`, pinned to the policy by a test): `m7i`, `m7g`, `c7i`,
`c7g`, `r7i`, `r7g` in `large` to `4xlarge`, and `t3`, `t4g` in `large` to `2xlarge`. No GPU,
accelerator or metal types. An onboarding with any other type is denied at apply; supporting one
means adding it to that table and to the three lists (`ec2:InstanceType` in the compute file,
`autoscaling:InstanceTypes` twice in the scaling file) together.
**Clearing workspace task definitions.** Removing a project leaves its revisions registered
(family `triple-c-<env>-<owner key>-<project id, 12 chars>`). The admin tier may deregister any
@@ -156,9 +224,11 @@ Deregistered revisions become INACTIVE; ECS keeps them until they are deleted wi
`ecs:DeleteTaskDefinitions`, which the admin tier does not grant (an account administrator can).
A task still running from a revision keeps running.
Not in v1, and so not in the policy: ECR (tasks pull from GHCR), custom ECS capacity providers,
and passing roles to EC2 (`iam:PassRole` is to `ecs-tasks.amazonaws.com` only). Phase 7's EC2
capacity adds them back.
Not in the policy: ECR (tasks pull from GHCR). Phase 7's EC2 capacity added custom ECS capacity
providers and passing roles to EC2. The `triple-c-*-instance` name on that PassRole keeps a
user's task and execution roles from being handed to an instance by mistake; it is **not** a
fence against the admin, who can create a bounded role under any `triple-c-*` name. What fences
an instance role is the permissions boundary, as for every Triple-C role.
### What each module resource needs
@@ -192,6 +262,16 @@ the proof; an action it shows missing is added to the table, the policy, the che
| `aws_iam_role` | `iam:CreateRole` (with the boundary), `TagRole`, `UntagRole`, `GetRole`, `ListRolePolicies`, `ListAttachedRolePolicies`, `ListInstanceProfilesForRole`, `UpdateRole`, `UpdateRoleDescription`, `DeleteRole` (`triple-c-*`) |
| `aws_iam_role_policy` | `iam:PutRolePolicy`, `DeleteRolePolicy` (with the boundary), `GetRolePolicy` (the role; untaggable) |
| data sources | `sts:GetCallerIdentity`, `ec2:DescribeAvailabilityZones`, `ec2:DescribeSubnets`, `ssm:GetParameter` |
| *Phase 7, per-user EC2 (onboarding module, expected; `EC2_COMPUTE_CALLS` until the resources exist)* | |
| `aws_ecs_cluster` (the user's) | as the environment's cluster, on `cluster/triple-c-<env>--u-<user tag>` (the capacity provider, Auto Scaling group and launch template use the same name) |
| `aws_ecs_capacity_provider` | `ecs:CreateCapacityProvider`, `TagResource` (at create), `ListTagsForResource`, `UpdateCapacityProvider`, `UntagResource`, `DeleteCapacityProvider` (`capacity-provider/triple-c-*--u-*`), `DescribeCapacityProviders` (`*`); `iam:CreateServiceLinkedRole` for `ecs.amazonaws.com` (managed scaling runs as ECS's service-linked role) |
| `aws_ecs_cluster_capacity_providers` (the user's) | `ecs:PutClusterCapacityProviders` (the cluster and, expected, the capacity provider) |
| `aws_launch_template` | `ec2:CreateLaunchTemplate` (request tag), `CreateTags` (at create), `CreateLaunchTemplateVersion`, `CreateTags`/`DeleteTags`, `DeleteLaunchTemplate` (tagged), `Describe*`. Never `ModifyLaunchTemplate`: the default version never moves (the module must not set `update_default_version` or `default_version`) |
| `aws_autoscaling_group` | `autoscaling:CreateAutoScalingGroup` (`triple-c-*--u-*`, request tag, `autoscaling:LaunchTemplateVersionSpecified = true`, no `TargetGroupARNs`/`LoadBalancerNames`, `autoscaling:InstanceTypes` supported when present — a template-only create sends none, verified live, so the type is held by the launch check below), `UpdateAutoScalingGroup` (tagged; the same, except that an update which leaves the template alone passes the version check), `SetInstanceProtection`, `CreateOrUpdateTags`/`DeleteTags`, `DeleteAutoScalingGroup` (tagged), `Describe*` (`*`); `iam:CreateServiceLinkedRole` for `autoscaling.amazonaws.com`. Creating a group checks that the caller could launch its template: `ec2:RunInstances` on the tagged launch template; on the instance and its volume only with the component request tag, `ec2:LaunchTemplate` set and (instance) a supported `ec2:InstanceType` (required: a launch whose type IAM doesn't see is denied); on the network interface and subnet with `ec2:LaunchTemplate`; on the security group only the template's own (`ec2:IsLaunchTemplateResource`) component-tagged group; on images only Amazon's, named by the template; `ec2:CreateTags` at `RunInstances`; `iam:PassRole` of the instance role to `ec2.amazonaws.com`. ASG tags are **not** the provider's `default_tags`: the module must set the component tag on the group itself |
| `aws_iam_role` (instance), `aws_iam_role_policy` | as the other per-user roles (bounded), on `triple-c-<env>-<owner key>-instance`, tagged `triple-c:cluster = triple-c-<env>--u-<user tag>` (`iam:TagRole`) |
| `aws_iam_instance_profile` | `iam:CreateInstanceProfile`, `TagInstanceProfile`, `AddRoleToInstanceProfile` (expected to need `iam:PassRole` of the role too), `GetInstanceProfile`, `ListInstanceProfileTags`, `UntagInstanceProfile`, `RemoveRoleFromInstanceProfile`, `DeleteInstanceProfile` (`instance-profile/triple-c-*`) |
| `data.aws_ssm_parameter` (AMI) | `ssm:GetParameter`/`GetParameters` on the public `parameter/aws/service/ecs/optimized-ami/*` |
| `aws_ssm_parameter` (user record) | as the environment record, on `/triple-c/users/<env>/<owner key>` |
| S3 backend | `s3:ListBucket` (the state bucket), `GetObject`/`PutObject`/`DeleteObject` (`triple-c-tfstate-*/*`, the `.tflock` objects included), in the caller's own account |
The app also creates the state bucket: `s3:CreateBucket` with a `triple-c:component` request tag
@@ -221,7 +301,21 @@ the role carries the `triple-c-permissions-boundary` managed policy as its permi
(tagged `user-onboarding` by their onboarding module); the file system's own policy admits
only its owner's task role, through an access point;
- `sts:GetCallerIdentity`;
- invoking Bedrock models.
- invoking Bedrock models;
- for a user's EC2 instance role (phase 7, E4): the ECS agent's calls
(`RegisterContainerInstance`, `DeregisterContainerInstance`, `Poll`, `StartTelemetrySession`,
`Submit*`, `UpdateContainerInstancesState`) only on **the role's own cluster**, the one its
`triple-c:cluster` tag names (`cluster/${aws:PrincipalTag/triple-c:cluster}` and that
cluster's container instances; the tag must match `triple-c-*--u-*`, so never the environment's
shared cluster), `ecs:TagResource` only on
such a container instance as it registers (`ecs:CreateAction = RegisterContainerInstance`),
and `ecs:DiscoverPollEndpoint`. A user with Docker access is root on their own instance and
so holds its role's credentials; the tag keeps them from registering that instance into
another user's cluster (and so having that user's tasks placed on it). An untagged instance
role can do nothing with ECS. The admin sets the tag when it creates the role (the admin
tier already tags `triple-c-*` roles freely, and a workspace user cannot tag roles at all).
**No Session Manager on the instance** (plan E4 amended in fix round 1): nobody is granted
host sessions, and ECS Exec into an EC2 task uses the task role, as on Fargate.
- The boundary grants nothing in IAM.
- Explicit Deny statements stop the admin from:
- removing the boundary from a `triple-c-*` role;
@@ -239,11 +333,19 @@ created only *inside* a Triple-C (tagged) VPC, subnet and address, and an intern
attached, a route added or a route table associated only on tagged ones, so the admin cannot
re-route a VPC that isn't Triple-C's. Tagging is allowed only as part of the creating
call (`ec2:CreateAction`, `elasticfilesystem:CreateAction`), or on a resource that is already
Triple-C's. Resources that cannot carry a tag are authorized through one that does: routes and
Triple-C's. Launch templates (phase 7) are created only with the component tag and versioned,
launched from or deleted only when they carry it; their default version never changes. Auto
Scaling groups must be named `triple-c-*` and also carry it (request tag to create; resource tag
to change, protect, tag or delete, so an existing untagged `triple-c-*` group can't be claimed by
tagging it). A group must name its template's version explicitly: a group on `$Latest` or
`$Default` would launch any later version, whose instance profile nobody checked, as Auto
Scaling's service-linked role. Resources
that cannot carry a tag are authorized through one that does: routes and
route-table associations through the route table (and subnet), mount targets and the file-system
policy through the file system, KMS aliases through the key, role policies through the
(boundary-carrying) role. Clusters, log groups, parameters, roles and the state bucket are scoped
by name (`triple-c-*`, `/triple-c/*`, `triple-c-tfstate-*`).
(boundary-carrying) role. Clusters, capacity providers, log groups, parameters, roles, instance
profiles and the state bucket are scoped by name (`triple-c-*`, `/triple-c/*`,
`triple-c-tfstate-*`).
**What is not scoped.** The scoping above does not cover every action. These actions apply
account-wide:
@@ -276,6 +378,18 @@ account-wide:
and mount targets are created and deleted under this statement alone.
- Reads on `*`: `ec2:Describe*`, `elasticfilesystem:Describe*`, the KMS key reads and
`ListAliases`, `logs:DescribeLogGroups` and `ssm:DescribeParameters`.
- `ec2:RunInstances`, directly as well as through a group: only from a component-tagged launch
template, the instance and its volume carrying the component tag, a supported instance type,
Amazon's image and the environment's (tagged) security group, both as the template names them
(`ec2:IsLaunchTemplateResource`), and only a `triple-c-*-instance` role passed. The network
interface and **subnet are not tag-checked**: the group supplies the subnets (not the template),
and an `existing`-network environment's subnets are the user's own, untagged. The modules never
call `RunInstances` themselves: creating a group checks that the caller could, and the group's
launches run as Auto Scaling's service-linked role.
- `iam:CreateServiceLinkedRole`, only for `autoscaling.amazonaws.com` and `ecs.amazonaws.com`
(AWS fixes those roles' policies).
- Reads on `*`: `autoscaling:Describe*`, `ecs:DescribeCapacityProviders`; and the public
ECS-optimized AMI parameters.
- `s3:PutBucketPolicy` on the state bucket (only a bucket in the admin's own account, enforced by
`s3:ResourceAccount`): an admin could open that bucket to other accounts. The state holds no
secrets (the modules create none), but it maps the account's Triple-C resources.
