This commit adds some rudimentary functionality for cleaning up
connections on controller failure, or more specifically, a failure to
communicate with the controllers.
In this scenario, when a status cannot be sent for any reason for longer
than 30 seconds, the worker will close all connections that are normally
not cleaned up (so for sessions that currently are *not* in the progress
of being canceled or terminated). Status requests now time out after 10
seconds if they are not made successfully.
Session status is not touched during this process, but natural cleanup
or connectionless sessions will still occur. This should be okay, as the
session should be added on re-connect if need be.
A few other changes are made to cancelConnections: first, error logging
has been moved to the function so that we can handle errors requesting
cancellation directly within the function; this entails simply just
locally marking these connections as closed with a warning. Second,
there was a minor logging bug where closed connections were not properly
being noted in the last trace log, this has been fixed.
Additional changes:
* Circle: Set acceptance test no_output_timeout to 30m
This ensures that cors_allowed_origins can be set when cors_enabled is
not explicitly specified and allowed to be its default (currently true).
The current behavior is to clobber this value when cors_enabled is not
explicitly set, ultimately setting it to the default of
"serve://boundary" only.
This PR changes delete behavior:
* On success, a 204 is returned rather than a 200 with an empty JSON object
* The CLI now passes through the 404 error if encountered instead of swallowing it
It also updates formatting for JSON to use the response body, which means that the item (or items) in the JSON format are actually the format across the wire, instead of being reinterpreted by unmarshaling into the Go SDK and then remarshaling.
Because of listing visibility, if no actions are assigned to a user for
an individual resource, it's excluded from the list. This is fine and
good. However, when we made a change in 0.2.0 to restrict the
information returned from listing for the anonymous user for scopes and
auth methods, it exposed the fact that because of the way this
visibility works we had to add `read` to scopes by default -- thereby
rather defeating the purpose of restricting what goes back to `u_anon`
requests.
The purpose of "no-op" is to never be used for real actions within the
API, but as something that can be granted to specifically enable
visibility for the targeted user. This makes the controls over list
visibility much more granular.
Current behavior is that you need list grant on the parent scope when
recursively listing, and this is sufficient. This is good for
administrative workflows but can be unintuitive for resources that only
exist at a project level. This change allows permissions on the child
scope to also be taken into account, so that you can start a recursive
list from a parent scope without direct list support on that scope but
still get back child scopes for which list access is granted.
This also fixes a bug where visibility of a scope for recursive listing
would be based on whether the user had list resources for roles within
that scope instead of for the resource type in question.
Joe Thompson
Jim
Jeff Mitchell
Chris Marchesi
Louis Ruch
Randall Morey
Jeff Malnick
Todd Knight