@@ -290,7 +404,7 @@ it is unpacked), and an admin can run the very same files by hand from a bundle.
|---|---|---|
| `terraform/permissions-boundary` | the account's `triple-c-permissions-boundary` policy (an account administrator applies it; the admin tier cannot) | `boundary/terraform.tfstate` |
| `terraform/environment` | what every user shares: ECS cluster (Fargate, Fargate Spot), the tasks security group (no inbound rules), KMS key and alias, log group `/triple-c/<env>`, the network (`network_mode`: `public`, `private-nat`, `existing`) and, last, the record `/triple-c/environments/<env>` (which carries the advisory `max_workspace_hours` lifetime cap, 0 = none, at most 168) | `environments/<env>/terraform.tfstate` |
| `terraform/user-onboard` | one user's task and execution roles (with the boundary) and their own EFS file system, its `-efs` security group, two mount targets and a file-system policy that admits only their task role | `users/<env>/<owner key>/terraform.tfstate` |
| `terraform/user-onboard` | one user's task and execution roles (with the boundary) and their own EFS file system, its `-efs` security group, two mount targets and a file-system policy that admits only their task role. With `ec2_instance_type` set (phase 7) also their own EC2 capacity: cluster `triple-c-<env>--u-<user tag>`, one capacity provider, an Auto Scaling group (min 0), a launch template, the instance role and profile, and the record `/triple-c/users/<env>/<owner key>` | `users/<env>/<owner key>/terraform.tfstate` |
**Workspace lifetime cap (advisory).** `max_workspace_hours` (module variable, 0 to 168, default 0 =
no cap) is stored in the environment record and passed to every workspace as
@@ -398,6 +512,13 @@ aws iam create-policy --policy-name triple-c-permissions-boundary \
--policy-document file://policies/permissions-boundary.json
```
**Phase 7 changed the boundary** (version 2: the instance role's ECS agent, on the cluster its
`triple-c:cluster` tag names). An account administrator re-applies it once (the boundary module from the app, or
`aws iam create-policy-version --policy-arn arn:aws:iam::<account>:policy/triple-c-permissions-boundary
--policy-document file://policies/permissions-boundary.json --set-as-default`) before anyone
onboards with EC2 capacity; Fargate users are unaffected. The module tags the policy
`triple-c:template-version = 2`.
Create the policy **without a path**. The admin policy expects exactly
`arn:aws:iam::<account>:policy/triple-c-permissions-boundary`. The `terraform/permissions-boundary`
module creates the same policy, applied by an account administrator; the admin tier can read it
@@ -408,11 +529,12 @@ module creates the same policy, applied by an account administrator; the admin t
Create each file as a **customer managed policy** and attach it. In the console, go to
**IAM → Policies → Create policy → JSON**, paste the file, and attach the policy to a user, group
or role. Inline policies do not fit: a role's inline policies share 10,240 characters and an
environment admin needs every tier file (about 16,400), and a user's share 2,048. A managed policy
environment admin needs every tier file (23,304 characters in all), and a user's share 2,048. A managed policy
may hold 6,144 characters, not counting whitespace. Measured 2026-11-16 without whitespace, `workspace-user.json` is
6,036 (108 characters of headroom), `workspace-user-tasks.json` 1,293 (including `ReadFargatePrices`),
`environment-admin.json` 5,805 (339)
and `environment-admin-network.json` 3,340; a unit test (`every_policy_fits_a_managed_policy`)
6,036 (108 characters of headroom), `workspace-user-tasks.json` 1,827 (including `ReadFargatePrices` and the EC2 cluster Denies),
`environment-admin.json` 5,805 (339), `environment-admin-network.json` 3,340,
`environment-admin-compute.json` 3,326, `environment-admin-scaling.json` 2,970 and
`permissions-boundary.json` 2,661; a unit test (`every_policy_fits_a_managed_policy`)
fails before any file outgrows the limit. A tier that outgrows its files gains another file, which
the app then lists with the others. For IAM Identity Center, add the files as inline policies on the permission set
(the "up to 32,768 characters" limit is unverified) or reference the managed policies.
@@ -461,6 +583,83 @@ aws iam attach-user-policy --user-name <user> \
`containerOverrides`: every workspace user can read them through `ecs:Describe*` on `*`. Use
Secrets Manager (`triple-c/*`) references instead.
**Phase 7 (per-user EC2 capacity): the live run's answers (Task 7, 2026-11-16)**
The unit tests' evaluator is not IAM. What the live run (see "Phase 7: ECS on EC2" below) settled.
Still open for phase 7: an Identity Center attribute as the `triple-c-user` source and a user
without the tag (only a role tag was used live), and creating the Auto Scaling and ECS
service-linked roles as the admin (the test account had both). The other points below were
checked live; each one's mark gives its verdict (✓ as expected, ✗→fixed found wrong and fixed,
— not verified):
- ✓ **Creating an Auto Scaling group checks the caller's launch permissions.** The onboarding's
group create as the environment admin (only the four admin files) passed with no denial. A group
pinned to a template version naming `p5.48xlarge` was refused with "You are not authorized to
use launch template", so the create does run the caller's `ec2:RunInstances` check, with
`ec2:LaunchTemplate`, `ec2:IsLaunchTemplateResource`, the template's tags and the instance role's
`iam:PassRole` all as the statements expect.
- ✓ **`iam:AddRoleToInstanceProfile`, `ecs:PutClusterCapacityProviders`,
`ecs:CreateCapacityProvider`**, the public AMI parameter read, and tags at group create
(`autoscaling:CreateOrUpdateTags` was not needed separately): all passed in the real apply,
with no denial. Which context keys each one sent is not visible; the policy is not narrowed further.
- — **Service-linked roles.** The test account already had both. Creating them as the admin is
still unverified.
- ✓ **The instance role's agent calls**, under the boundary: the instance registered into the user's
cluster, polled, and ran tasks. ECR pulls through the execution role and logs worked too. So the
`triple-c:cluster` role tag resolves as `${aws:PrincipalTag/triple-c:cluster}`.
- ✓ **Version pinning.** `$Latest` and `$Default` are denied, and a numeric version is allowed (a
probe group with min/max 0, created and deleted as the admin).
- ✗→fixed **`autoscaling:InstanceTypes` required on a group create.** A template-only create sends
**no** `autoscaling:InstanceTypes`, so the `Null: false` denied every group (the first live
onboarding failed on it). It is dropped. Creating a group runs the caller's own
`ec2:RunInstances` check on the template, so `ec2:InstanceType` holds the type. A
`MixedInstancesPolicy` override of an unsupported type is denied by both checks
(`ForAllValues` on the key once present, and the launch check).
- ✓ **Instance types in templates.** The admin *can* create a template version naming
`p5.48xlarge` (as expected, no condition key). A group pinned to it is denied by the launch check.
- ✓ **Updating a group to an unsupported template version** runs the same launch check. In a
probe (a throwaway role whose `ec2:RunInstances` allows only `m7i.large`), a group pinned to an
`m7i.large` version updated to a `p5.48xlarge` version was refused ("You are not authorized to
use launch template"). A size-only update (no template) passed.
- ✓ **`aws:PrincipalTag/triple-c-user`** from a **role tag**: the tagged workspace role's session
ran tasks on its own `--u-<tag>` cluster. `RunTask` on another `--u-<other>` cluster was denied,
and the app shows the wrong-triple-c-user message. — An Identity Center attribute and an
untagged user were not tried. The policy's tag test is the same for both, so this is expected
to hold.
- ✓ **A `triple-c-*` capacity provider over someone else's group** is refused. ECS checks the
caller's `autoscaling:CreateOrUpdateTags` on the named group ("you do not have
autoscaling:CreateOrUpdateTags permissions to create tags on the Auto Scaling group"), which the
admin tier allows only on component-tagged `triple-c-*--u-*` groups.
- **IMDS and Docker access.** EC2 tasks use `bridge` networking (R2), and the template sets IMDSv2
with a hop limit of 1. Live: from the workspace container, the token `PUT` gets no answer
(curl times out) and an IMDSv1 `GET` gets 401, so no instance-role credentials. The task role's
credentials (169.254.170.2) work. The ECS agent's introspection API **is** reachable on the
bridge gateway (`172.17.0.1:51678`, 200). It returns the cluster, the container instance and
the instance's tasks, and no credentials. That is accepted: the instance runs only its user's
tasks (E1). A project with Docker access has the host's socket (group `docker-host`, `docker
ps` works as `claude`), so its user can start a host-network container and read IMDS (and so the
instance role). That is accepted too: Docker access is root on the user's own instance, and
the boundary keeps that role on the user's own cluster.
- **Logs are written per user.** All users' tasks log into the environment's one log group
`/triple-c/<env>`. The instance role has **no** log rights: the template sets
`ECS_ENABLE_AWSLOGS_EXECUTIONROLE_OVERRIDE=true`, so the awslogs driver writes with the task's
execution role (the live run asserts the task's log has lines). The task and execution roles may
create streams in the group but `PutLogEvents` only into `log-stream:<owner_key>/*`, the prefix
every awslogs stream of theirs has (`<owner_key>/<project>/<container>/<task id>`), so no
user's container, instance or host-network container can write into another user's streams.
Known limit: `logs:CreateLogStream` is authorized against the log group ARN only (the Service
Authorization Reference lists no log-stream resource type for it), so it cannot be scoped to
`log-stream:<owner_key>/*`. A user can therefore create a stream under another user's prefix
but cannot write events to it (`PutLogEvents` is scoped); the worst case is an empty stream.
This also changed the version-1 (Fargate) task and execution role policies, in place.
- **Orphaned cold-start tasks and clock skew.** A task with ECS Exec off still waiting for an
instance counts as another start's (the start says so and waits) until it is older than the EC2
placement wait (10 min) plus `ORPHAN_MARGIN` (10 min); after that a start or the app's startup
reconcile replaces it as an orphan. Its age is ECS's `createdAt` against the local clock, so a
local clock running more than 10 minutes fast could replace another machine's start mid cold
start (that start then reports it was replaced; Start again connects). A slow clock only
delays the replacement.
**Phase 6**
- **Access-point roots.** A workspace user creates their own access points, only on their own file
@@ -615,6 +814,7 @@ and the secret and removes the ECR policy; then the fixture destroys the modules
| Bubblewrap (P9) | ✗ `bwrap --unshare-user` fails: "No permissions to create new namespace" (kernel 6.1, `max_user_namespaces` 16384, no `CAP_SYS_ADMIN` on Fargate). The sandbox stays off on AWS |
| A revision per configuration | turning idle stop on adds a secret key, so the next start registers a new revision (by design) |
| Removal (R8) | ✓ final run: `remove_aws_resources` succeeds; the secret is scheduled for deletion, the access point deleted, the revision left ACTIVE. Since R10 the secret is deleted without a recovery window (the test now asserts it is gone) |
| Logs under the narrowed scope (2026-11-16, after phase 7's review) | ✓ re-run with the task and execution roles' `PutLogEvents` limited to `log-stream:<owner_key>/*` (`CreateLogStream` on the group): passed in 913 s; awslogs wrote both tasks' streams (the test reads and asserts task 1's and task 2's stop lines), cold start 109 s, idle stop exit 75; nothing left but the test secret already pending deletion |
| Cleanup | ✓ nothing `triple-c` left in us-west-2; environment KMS keys pending deletion |
### Phase 6: the cost estimate's price reads (2026-11-16)
@@ -660,3 +860,44 @@ spend under $0.05.
| Cost estimate | ✓ `fetch_rates` as the workspace user + `estimate`: $0.0495/hour for 1 vCPU / 2 GiB / 21 GiB x86, `efs_billed_separately` |
| Removal with discard | ✓ `remove_aws_resources(…, true)` Ok; both access points gone; the secret deleted |
| Cleanup | ✓ nothing `triple-c` left in us-west-2 (the run's environment KMS key pending deletion). The guard's task-definition retirement had never matched anything (ECS's `familyPrefix` is a whole family name); fixed, and 13 revisions earlier runs had left ACTIVE were deregistered and deleted |
### Phase 7: ECS on EC2 (2026-11-16)
```sh
TRIPLEC_E2E_TEST=aws::live_tests::environment::a_phase7_ec2_workspace_round_trips \
TRIPLEC_AWS_E2E_PROFILE=<admin profile> TRIPLEC_AWS_E2E_REGION=us-west-2 scripts/aws-e2e-workspace.sh
```
Same fixture as phase 5. The account admin applies the current boundary (the account had none,
so it is created and destroyed again). The environment admin role holds all four admin files.
The workspace role carries the role tag `triple-c-user=<tag>`, which an assumed-role session has
as a principal tag. The admin onboards it with `m7i.large`, at most 1 instance, through the app's
own pre-apply step (`prepare_onboarding`: record, choice, name-collision check). The project is
EC2, Hybrid, 1024 CPU units / 4096 MiB reserved, with Docker access, VPN and sandbox mode on, and
no repos. The guard (`ec2_cleanup`, as the account admin) stops the per-user clusters' tasks,
force-deletes their groups (waiting until they are gone), terminates any instance tagged with the
environment, then deletes capacity providers, container instances, clusters (retried while busy),
launch templates, the user record and the instance role and profile. After that the fixture
destroys the modules. `TRIPLEC_E2E_PAUSE_SECS` holds a failed first start for inspection before
the guard runs.
| Point | Result |
|---|---|
| Onboarding with EC2 | ✓ after the `InstanceTypes` fix (the first run's group create was denied, see above). Applied in 93 s: cluster `triple-c-<env>--u-<tag>` ACTIVE with exactly its capacity provider as the default strategy; group min 0, max 1, desired 0, template pinned to version 1, scale-in protection on, tagged with component/environment/owner/version; capacity provider with managed scaling at target 100 and managed termination protection; launch template `m7i.large`, the current ECS-optimized AL2023 AMI, IMDSv2 required, hop limit 1; instance profile holding the instance role; the user record reads `Valid` as the workspace user |
| Collision check | ✓ `DescribeClusters include=TAGS` as the admin tier returns the cluster's `triple-c:owner` tag. A missing name answers `None` for both the cluster (a `failures` entry) and the group. Onboarding another owner with the same `triple-c-user` value is refused: "An ECS cluster named … already belongs to another onboarding" |
| Admin-tier probes | ✓ `$Latest` denied; pinned version allowed and deleted; a `p5.48xlarge` template version created but a group pinned to it denied; a capacity provider over an untagged group denied (see "Phase 7" under Known gaps) |
| **ECS Exec on a cold start** | ✗→fixed (app). ECS turns `enableExecuteCommand` **off** on a RunTask that has to wait for an instance (the task goes PROVISIONING behind the capacity provider). CloudTrail shows `true` in the request and `false` in the answer, and the task keeps it off once placed, so the ExecuteCommandAgent never starts and every cold start timed out ("ECS Exec agent not running yet"). The same RunTask with an instance up and room keeps it `true`, and one that has to wait gets `false` even with an instance up. A cold start now lets that first task bring the instance up. Once the task is placed, the app stops it, waits until it has stopped and runs the task again onto the instance (`rerun_once_placed`). A live task found with ECS Exec off is stopped and replaced |
| Cold start | ✓ 249 s (run 4) and 203 s (run 5) from Start to connected. Run 4: RunTask at +4 s; "Starting an instance" until +186 s (instance launch, ECS registration and placement of the first task, which is then stopped); the second task ready and the tunnel at +248 s |
| Agent tunnel, bridge mode | ✓ the ECS Exec port forward reaches the agent on the container's own loopback (`echo` through the agent as `claude`); the task definition is `bridge`/`EC2` with the `docker-socket` host volume |
| Docker access | ✓ `/var/run/docker.sock` is `root:docker-host 0660`, and `claude` is in `docker-host`; `docker ps` as `claude` lists the workspace and the ECS agent (Docker 25.0.18) |
| VPN | ✓ as root: `ip link add wg0 type wireguard` and `ip tuntap add tun0 mode tun` both work; `/dev/net/tun` is there; `net.ipv4.conf.all.src_valid_mark = 1`; `sudo ip link add … wireguard` works as `claude` |
| E8 bubblewrap | ✗→decided. `bwrap --unshare-user` and `--unshare-all` fail with "No permissions to create new namespace", although the kernel (6.1, AL2023) allows user namespaces (`max_user_namespaces` 30937). Docker's default seccomp profile refuses `unshare` without `CAP_SYS_ADMIN`. The sandbox is kept off on EC2 as on Fargate (`sandbox_allowed`, the Runtime toggle) |
| IMDS / introspection | see "IMDS and Docker access" above: no token (hop limit 1), IMDSv1 401, task-role credentials work, introspection reachable on the bridge gateway (metadata only) |
| Isolation | ✓ `RunTask` on another user's cluster (`triple-c-<env>--u-<other>`, created empty by the admin) as the tagged workspace user is denied, and the app's error is the wrong-triple-c-user one ("This EC2 capacity belongs to a different triple-c-user value than your sign-in's …") |
| Stop | ✓ STOPPED 6 s after `stop_project` |
| Scale-in to zero | ✓ from the group's activity history (run 4). The task STOPPED at 00:55:42Z. At 01:11:08Z managed scaling's low alarm set desired capacity 1 → 0. Scale-in protection first held the instance ("all remaining instances are protected from scale-in"). ECS then removed it from the now-idle instance, which was terminated at 01:11:59–01:12:44Z. That is **about 15.5 minutes to the scale-in decision and 17 minutes until the instance is gone**. The cold start's scale-out is similar: the high alarm fired about 3 minutes after the RunTask, and the instance was running a minute later |
| Second start | ✓ warm, right after stop 1 with the instance still up: 29 s (no instance wait, image cached, ECS Exec on at once). After a scale-in a start is cold again (≈ 3.5–4 min) |
| Removal | ✓ `remove_aws_resources(…, true)` after stop 2: Ok (task, secret, access point) |
| Re-apply without EC2, instance still up | ✓ `prepare_onboarding(…, None)` then the apply as the environment admin, 79 s: "0 added, 0 changed, 9 destroyed". The provider drained the group (the instance was InService, scale-in protected) and deleted it; after the apply no group, no live instance, capacity provider INACTIVE, cluster INACTIVE, launch template gone, instance profile and role gone, the user record Absent (as the workspace user). No denied action |
| Offboard and destroy | ✓ as the environment admin: onboarding destroy (9) and environment destroy (15) with no denied action; the account admin's boundary destroy after |
| Runs and cleanup | Five runs. Run 1: the `InstanceTypes` denial. Run 2: ECS Exec off on the cold start. Run 3: inspection over SSM (Session Manager core attached by hand to the instance role, boundary removed, both undone before the teardown), the ECS Exec diagnosis and the in-task checks. Run 4: everything to the scale-in, which the guard could not finish because the credentials expired. Run 5, `TRIPLEC_E2E_PHASE7_SHORT=1` (life cycle only): passed in 577 s. Run 6 (short, after the review fixes: the rerun's launch through `find_or_run`, exactly one live task after the cold start, `sandbox.enabled = false` in the task's settings): passed in 583 s, with a cold start of 222 s and a warm one of 19 s. Run 7 (short, after the final review's log scoping: no log rights on the instance role, `PutLogEvents` only into `<owner_key>/*` for the task and execution roles): passed in 527 s, cold start 182 s, warm 19 s, and task 1's awslogs stream had its lines (written with the execution role). Every run's guard, or by hand with tofu destroy from the run's work folders once the credentials were renewed, left nothing: the final inventory of us-west-2 has no instance, group, launch template, Triple-C capacity provider, cluster, `triple-c*` role, policy or instance profile, `/triple-c` parameter, log group, ECR repository, file system, secret, task definition, VPC, state bucket or boundary. The environments' KMS keys are pending deletion (7 days). Spend: about 1.5 m7i.large instance-hours plus minutes of EFS and public IPv4, under $0.30 |
+230
View File
@@ -0,0 +1,230 @@
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "ManageTripleCCapacityProviders",
"Effect": "Allow",
"Action": [
"ecs:CreateCapacityProvider",
"ecs:UpdateCapacityProvider",
"ecs:DeleteCapacityProvider",
"ecs:PutClusterCapacityProviders",
"ecs:TagResource",
"ecs:UntagResource",
"ecs:ListTagsForResource"
],
"Resource": [
"arn:aws:ecs:*:*:capacity-provider/triple-c-*--u-*"
]
},
{
"Sid": "CreateTaggedLaunchTemplates",
"Effect": "Allow",
"Action": [
"ec2:CreateLaunchTemplate"
],
"Resource": [
"arn:aws:ec2:*:*:launch-template/*"
],
"Condition": {
"Null": {
"aws:RequestTag/triple-c:component": "false"
}
}
},
{
"Sid": "TagComputeAtCreate",
"Effect": "Allow",
"Action": [
"ec2:CreateTags"
],
"Resource": [
"arn:aws:ec2:*:*:launch-template/*",
"arn:aws:ec2:*:*:instance/*",
"arn:aws:ec2:*:*:volume/*",
"arn:aws:ec2:*:*:network-interface/*"
],
"Condition": {
"StringEquals": {
"ec2:CreateAction": [
"CreateLaunchTemplate",
"RunInstances"
]
}
}
},
{
"Sid": "ManageAndUseTripleCLaunchTemplates",
"Effect": "Allow",
"Action": [
"ec2:CreateLaunchTemplateVersion",
"ec2:DeleteLaunchTemplate",
"ec2:CreateTags",
"ec2:DeleteTags",
"ec2:RunInstances"
],
"Resource": [
"arn:aws:ec2:*:*:launch-template/*"
],
"Condition": {
"Null": {
"aws:ResourceTag/triple-c:component": "false"
}
}
},
{
"Sid": "LaunchSupportedInstances",
"Effect": "Allow",
"Action": [
"ec2:RunInstances"
],
"Resource": [
"arn:aws:ec2:*:*:instance/*"
],
"Condition": {
"Null": {
"aws:RequestTag/triple-c:component": "false"
},
"ArnLike": {
"ec2:LaunchTemplate": "arn:aws:ec2:*:*:launch-template/*"
},
"StringEquals": {
"ec2:InstanceType": [
"m7i.large",
"m7i.xlarge",
"m7i.2xlarge",
"m7i.4xlarge",
"m7g.large",
"m7g.xlarge",
"m7g.2xlarge",
"m7g.4xlarge",
"c7i.large",
"c7i.xlarge",
"c7i.2xlarge",
"c7i.4xlarge",
"c7g.large",
"c7g.xlarge",
"c7g.2xlarge",
"c7g.4xlarge",
"r7i.large",
"r7i.xlarge",
"r7i.2xlarge",
"r7i.4xlarge",
"r7g.large",
"r7g.xlarge",
"r7g.2xlarge",
"r7g.4xlarge",
"t3.large",
"t3.xlarge",
"t3.2xlarge",
"t4g.large",
"t4g.xlarge",
"t4g.2xlarge"
]
}
}
},
{
"Sid": "LaunchTaggedVolumes",
"Effect": "Allow",
"Action": [
"ec2:RunInstances"
],
"Resource": [
"arn:aws:ec2:*:*:volume/*"
],
"Condition": {
"Null": {
"aws:RequestTag/triple-c:component": "false"
},
"ArnLike": {
"ec2:LaunchTemplate": "arn:aws:ec2:*:*:launch-template/*"
}
}
},
{
"Sid": "LaunchIntoSubnets",
"Effect": "Allow",
"Action": [
"ec2:RunInstances"
],
"Resource": [
"arn:aws:ec2:*:*:network-interface/*",
"arn:aws:ec2:*:*:subnet/*"
],
"Condition": {
"ArnLike": {
"ec2:LaunchTemplate": "arn:aws:ec2:*:*:launch-template/*"
}
}
},
{
"Sid": "LaunchWithTripleCGroups",
"Effect": "Allow",
"Action": [
"ec2:RunInstances"
],
"Resource": [
"arn:aws:ec2:*:*:security-group/*"
],
"Condition": {
"ArnLike": {
"ec2:LaunchTemplate": "arn:aws:ec2:*:*:launch-template/*"
},
"Bool": {
"ec2:IsLaunchTemplateResource": "true"
},
"Null": {
"aws:ResourceTag/triple-c:component": "false"
}
}
},
{
"Sid": "LaunchAmazonImages",
"Effect": "Allow",
"Action": [
"ec2:RunInstances"
],
"Resource": [
"arn:aws:ec2:*::image/*"
],
"Condition": {
"StringEquals": {
"ec2:Owner": "amazon"
},
"ArnLike": {
"ec2:LaunchTemplate": "arn:aws:ec2:*:*:launch-template/*"
},
"Bool": {
"ec2:IsLaunchTemplateResource": "true"
}
}
},
{
"Sid": "PassInstanceRolesToEc2",
"Effect": "Allow",
"Action": [
"iam:PassRole"
],
"Resource": [
"arn:aws:iam::*:role/triple-c-*-instance"
],
"Condition": {
"StringEquals": {
"iam:PassedToService": "ec2.amazonaws.com"
}
}
},
{
"Sid": "ReadEcsOptimizedAmis",
"Effect": "Allow",
"Action": [
"ssm:GetParameter",
"ssm:GetParameters"
],
"Resource": [
"arn:aws:ssm:*::parameter/aws/service/ecs/optimized-ami/*"
]
}
]
}
+178
View File
@@ -0,0 +1,178 @@
{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "CreatePinnedGroups",
"Effect": "Allow",
"Action": [
"autoscaling:CreateAutoScalingGroup"
],
"Resource": [
"arn:aws:autoscaling:*:*:autoScalingGroup:*:autoScalingGroupName/triple-c-*--u-*"
],
"Condition": {
"Null": {
"aws:RequestTag/triple-c:component": "false",
"autoscaling:TargetGroupARNs": "true",
"autoscaling:LoadBalancerNames": "true"
},
"Bool": {
"autoscaling:LaunchTemplateVersionSpecified": "true"
},
"ForAllValues:StringEquals": {
"autoscaling:InstanceTypes": [
"m7i.large",
"m7i.xlarge",
"m7i.2xlarge",
"m7i.4xlarge",
"m7g.large",
"m7g.xlarge",
"m7g.2xlarge",
"m7g.4xlarge",
"c7i.large",
"c7i.xlarge",
"c7i.2xlarge",
"c7i.4xlarge",
"c7g.large",
"c7g.xlarge",
"c7g.2xlarge",
"c7g.4xlarge",
"r7i.large",
"r7i.xlarge",
"r7i.2xlarge",
"r7i.4xlarge",
"r7g.large",
"r7g.xlarge",
"r7g.2xlarge",
"r7g.4xlarge",
"t3.large",
"t3.xlarge",
"t3.2xlarge",
"t4g.large",
"t4g.xlarge",
"t4g.2xlarge"
]
}
}
},
{
"Sid": "UpdatePinnedGroups",
"Effect": "Allow",
"Action": [
"autoscaling:UpdateAutoScalingGroup"
],
"Resource": [
"arn:aws:autoscaling:*:*:autoScalingGroup:*:autoScalingGroupName/triple-c-*--u-*"
],
"Condition": {
"Null": {
"aws:ResourceTag/triple-c:component": "false",
"autoscaling:TargetGroupARNs": "true",
"autoscaling:LoadBalancerNames": "true"
},
"BoolIfExists": {
"autoscaling:LaunchTemplateVersionSpecified": "true"
},
"ForAllValues:StringEquals": {
"autoscaling:InstanceTypes": [
"m7i.large",
"m7i.xlarge",
"m7i.2xlarge",
"m7i.4xlarge",
"m7g.large",
"m7g.xlarge",
"m7g.2xlarge",
"m7g.4xlarge",
"c7i.large",
"c7i.xlarge",
"c7i.2xlarge",
"c7i.4xlarge",
"c7g.large",
"c7g.xlarge",
"c7g.2xlarge",
"c7g.4xlarge",
"r7i.large",
"r7i.xlarge",
"r7i.2xlarge",
"r7i.4xlarge",
"r7g.large",
"r7g.xlarge",
"r7g.2xlarge",
"r7g.4xlarge",
"t3.large",
"t3.xlarge",
"t3.2xlarge",
"t4g.large",
"t4g.xlarge",
"t4g.2xlarge"
]
}
}
},
{
"Sid": "ManageTripleCAutoScalingGroups",
"Effect": "Allow",
"Action": [
"autoscaling:SetInstanceProtection",
"autoscaling:CreateOrUpdateTags",
"autoscaling:DeleteTags",
"autoscaling:DeleteAutoScalingGroup"
],
"Resource": [
"arn:aws:autoscaling:*:*:autoScalingGroup:*:autoScalingGroupName/triple-c-*--u-*"
],
"Condition": {
"Null": {
"aws:ResourceTag/triple-c:component": "false"
}
}
},
{
"Sid": "CreateComputeServiceLinkedRoles",
"Effect": "Allow",
"Action": [
"iam:CreateServiceLinkedRole"
],
"Resource": [
"arn:aws:iam::*:role/aws-service-role/autoscaling.amazonaws.com/AWSServiceRoleForAutoScaling*",
"arn:aws:iam::*:role/aws-service-role/ecs.amazonaws.com/AWSServiceRoleForECS*"
],
"Condition": {
"StringEquals": {
"iam:AWSServiceName": [
"autoscaling.amazonaws.com",
"ecs.amazonaws.com"
]
}
}
},
{
"Sid": "ManageTripleCInstanceProfiles",
"Effect": "Allow",
"Action": [
"iam:CreateInstanceProfile",
"iam:DeleteInstanceProfile",
"iam:GetInstanceProfile",
"iam:AddRoleToInstanceProfile",
"iam:RemoveRoleFromInstanceProfile",
"iam:TagInstanceProfile",
"iam:UntagInstanceProfile",
"iam:ListInstanceProfileTags"
],
"Resource": [
"arn:aws:iam::*:instance-profile/triple-c-*"
]
},
{
"Sid": "ReadCompute",
"Effect": "Allow",
"Action": [
"ecs:DescribeCapacityProviders",
"autoscaling:Describe*"
],
"Resource": [
"*"
]
}
]
}
+49
View File
@@ -116,6 +116,55 @@
"arn:aws:bedrock:*:*:inference-profile/*",
"arn:aws:bedrock:*:*:application-inference-profile/*"
]
},
{
"Sid": "EcsAgentOnItsOwnCluster",
"Effect": "Allow",
"Action": [
"ecs:RegisterContainerInstance",
"ecs:DeregisterContainerInstance",
"ecs:Poll",
"ecs:StartTelemetrySession",
"ecs:Submit*",
"ecs:UpdateContainerInstancesState"
],
"Resource": [
"arn:aws:ecs:*:*:cluster/${aws:PrincipalTag/triple-c:cluster}",
"arn:aws:ecs:*:*:container-instance/${aws:PrincipalTag/triple-c:cluster}/*"
],
"Condition": {
"StringLike": {
"aws:PrincipalTag/triple-c:cluster": "triple-c-*--u-*"
}
}
},
{
"Sid": "EcsAgentTagsItsInstance",
"Effect": "Allow",
"Action": [
"ecs:TagResource"
],
"Resource": [
"arn:aws:ecs:*:*:container-instance/${aws:PrincipalTag/triple-c:cluster}/*"
],
"Condition": {
"StringEquals": {
"ecs:CreateAction": "RegisterContainerInstance"
},
"StringLike": {
"aws:PrincipalTag/triple-c:cluster": "triple-c-*--u-*"
}
}
},
{
"Sid": "EcsAgentDiscovery",
"Effect": "Allow",
"Action": [
"ecs:DiscoverPollEndpoint"
],
"Resource": [
"*"
]
}
]
}
+38
View File
@@ -85,6 +85,44 @@
"aws:ResourceTag/triple-c:component": "false"
}
}
},
{
"Sid": "RunOnlyOnOwnEc2Cluster",
"Effect": "Deny",
"Action": [
"ecs:RunTask",
"ecs:StartTask"
],
"Resource": [
"*"
],
"Condition": {
"ArnLike": {
"ecs:cluster": "arn:aws:ecs:*:*:cluster/triple-c-*--u-*"
},
"ArnNotLike": {
"ecs:cluster": "arn:aws:ecs:*:*:cluster/triple-c-*--u-${aws:PrincipalTag/triple-c-user}"
}
}
},
{
"Sid": "NoEc2ClusterWithoutUserTag",
"Effect": "Deny",
"Action": [
"ecs:RunTask",
"ecs:StartTask"
],
"Resource": [
"*"
],
"Condition": {
"ArnLike": {
"ecs:cluster": "arn:aws:ecs:*:*:cluster/triple-c-*--u-*"
},
"Null": {
"aws:PrincipalTag/triple-c-user": "true"
}
}
}
]
}
+8 -1
View File
@@ -1,12 +1,19 @@
# The permissions boundary every role a Triple-C module creates must carry:
# the most any such role may ever hold. One per account.
locals {
# BOUNDARY_MODULE_VERSION in app/src-tauri/src/aws/tofu_modules.rs.
# 2: the instance role's ECS agent actions, on its own cluster (phase 7).
template_version = 2
}
provider "aws" {
region = var.region
default_tags {
tags = {
"triple-c:component" = "boundary"
"triple-c:component" = "boundary"
"triple-c:template-version" = tostring(local.template_version)
}
}
}
+311
View File
@@ -0,0 +1,311 @@
# The user's own EC2 capacity (phase 7, E1–E4, E11), made only when
# `ec2_instance_type` is set: every resource and data source here has
# `count = local.ec2_count`, so an onboarding without it plans exactly as
# before. One cluster per user (`triple-c-<env>--u-<user tag>`, the name the
# workspace policy lets only that `triple-c-user` principal tag run tasks on)
# with exactly one capacity provider: an Auto Scaling group of 0 to
# `ec2_max_instances` instances that ECS's managed scaling grows when a task
# needs one and shrinks back to zero when none does.
#
# Removing it (re-applying with an empty type, or destroying) goes in
# dependency order: the user record first, then the cluster's provider list
# (ECS refuses while a task still runs on the provider, so a running
# workspace stops the removal before anything is torn down), the capacity
# provider, the group (`force_delete` off: the provider first drains it to
# zero, clearing scale-in protection, and waits for the instances to
# terminate), the launch template, the instance profile and role, and last
# the cluster, which the launch template's user data references so that it
# outlives every instance registered into it.
locals {
ec2 = var.ec2_instance_type != ""
ec2_count = local.ec2 ? 1 : 0
# `user_ec2_stem` and `user_instance_role_name` in tofu_modules.rs.
ec2_name = "triple-c-${var.environment_name}--u-${var.ec2_user_tag}"
instance_role_name = "${local.name}-instance"
# Graviton families (m7g, c7g, r7g, t4g: a `g` after the generation) run
# the arm64 image.
ec2_arm64 = can(regex("^[a-z]+[0-9]+g[a-z]*\\.", var.ec2_instance_type))
ecs_ami_parameter = (
local.ec2_arm64
? "/aws/service/ecs/optimized-ami/amazon-linux-2023/arm64/recommended/image_id"
: "/aws/service/ecs/optimized-ami/amazon-linux-2023/recommended/image_id"
)
}
# The provider's default tags (component, environment, owner, version), for
# the places they don't reach by themselves: the launch template's tag
# specifications and the group's tags.
data "aws_default_tags" "current" {
count = local.ec2_count
}
# The current ECS-optimized Amazon Linux 2023 image, resolved at apply time.
data "aws_ssm_parameter" "ecs_ami" {
count = local.ec2_count
name = local.ecs_ami_parameter
}
resource "aws_ecs_cluster" "ec2" {
count = local.ec2_count
name = local.ec2_name
setting {
name = "containerInsights"
value = "disabled"
}
}
# The instance's role: the ECS agent on this user's cluster only (the
# boundary allows the agent's calls only on the cluster this role's
# `triple-c:cluster` tag names), image pulls and the environment's logs.
# ECS Exec needs nothing here: its channels run on the task role.
resource "aws_iam_role" "instance" {
count = local.ec2_count
name = local.instance_role_name
description = "Triple-C ${var.environment_name}: EC2 instances of ${var.owner_id}"
permissions_boundary = local.boundary_arn
assume_role_policy = jsonencode({
Version = "2012-10-17"
Statement = [{
Effect = "Allow"
Principal = { Service = "ec2.amazonaws.com" }
Action = "sts:AssumeRole"
}]
})
tags = { "triple-c:cluster" = local.ec2_name }
}
resource "aws_iam_role_policy" "instance" {
count = local.ec2_count
name = "triple-c-instance"
role = aws_iam_role.instance[0].id
policy = jsonencode({
Version = "2012-10-17"
Statement = [
{
Sid = "EcsAgentOnOwnCluster"
Effect = "Allow"
Action = [
"ecs:RegisterContainerInstance",
"ecs:DeregisterContainerInstance",
"ecs:Poll",
"ecs:StartTelemetrySession",
"ecs:Submit*",
"ecs:UpdateContainerInstancesState",
]
Resource = [
"arn:${local.partition}:ecs:${var.region}:${local.account}:cluster/${local.ec2_name}",
"arn:${local.partition}:ecs:${var.region}:${local.account}:container-instance/${local.ec2_name}/*",
]
},
{
Sid = "EcsAgentDiscovery"
Effect = "Allow"
Action = "ecs:DiscoverPollEndpoint"
Resource = "*"
},
{
Sid = "EcrAuthorization"
Effect = "Allow"
Action = "ecr:GetAuthorizationToken"
Resource = "*"
},
{
Sid = "PullTripleCImages"
Effect = "Allow"
Action = [
"ecr:BatchGetImage",
"ecr:GetDownloadUrlForLayer",
"ecr:BatchCheckLayerAvailability",
]
Resource = [
"arn:${local.partition}:ecr:${var.region}:${local.account}:repository/triple-c-*",
"arn:${local.partition}:ecr:${var.region}:${local.account}:repository/triple-c/*",
]
},
]
})
}
resource "aws_iam_instance_profile" "instance" {
count = local.ec2_count
name = aws_iam_role.instance[0].name
role = aws_iam_role.instance[0].name
}
resource "aws_launch_template" "ec2" {
count = local.ec2_count
name = local.ec2_name
description = "Triple-C ${var.environment_name}: EC2 capacity of ${var.owner_id}"
image_id = data.aws_ssm_parameter.ecs_ami[0].insecure_value
instance_type = var.ec2_instance_type
iam_instance_profile {
name = aws_iam_instance_profile.instance[0].name
}
# The environment's tasks group (outbound only), named in the template so
# the admin policy sees it as a launch-template resource. A public IP only
# where the environment's tasks get one: in a public-subnet environment
# the agent, and bridge-network tasks through the instance, reach out
# through the internet gateway.
network_interfaces {
device_index = 0
security_groups = [local.env.security_group_id]
associate_public_ip_address = local.env.assign_public_ip
delete_on_termination = true
}
# IMDSv2 only, one hop: the agent (host network) reaches it, while
# bridge-network containers are one hop further and their token request
# never comes back, so they can't take the instance role's credentials.
metadata_options {
http_endpoint = "enabled"
http_tokens = "required"
http_put_response_hop_limit = 1
instance_metadata_tags = "disabled"
}
block_device_mappings {
device_name = "/dev/xvda"
ebs {
volume_size = 50
volume_type = "gp3"
encrypted = true
delete_on_termination = true
}
}
user_data = base64encode(<<-EOT
#!/bin/bash
cat >>/etc/ecs/ecs.config <<'EOF'
ECS_CLUSTER=${aws_ecs_cluster.ec2[0].name}
ECS_ENABLE_TASK_IAM_ROLE=true
ECS_ENABLE_AWSLOGS_EXECUTIONROLE_OVERRIDE=true
EOF
EOT
)
# Default tags don't reach what a launch creates: the instance, its
# volume and its interface are tagged here (the admin policy lets a launch
# create only component-tagged instances and volumes).
tag_specifications {
resource_type = "instance"
tags = data.aws_default_tags.current[0].tags
}
tag_specifications {
resource_type = "volume"
tags = data.aws_default_tags.current[0].tags
}
tag_specifications {
resource_type = "network-interface"
tags = data.aws_default_tags.current[0].tags
}
}
resource "aws_autoscaling_group" "ec2" {
count = local.ec2_count
name = local.ec2_name
vpc_zone_identifier = local.env.subnet_ids
# Zero when idle; ECS's managed scaling sets the desired count.
min_size = 0
max_size = var.ec2_max_instances
protect_from_scale_in = true
force_delete = false
# A fixed version, never $Latest/$Default (the admin policy refuses a
# group that would follow later versions). Running instances keep theirs
# until they scale in.
launch_template {
id = aws_launch_template.ec2[0].id
version = aws_launch_template.ec2[0].latest_version
}
# Provider default tags don't apply to groups.
dynamic "tag" {
for_each = data.aws_default_tags.current[0].tags
content {
key = tag.key
value = tag.value
propagate_at_launch = false
}
}
# ECS adds this when the capacity provider takes the group over; declared
# so a re-apply doesn't remove it.
tag {
key = "AmazonECSManaged"
value = "true"
propagate_at_launch = true
}
# The drain before deletion waits for the instances to terminate.
timeouts {
delete = "20m"
}
lifecycle {
ignore_changes = [desired_capacity]
}
}
resource "aws_ecs_capacity_provider" "ec2" {
count = local.ec2_count
name = local.ec2_name
auto_scaling_group_provider {
auto_scaling_group_arn = aws_autoscaling_group.ec2[0].arn
# ECS protects an instance from scale-in while a task runs on it.
managed_termination_protection = "ENABLED"
managed_scaling {
status = "ENABLED"
target_capacity = 100
minimum_scaling_step_size = 1
maximum_scaling_step_size = 1
instance_warmup_period = 60
}
}
}
resource "aws_ecs_cluster_capacity_providers" "ec2" {
count = local.ec2_count
cluster_name = aws_ecs_cluster.ec2[0].name
capacity_providers = [aws_ecs_capacity_provider.ec2[0].name]
default_capacity_provider_strategy {
capacity_provider = aws_ecs_capacity_provider.ec2[0].name
weight = 1
}
}
# The user's record (E11): what the app reads to run EC2 projects, since
# workspace users can't read the state bucket. Outside
# /triple-c/environments/, so environment detection never sees it. Written
# last, removed first.
resource "aws_ssm_parameter" "user" {
count = local.ec2_count
name = "/triple-c/users/${var.environment_name}/${var.owner_key}"
description = "Triple-C ${var.environment_name}: EC2 capacity of ${var.owner_id}"
type = "String"
value = jsonencode({
schema = 1
instance_type = var.ec2_instance_type
cluster_arn = aws_ecs_cluster.ec2[0].arn
capacity_provider = aws_ecs_capacity_provider.ec2[0].name
user_tag = var.ec2_user_tag
max_instances = var.ec2_max_instances
})
depends_on = [aws_ecs_cluster_capacity_providers.ec2]
}
+20 -4
View File
@@ -37,11 +37,19 @@ resource "aws_iam_role_policy" "task" {
Resource = "*"
},
{
Sid = "WriteEnvironmentLogs"
# CreateLogStream is authorized on the log group; writing only into
# the user's own streams (other users' are in the same group).
Sid = "CreateEnvironmentLogStreams"
Effect = "Allow"
Action = ["logs:CreateLogStream", "logs:PutLogEvents"]
Action = ["logs:CreateLogStream"]
Resource = local.log_group_arn
},
{
Sid = "WriteOwnLogStreams"
Effect = "Allow"
Action = ["logs:PutLogEvents"]
Resource = local.own_log_streams_arn
},
{
Sid = "MountOwnFileSystem"
Effect = "Allow"
@@ -115,11 +123,19 @@ resource "aws_iam_role_policy" "exec" {
}
},
{
Sid = "WriteEnvironmentLogs"
# CreateLogStream is authorized on the log group; writing only into
# the user's own streams (other users' are in the same group).
Sid = "CreateEnvironmentLogStreams"
Effect = "Allow"
Action = ["logs:CreateLogStream", "logs:PutLogEvents"]
Action = ["logs:CreateLogStream"]
Resource = local.log_group_arn
},
{
Sid = "WriteOwnLogStreams"
Effect = "Allow"
Action = ["logs:PutLogEvents"]
Resource = local.own_log_streams_arn
},
]
})
}
+10 -5
View File
@@ -1,10 +1,11 @@
# One user's place in an environment: their task and execution roles and
# their own EFS file system (D2, strict tenancy). The environment itself is
# read only from its record, never from its state.
# their own EFS file system (D2, strict tenancy), and optionally their own
# EC2 capacity (ec2.tf, phase 7). The environment itself is read only from
# its record, never from its state.
locals {
# ONBOARD_MODULE_VERSION in app/src-tauri/src/aws/tofu_modules.rs.
template_version = 1
template_version = 2
name = "triple-c-${var.environment_name}-${var.owner_key}"
partition = data.aws_partition.current.partition
@@ -15,6 +16,9 @@ locals {
boundary_arn = "arn:${local.partition}:iam::${local.account}:policy/triple-c-permissions-boundary"
log_group_arn = "arn:${local.partition}:logs:${var.region}:${local.account}:log-group:${local.env.log_group}:*"
# The user's own streams in it: every awslogs stream of theirs is
# `<owner_key>/<project>/<container>/<task id>` (log_stream_prefix).
own_log_streams_arn = "arn:${local.partition}:logs:${var.region}:${local.account}:log-group:${local.env.log_group}:log-stream:${var.owner_key}/*"
}
provider "aws" {
@@ -64,10 +68,11 @@ data "aws_ssm_parameter" "environment" {
))
&& jsondecode(self.insecure_value).log_group == "/triple-c/${var.environment_name}"
&& can(regex("^sg-[0-9a-f]{8,17}$", jsondecode(self.insecure_value).security_group_id))
&& alltrue([for s in jsondecode(self.insecure_value).subnet_ids : can(regex("^subnet-[0-9a-f]{8,17}$", s))]),
&& alltrue([for s in jsondecode(self.insecure_value).subnet_ids : can(regex("^subnet-[0-9a-f]{8,17}$", s))])
&& (jsondecode(self.insecure_value).assign_public_ip == true || jsondecode(self.insecure_value).assign_public_ip == false),
false
)
error_message = "The environment record has a malformed kms_key_arn (not a key in this region and account), log_group (not /triple-c/<environment>), security_group_id or subnet id."
error_message = "The environment record has a malformed kms_key_arn (not a key in this region and account), log_group (not /triple-c/<environment>), security_group_id, subnet id or assign_public_ip."
}
}
}
+23
View File
@@ -18,6 +18,20 @@ output "efs_security_group_id" {
value = aws_security_group.efs.id
}
# The user's EC2 capacity, null without it. The app reads these from the
# user record (/triple-c/users/<env>/<owner key>), not from the state.
output "ec2_cluster_arn" {
value = one(aws_ecs_cluster.ec2[*].arn)
}
output "ec2_capacity_provider" {
value = one(aws_ecs_capacity_provider.ec2[*].name)
}
output "ec2_instance_type" {
value = local.ec2 ? var.ec2_instance_type : null
}
# Written only when every resource above was applied: a run that fails
# part-way never evaluates it, so the app reads its absence from the state
# as "the last onboarding didn't finish" (Incomplete, Join re-applies)
@@ -34,5 +48,14 @@ output "onboarding_complete" {
aws_iam_role_policy.task_bedrock,
aws_iam_role.exec,
aws_iam_role_policy.exec,
aws_ecs_cluster.ec2,
aws_iam_role.instance,
aws_iam_role_policy.instance,
aws_iam_instance_profile.instance,
aws_launch_template.ec2,
aws_autoscaling_group.ec2,
aws_ecs_capacity_provider.ec2,
aws_ecs_cluster_capacity_providers.ec2,
aws_ssm_parameter.user,
]
}
+55
View File
@@ -54,3 +54,58 @@ variable "bedrock_on_task_role" {
type = bool
default = false
}
# Optional per-user EC2 capacity (phase 7). Empty = none: the module then
# makes nothing for EC2 and the user's projects run on Fargate only.
variable "ec2_instance_type" {
description = "The instance type of the user's own EC2 capacity (e.g. m7i.large or m7g.large), or empty for none. Only the types the app can size are accepted."
type = string
default = ""
# The app's table (`ec2_instance_spec`) and the admin policy's lists.
validation {
condition = var.ec2_instance_type == "" || contains([
"m7i.large", "m7i.xlarge", "m7i.2xlarge", "m7i.4xlarge",
"m7g.large", "m7g.xlarge", "m7g.2xlarge", "m7g.4xlarge",
"c7i.large", "c7i.xlarge", "c7i.2xlarge", "c7i.4xlarge",
"c7g.large", "c7g.xlarge", "c7g.2xlarge", "c7g.4xlarge",
"r7i.large", "r7i.xlarge", "r7i.2xlarge", "r7i.4xlarge",
"r7g.large", "r7g.xlarge", "r7g.2xlarge", "r7g.4xlarge",
"t3.large", "t3.xlarge", "t3.2xlarge",
"t4g.large", "t4g.xlarge", "t4g.2xlarge",
], var.ec2_instance_type)
error_message = "ec2_instance_type must be empty or one of the m7i, m7g, c7i, c7g, r7i, r7g (large to 4xlarge), t3 or t4g (large to 2xlarge) types."
}
}
variable "ec2_max_instances" {
description = "How many of the user's instances may run at once (1 to 4); each holds one workspace as expected use."
type = number
default = 1
validation {
condition = var.ec2_max_instances >= 1 && var.ec2_max_instances <= 4 && floor(var.ec2_max_instances) == var.ec2_max_instances
error_message = "ec2_max_instances must be a whole number from 1 to 4."
}
}
variable "ec2_user_tag" {
description = "The value of the triple-c-user principal tag the user signs in with; names their EC2 cluster (triple-c-<environment>--u-<tag>). Required with ec2_instance_type."
type = string
default = ""
validation {
condition = var.ec2_user_tag == "" || (
can(regex("^[A-Za-z0-9-]{1,32}$", var.ec2_user_tag))
&& !startswith(var.ec2_user_tag, "-")
&& !endswith(var.ec2_user_tag, "-")
&& !strcontains(var.ec2_user_tag, "--")
)
error_message = "ec2_user_tag must be 1 to 32 letters, digits and single hyphens, not starting or ending with a hyphen."
}
validation {
condition = var.ec2_instance_type == "" || var.ec2_user_tag != ""
error_message = "ec2_user_tag (the user's triple-c-user principal tag) is required when ec2_instance_type is set."
}
}
+7 -1
View File
@@ -459,6 +459,7 @@ fi
# That case says so rather than returning silently: the toggle is on, the
# capability is there, and the skill simply never appears — which is impossible
# to work out from the outside.
# >>> feature_skills
install_feature_skill() {
local _name="$1"
local _enabled="$2"
@@ -509,6 +510,7 @@ install_feature_skill() {
install_feature_skill pia-vpn "${VPN_SUPPORT_ENABLED:-0}"
unset VPN_SUPPORT_ENABLED
# <<< feature_skills
# ── Claude Code settings ────────────────────────────────────────────────────
# Apply the managed Claude Code settings to ~/.claude/settings.json, keeping
@@ -655,7 +657,10 @@ else
fi
# ── Docker socket permissions ────────────────────────────────────────────────
if [ "$triplec_aws" != 1 ] && [ -S /var/run/docker.sock ]; then
# Docker mode and AWS alike: an EC2 project with Docker access mounts its own
# instance's socket here (phase 7 E6); Fargate never has one.
# >>> docker_socket
if [ -S /var/run/docker.sock ]; then
DOCKER_GID=$(stat -c '%g' /var/run/docker.sock)
if ! getent group "$DOCKER_GID" > /dev/null 2>&1; then
groupadd -g "$DOCKER_GID" docker-host
@@ -663,6 +668,7 @@ if [ "$triplec_aws" != 1 ] && [ -S /var/run/docker.sock ]; then
DOCKER_GROUP=$(getent group "$DOCKER_GID" | cut -d: -f1)
usermod -aG "$DOCKER_GROUP" claude
fi
# <<< docker_socket
# ── Timezone setup ───────────────────────────────────────────────────────────
if [ -n "${TZ:-}" ]; then
@@ -0,0 +1,95 @@
# AWS Host Phase 7 (ECS on EC2) Implementation Plan
> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
**Goal:** A project can run on its owner's dedicated EC2 capacity instead of Fargate, which makes Docker access, VPN support and the full Claude Code sandbox available on AWS; the capacity scales to zero when unused.
**Architecture:** The onboarding module gains an optional per-user compute block: the user's own ECS cluster, an instance role and profile, a launch template on the ECS-optimized AMI, an Auto Scaling group (min 0) and a capacity provider with managed scaling, so instances exist only while a task needs one. An EC2 project registers an EC2-compatible task definition (awsvpc, the host's Docker socket and/or VPN device when enabled) and runs on that cluster through the capacity provider. Everything else (secret, EFS, agent tunnel, stop guard, idle stop, follower, port mappings) is phase 5–6 unchanged.
**Tech Stack:** OpenTofu (user-onboard module, hashicorp/aws 6.73.0 pinned), Rust (aws-sdk-ecs/-ec2 read, pricing), bash entrypoint, React/TS.
**Spec:** `docs/superpowers/specs/2026-11-13-aws-ecs-host-design.md` (Multi-user accounts → strict tenancy "Privileged and Docker allowed only on a per-user capacity provider"; ECS on EC2 vs Fargate; Feature compatibility "EC2 (dedicated)" column; Phase 7). Phase 5 and 6 rulings still bind.
## User decisions (2026-11-16)
- **Idle capacity scales to zero.** The Auto Scaling group has min 0; the capacity provider's managed scaling removes instances when no task needs them. Starts after idle take about 2–3 minutes longer (instance boot + image pull).
- **The instance type is set at onboarding** (default `m7i.large` for x86, `m7g.large` for ARM). Projects only choose Fargate or EC2.
- **No GPU in phase 7.**
## Decisions this plan makes (rulings)
- **E1. A per-user ECS cluster, not a capacity provider on the environment cluster.** `PutClusterCapacityProviders` replaces a cluster's whole list, and the environment module owns the shared cluster's list; two modules writing it would race and undo each other. The user's onboarding module creates `triple-c-<env>-<owner_key>` (the workspace-user policy's `ecs:cluster` condition `cluster/triple-c-*` already matches) with exactly one capacity provider. Strict isolation follows for free: nobody else's task can be placed on the user's instances.
- **E2. Compute is opt-in per onboarding.** New onboarding variables `ec2_instance_type` (empty = no EC2 compute, the default; the module creates nothing for EC2) and `ec2_max_instances` (1–4, default 1). Adding or removing it is an onboarding re-apply. `ONBOARD_MODULE_VERSION` 1 → 2 (re-apply shows "Update available"; existing users keep working on Fargate without it).
- **E3. Instance specifics:** ECS-optimized Amazon Linux 2023 AMI resolved at apply time from the public SSM parameter (`/aws/service/ecs/optimized-ami/amazon-linux-2023/recommended/image_id`, or `…/arm64/…` by the type's architecture); IMDSv2 required, hop limit 2 (tasks can't reach the instance role at hop 1 with awsvpc anyway — awsvpc tasks get no IMDS route; set `ECS_AWSVPC_BLOCK_IMDS=true` in the agent config); root volume gp3 50 GiB encrypted; instances in the environment's two subnets with the environment's tasks security group; `ECS_ENABLE_TASK_ENI` trunking is **not** enabled (one workspace per instance is the expected use; `ec2_max_instances` bounds concurrency).
- **E4. The instance role** `triple-c-<env>-<owner_key>-instance` carries the managed-policy-equivalent inline rights for the ECS agent (register/poll/submit, ECR pull, logs) and SSM core (for ECS Exec on EC2), under the permissions boundary — which therefore grows: the boundary module gains those actions (boundary version bump; an account admin re-applies it once). Instance profile of the same name.
- **E5. Admin tier grows a third file** `aws/policies/environment-admin-compute.json` (the two existing files are at 5,805 and 3,340 of 6,144): `autoscaling:*` on `autoScalingGroup:*:autoScalingGroupName/triple-c-*` with tag conditions, EC2 launch templates tag-scoped (`triple-c:component`), `ecs:CreateCapacityProvider`/`DeleteCapacityProvider`/`UpdateCapacityProvider`/`PutClusterCapacityProviders` on `capacity-provider/triple-c-*` and `cluster/triple-c-*`, `iam:*InstanceProfile*` on `instance-profile/triple-c-*`, `iam:PassRole` to `ec2.amazonaws.com` for `role/triple-c-*`, `ssm:GetParameter` on the public AMI parameters, `ec2:RunInstances` (with launch-template + tag conditions), `iam:CreateServiceLinkedRole` for `autoscaling.amazonaws.com`. Destruction tag-scoped as in phase 3. MODULE_CALLS gains every new provider call (drift test).
- **E6. Docker access on EC2 = the host's socket.** `allow_docker_access` on an EC2 project mounts `/var/run/docker.sock` from the instance (host volume) at the same path, and the entrypoint's existing Docker-socket group step runs in aws mode when the socket exists. That is root on the user's own instance — acceptable only because the instance is theirs alone (E1); the UI says so. **Privileged mode is not offered** separately (nothing in Triple-C needs it beyond Docker/VPN).
- **E7. VPN on EC2 = the same three pieces as local** (`vpn_host_config` parity): `NET_ADMIN`, the `/dev/net/tun` device, and the `net.ipv4.conf.all.src_valid_mark=1` system control (ECS `systemControls`, allowed for awsvpc). The AMI must have the `tun` module (AL2023 does); the live test checks a `wg` interface can be created.
- **E8. The Claude Code sandbox is allowed on EC2** (bubblewrap's user namespaces work on the ECS-optimized AMI; verify live — if not, keep it off like Fargate, P9).
- **E9. EC2 task sizing:** CPU/memory become reservations that must fit the instance type (the app knows the type's vCPU/memory from a small built-in table for the default families, validated at save; unknown types → only a "must fit" warning), no `ephemeralStorage` (instance disk), no Spot (EC2 Spot is a later item).
- **E10. Cost estimate for EC2** = the instance type's on-demand hourly price (Pricing API `AmazonEC2`, Linux, shared tenancy) while running; shown with "instances scale to zero when no workspace is running".
## Global Constraints
- Phase 5/6 constraints hold (sdk_config only; three tag keys on per-user resources; names only in `aws/tofu_modules.rs`; P1; policy + checks + README together, ≤ 6,144 each; three-part command registration; serde defaults + TS mirrors; live AWS only via the scratchpad `awsx`, cleaned up, spend cap $10 overall).
- Module versions and pinning tests move together (HCL `template_version` ↔ Rust consts ↔ default tags).
- Fargate projects behave exactly as before (fingerprints of existing Fargate plans unchanged — pinned tests).
- Every per-user EC2 resource carries `triple-c:component=user-onboarding` via default tags (admin destruction scoping).
## Review Focus
1. A user without EC2 compute picks EC2 for a project: refused at save with "Ask an admin to add EC2 capacity to your onboarding" — Task 4/6.
2. First start after idle: the capacity provider has to launch an instance; the start waits (progress "Starting an instance (about 3 minutes)…") instead of timing out at the Fargate wait — Task 5.
3. A task that doesn't fit the instance type: refused at save, not stuck PROVISIONING forever — Task 4/5.
4. Removing EC2 compute from an onboarding while an EC2 project exists: the next start says why — Task 5.
5. Docker access on: `docker ps` in the workspace works and lists only the user's own instance's containers — Task 7 (live).
---
### Task 1: Model and validation
**Files:** `models/project.rs`, `lib/types.ts`, `lib/awsWorkspace.ts`, `commands/project_commands.rs` (validate_aws_update).
**Produces:** `AwsWorkspaceConfig.launch_type: LaunchType { Fargate (default), Ec2 }`; validation: Ec2 forbids `spot` and `ephemeral_gib` ≠ default is ignored; Docker access and VPN allowed only when `launch_type == Ec2` for AWS projects; `ec2_instance_spec(type) -> Option<(vcpu_units, memory_mib)>` table for m7i/m7g/c7i/c7g/r7i/r7g/t3/t4g common sizes (large…4xlarge); fit check reserve 256 MiB + 128 CPU units for the agent/OS.
- [ ] Tests: old JSON → Fargate; Ec2 + spot refused; Docker/VPN refused on Fargate AWS projects, allowed on Ec2; fit check pass/fail; unknown type → Ok with warning flag.
- [ ] Commit `feat(aws): EC2 launch type setting`.
### Task 2: Policies — boundary and the admin compute file
**Files:** `aws/policies/permissions-boundary.json`, new `aws/policies/environment-admin-compute.json`, `aws/terraform/permissions-boundary/*` (version), `aws/policies.rs` (+ checks, `policy_files(EnvironmentAdmin)` three files), `aws/README.md`, `aws/live_tests.rs` (admin role attaches all three).
**Produces:** E4/E5 statements; boundary version bump; checks in lock-step; destruction tag-scoped; sizes recorded.
- [ ] Tests: lock-step + evaluator; another owner's ASG/launch template/capacity provider denied; PassRole to ec2 only for `role/triple-c-*`; all files ≤ 6,144.
- [ ] Commit `feat(aws): admin and boundary permissions for per-user EC2 capacity`.
### Task 3: Onboarding module — per-user EC2 compute
**Files:** `aws/terraform/user-onboard/{variables.tf,main.tf,ec2.tf (new),outputs.tf}`, `.terraform.lock.hcl` if providers change (they shouldn't), `aws/tofu_modules.rs` (+ tests: version 2, variables, `MODULE_CALLS`, names `user_cluster_name(env, owner_key)`, `user_capacity_provider_name`, `user_instance_role_name`), `aws/environments.rs` (onboarding outputs: `ec2_cluster_arn`, `ec2_capacity_provider`, `ec2_instance_type` — readable how? see note).
**Produces:** E1–E4 resources gated on `ec2_instance_type != ""` (`count`); outputs; `onboarding_complete` still depends on everything.
**Note (E11):** workspace users can't read the state bucket, so the module also writes a per-user SSM record `/triple-c/users/<env>/<owner_key>` (`{schema:1, instance_type, cluster_arn, capacity_provider}`, String, tagged) that the workspace user can read (`ssm:GetParameter*` on `parameter/triple-c/*`). It lives outside `/triple-c/environments/` so environment detection (GetParametersByPath there) never sees it. Task 2 must make sure the admin tier can put/delete parameters under `parameter/triple-c/users/*` (tag-scoped like the environment record).
- [ ] Tests (HCL + Rust): resources present only with the variable set; ASG min 0, max `ec2_max_instances`, `protect_from_scale_in` true with managed termination protection; capacity provider managed scaling target 100; IMDSv2 required; AMI parameter by arch; user record field list pinned.
- [ ] Commit `feat(aws): optional per-user EC2 capacity in the onboarding module`.
### Task 4: Onboarding UI + commands
**Files:** `commands/environment_commands.rs` (onboard + admin bundle carry `ec2_instance_type`, `ec2_max_instances`), `components/settings/hosts/*` (onboarding dialog: "EC2 capacity: none | m7i.large | m7g.large | other…", max instances), `aws/environments.rs` (`OnboardingState` gains `ec2: Option<UserCompute>` from the user record).
- [ ] Tests: vars passed through; re-apply keeps existing values unless changed; UI shows the user's compute.
- [ ] Commit `feat(aws): choose EC2 capacity when onboarding`.
### Task 5: EC2 task definition and run
**Files:** `aws/workspace.rs` (EC2 request: `requiresCompatibilities EC2`, awsvpc, no ephemeralStorage, Docker socket host volume + mount when enabled, VPN linuxParameters/devices/systemControls when enabled, sandbox per E8), `aws/tasks.rs` (run on the user's cluster with `capacityProviderStrategy=[{user CP, weight 1}]`; wait timeout 10 min for EC2 with the "Starting an instance…" progress while the task is PROVISIONING/PENDING without a container instance), `host/aws_host.rs` (cluster per project: environment cluster for Fargate, user cluster for EC2; reconcile/watch/stop use the project's cluster), `aws/runtime.rs` (user record read).
- [ ] Tests: Fargate request bytes unchanged (pinned fingerprints); EC2 request shape for each toggle combination; run request uses the user's cluster + CP; watch groups by cluster; missing user compute → clear error; wait progress strings.
- [ ] Commit `feat(aws): run projects on per-user EC2 capacity`.
### Task 6: Container + UI for EC2 projects
**Files:** `container/entrypoint.sh` (aws mode: Docker socket group step when `/var/run/docker.sock` exists; VPN skill install when `VPN_SUPPORT_ENABLED=1` as on Docker), `docker/aws_mode_tests.rs`, `HostSection.tsx` (Launch type: Fargate | EC2 (your instance: m7i.large) — disabled with the reason when the user has no EC2 compute; EC2 notes: scale-to-zero start delay; Docker access = root on your instance), `AccessSection.tsx`/`RuntimeSection.tsx` (Docker access / VPN enabled for EC2 projects), cost estimate (E10) in `aws/pricing.rs` + `AwsCostEstimate`.
- [ ] Tests: entrypoint gating; UI enable/disable per launch type; EC2 estimate arithmetic with a recorded `AmazonEC2` price fixture.
- [ ] Commit `feat(aws): EC2 projects in the UI, Docker access and VPN`.
### Task 7: Live verification
Via `scripts/aws-e2e-workspace.sh` (extend or add a test): onboard a workspace user with `ec2_instance_type = m7i.large`, start an EC2 project (Hybrid) with Docker access and VPN on: cold start time (instance launch), `docker ps` works, `ip link add wg0 type wireguard` (or `ip tuntap add`) works, bubblewrap check (E8), stop → capacity scales back to 0 instances within the managed-scaling window (record the time), second start time; remove; destroy; inventory incl. EC2 instances, ASGs, launch templates, capacity providers, instance profiles. Budget < $2.
### Task 8: Docs
Spec (phase 7 done, E1–E10 + user decisions + execution rulings, data model), `CLAUDE.md` (aws/, host/, container), `aws/README.md` (third admin file, boundary re-apply, EC2 compute, live results).
@@ -492,13 +492,14 @@ fails because of a quota, the app says so and doesn't retry.
|---|---|---|
| Start time | About 30–90 s (image pull every time; SOCI lazy-loading **to evaluate**) | Fast when the image is cached on a warm instance |
| Docker access / privileged / VPN | No | Yes, on dedicated capacity |
| GPU | No | Yes |
| Ops | None | ASG + capacity provider, defined in the environment module. The app only picks a capacity provider |
| Cost when idle | 0 (task stopped) | Instances keep running unless the ASG scales in |
| GPU | No | Not in phase 7 (no GPU, accelerator or metal types) |
| Ops | None | A per-user cluster, capacity provider, ASG and launch template, created by the user's onboarding module (phase 7). The app only picks the launch type |
| Cost when idle | 0 (task stopped) | The ASG has min 0 and managed scaling removes the instance about 17 minutes after the last task stops; a start after that is cold (3.5-4 min) |
| Max size | 16 vCPU / 120 GB | Instance type |
ECS on EC2 uses `awsvpc` network mode too, so security groups and the no-inbound rule work the same
way.
ECS on EC2 uses `bridge` network mode (phase 7, R2), not `awsvpc`: the instance is the user's alone,
so per-task ENIs buy nothing. The task egresses through the instance (the environment's tasks
security group, which has no inbound rules, is on the instance), and the no-inbound rule holds.
### Lifecycle and status
@@ -556,11 +557,11 @@ way.
| Speech-to-text | Yes (local STT, audio over the agent) | Yes | Yes |
| Scheduler, Mission Control | Yes, and they keep running with the app closed | Yes | Yes |
| Marketplace installs | Yes (files copied through the agent) | Yes | Yes |
| LLM gateway (LiteLLM) | Sidecar container in the task definition | Sidecar | Sidecar |
| LLM gateway (LiteLLM) | Sidecar container in the task definition | Sidecar | **No** (refused: bridge containers share no loopback) |
| Ollama / llama.cpp on the user's machine | **No**, unless the URL is reachable from AWS | No | No |
| Claude Code sandbox (bubblewrap) | Off (bubblewrap fails on Fargate, P9) | Weaker nested mode | Full |
| Docker access | No | No | Yes |
| VPN support | No | No | Yes |
| Claude Code sandbox (bubblewrap) | Off (bubblewrap fails on Fargate, P9) | Weaker nested mode | **Off** (bubblewrap fails under Docker's seccomp profile on EC2 too, measured) |
| Docker access | No | No | Yes (the instance's own socket; root on the user's instance) |
| VPN support | No | No | Yes (`NET_ADMIN`, `/dev/net/tun`, `src_valid_mark`; WireGuard device created live) |
| Snapshots | Setup script instead | Setup script | Setup script |
## Fit with existing conventions (CLAUDE.md)
@@ -611,7 +612,10 @@ pub host_id: Option<String>, // None = Local (all existing project
pub aws_placement: Option<AwsPlacement>, // { region, environment } — set when host_id is an AWS connection
pub aws: Option<AwsWorkspaceConfig>,
pub enum LaunchType { Fargate /* default */, Ec2 } // serde snake_case; absent in old JSON = Fargate
pub struct AwsWorkspaceConfig {
pub launch_type: LaunchType, // Ec2 = the user's own instance (see user record below)
pub cpu: u32, pub memory_mib: u32, pub arch: Arch, // X86_64 | Arm64 (Fargate's valid pairs)
pub storage: StorageMode, // Ephemeral | Hybrid (default) | Efs (Hybrid + /workspace on EFS)
pub ephemeral_gib: u32, // 21-200
@@ -628,7 +632,12 @@ pub struct AwsWorkspaceConfig {
pub workspace_access_point_id: Option<String>, // the EFS-mode /workspace access point
}
// EnvironmentInfo (the SSM record, parsed): max_workspace_hours: u32 (0 = no cap).
// Not built: launch_type (Fargate only; EC2 is phase 7).
// UserCompute (the user's EC2 record, SSM `/triple-c/users/<env>/<owner_key>`, written by their
// onboarding module only while it has EC2 capacity; workspace users can read it): schema,
// instance_type, cluster_arn (`triple-c-<env>--u-<tag>`), capacity_provider (same name),
// user_tag, max_instances. Absent = no EC2 capacity. On EC2, cpu/memory are reservations that
// must fit the instance type (10% memory margin; ECS registers less than nominal); no ephemeral
// size, no spot, no gateway sidecar. Switching the launch type back to Fargate clears Docker/VPN (R1).
```
Existing `projects.json` files deserialize unchanged (`host_id` missing → Local).
@@ -678,6 +687,99 @@ Execution rulings:
- **R7** Switching to copy-at-connect recreates the project secret (force delete, then create), because a new secret version would keep the old credentials as AWSPREVIOUS. Cost: a short delay on the first start after the switch.
- **R8** The sidecar's master key is random per start, so the local Docker gateway's keychain key never goes to AWS.
### Phase 7: rulings (2026-11-16)
**Done 2026-11-16.** Verified live (`aws/README.md` "Phase 7: ECS on EC2"): cold start about
3.5-4 minutes including the instance launch, a warm start 20-30 s, scale to zero about 17 minutes
after the last task stops, Docker access and VPN work, IMDS is blocked for the container, another
user's cluster is refused (RunTask denied), and the sandbox stays off on EC2 too.
User decisions:
- **Idle capacity scales to zero.** The group has min 0 and managed scaling removes the instance
when no task needs it; the next start waits for an instance (boot, ECS registration, image pull).
- **The instance type is chosen at onboarding** (`ec2_instance_type`, default `m7i.large` for x86 or
`m7g.large` for ARM, from a built-in table of supported families); a project only chooses
Fargate or EC2. Compute is opt-in per onboarding (`ec2_instance_type` empty = none).
- **No GPU** (and no accelerator or metal types).
- **Isolation by the `triple-c-user` principal tag.** An admin gives each EC2 user a tag value
(an Identity Center attribute for access control, or a tag on the IAM user or role); the user's
cluster is named after it, and the workspace policy denies `RunTask`/`StartTask` on every
per-user cluster except `cluster/triple-c-*--u-${aws:PrincipalTag/triple-c-user}`, and on all of
them when the tag is missing. Users without the tag can't use EC2. This amended E1.
Plan decisions, as built (E1-E11):
- **E1** A **per-user cluster**, not a capacity provider on the environment cluster:
`PutClusterCapacityProviders` replaces a cluster's whole list, so two modules writing the shared
cluster would undo each other. The cluster is `triple-c-<env>--u-<tag>` (environment names and
tags never hold `--`, so the pattern matches exactly the per-user clusters) with exactly one
capacity provider; the Auto Scaling group and launch template share the name.
- **E2** Opt-in per onboarding: `ec2_instance_type`, `ec2_max_instances` (1-4), `ec2_user_tag`.
`ONBOARD_MODULE_VERSION` 1 -> 2; existing users keep working on Fargate.
- **E3** The ECS-optimized AL2023 AMI (public SSM parameter, by architecture), IMDSv2 required,
**hop limit 1 and bridge networking (R2) instead of awsvpc**, gp3 50 GiB encrypted root, the
environment's subnets and tasks security group, no ENI trunking.
- **E4** The instance role `triple-c-<env>-<owner_key>-instance` (named by owner key: role names
are account-global) carries only the ECS agent's calls, which the **boundary scopes to the
cluster named by the role's `triple-c:cluster` tag** (`aws:PrincipalTag/triple-c:cluster`; the tag
must match `triple-c-*--u-*`, never the shared cluster). **The SSM core is dropped**: nobody gets
host sessions, and ECS Exec uses the task role. The boundary changed (version 2); an account
administrator re-applies it once.
- **E5** The admin tier grows to **four files**: `environment-admin-compute.json` (capacity
providers, launch templates, `RunInstances` rules, the instance role's PassRole to EC2, the AMI
parameter) and `environment-admin-scaling.json` (Auto Scaling groups, instance profiles, the
service-linked roles). Every launch template and group is **pinned to a numeric template
version** and checked on create: `$Latest`/`$Default` would let a template edited after the check
carry any instance type or role (a PassRole bypass), so they are denied. A capacity provider over
someone else's group is refused (ECS checks `CreateOrUpdateTags` on the group).
- **E6** Docker access on EC2 is the instance's own socket (host volume), the entrypoint's
docker-socket group step running in aws mode. That is root on the user's own instance, and the
UI says so.
- **E7** VPN on EC2 is `NET_ADMIN`, the `/dev/net/tun` device and `src_valid_mark=1`, as locally.
- **E8** **The sandbox is off on EC2 as on Fargate.** Bubblewrap fails ("No permissions to create
new namespace") because Docker's default seccomp profile refuses `unshare` without
`CAP_SYS_ADMIN`, although the kernel allows user namespaces; the toggle is locked off for every
AWS project.
- **E9** EC2 CPU/memory are reservations that must fit the instance type (ECS registers less
memory than nominal, so a 10% margin), no ephemeral size, no Spot.
- **E10** The cost estimate is the instance type's on-demand hourly price while it runs.
- **E11** **The user record** `/triple-c/users/<env>/<owner_key>` (schema, instance_type,
cluster_arn, capacity_provider, user_tag, max_instances) is written by the onboarding module only
while it has EC2 capacity. Workspace users may read it (they can't read the state bucket), and it
lives outside `/triple-c/environments/` so environment detection doesn't list it.
Execution rulings:
- **R1** A Fargate AWS project can't keep Docker access or VPN: no grandfathering. Switching the
launch type to Fargate clears both flags (the user re-enables them after switching back).
- **R2** EC2 tasks use **bridge** networking. The instance is single-tenant, so per-task ENIs buy
nothing, awsvpc tasks on EC2 get no public IP (no internet in public-mode environments without
a NAT), and bridge keeps IMDS out of reach through the hop limit. Cost: the ECS agent's
introspection port (`172.17.0.1:51678`: cluster, tasks, no credentials) is reachable from the
container; accepted, since the instance runs only its user's tasks.
- **R3/R4** A pre-apply **collision check**: onboarding refuses a `triple-c-<env>--u-<tag>` cluster
or group that belongs to someone else, because the module would adopt it (R3). Ownership is the
resources' `triple-c:owner` tag, with the user record as fallback; only other-owned or
unowned names are refused, so an onboarding's own half-built cluster doesn't block its re-apply
(R4, amending R3). Admin bundles carry the same check as commands in their `README.txt`.
- **The gateway sidecar is refused on EC2.** In bridge mode a task's containers don't share a
loopback, so the sidecar on 127.0.0.1:4000 wouldn't be reachable; a host port would expose it on
the instance's network. Saves and starts of an EC2 project with it are refused.
- **The cold-start rerun.** ECS turns `enableExecuteCommand` **off** for a RunTask that has to
wait for an instance (measured in CloudTrail: true in the request, false in the answer, and the
task keeps it off), so the agent tunnel can't open. On a cold start the first task only brings
the instance up; once it is placed on an instance and desired RUNNING, the app stops it, waits
for STOPPED and runs the task again, requiring exec on (else it stops that task and fails with
`EXEC_OFF`). A live task found with exec off is stopped and replaced. Cost: about 30-60 s on a
cold start.
- **Race rules for the rerun.** Both launches go through `find_or_run`, so a concurrent start's
live task is never doubled (`START_RACE`, with one RunTask only). A first task that stops by
itself ends the start with its stop reason, with no second RunTask. A task still waiting for an
instance with exec off is another start's (`INSTANCE_RACE`) while it is younger than the
instance wait plus 2 minutes; an older one is an orphan and is replaced. A first task stopped
with `REPLACED_REASON` says another start replaced it. A stop gets its own 3-minute wait.
## Shipped AWS artefacts
```
@@ -723,8 +825,13 @@ TLS only, tagged `triple-c:component=state`), one object per deployment (D9).
"Phase 6: one workspace with every option"): EFS persistence, the setup script, port mapping,
the sidecar, copy-at-connect and the cost estimate. Not exercised live: the lifetime cap firing
(unit-tested), a v1 → v2 environment update, a private clone retried after connect.
7. **ECS on EC2.** Per-user capacity providers in the onboarding module, privileged/Docker/VPN on
dedicated capacity.
7. **ECS on EC2.** Per-user capacity (cluster, capacity provider, Auto Scaling group at min 0,
launch template) in the onboarding module, Docker and VPN on that dedicated capacity.
**Done 2026-11-16** (rulings: "Phase 7: rulings"). Verified live (`aws/README.md` "Phase 7:
ECS on EC2"): cold start about 3.5-4 min including the instance launch, warm 20-30 s, scale to
zero about 17 min after the last task, Docker and VPN work, IMDS blocked, cross-user RunTask
denied, sandbox off on EC2 too. Not exercised live: an Identity Center attribute as the
`triple-c-user` source, an untagged user, creating the service-linked roles.
8. **Standard tenancy** (opt-in), only if demand appears.
## To verify before or during implementation
@@ -834,6 +941,21 @@ detached runner, detection, self-onboarding, an update ("No changes"), destroy;
file-system policy accepted the just-created task role; conditional `DeleteObject` (`If-Match`)
on the versioned bucket returns 412 on a stale ETag.
- **ECS on EC2, measured 2026-11-16** (phase 7, `aws/README.md` "Phase 7: ECS on EC2"):
- A per-user capacity (cluster, capacity provider, Auto Scaling group at 0, launch template)
applies as the environment admin once the scaling file stops requiring
`autoscaling:InstanceTypes`. A template-only group create doesn't send it, so the create's
own launch check (`ec2:InstanceType`) is what holds the type.
- ECS turns `enableExecuteCommand` off on a RunTask that has to wait for an instance, so a cold
start runs the task a second time once the instance is up. A cold start takes about 4 minutes.
- Bridge networking works for the agent tunnel (ECS Exec port forward), Docker access (host
socket) and VPN (`NET_ADMIN`, `/dev/net/tun`, `src_valid_mark`).
- Bubblewrap **does not** work in an ECS task on EC2 either (Docker's seccomp profile), so the
sandbox stays off on AWS.
- IMDS gives a bridge container no token. The agent's introspection port is reachable from it
(metadata only).
- The `triple-c-user` principal tag (a role tag) confines RunTask to the user's own cluster.
Still to verify: the `existing` network mode against a user's own VPC. Fargate tasks in the
deployed environment were run in phase 5 (`aws/README.md`).