Add an extra cleanup task for sessions stuck in canceling state (#2229)

* Add an extra cleanup task for sessions stuck in canceling state

See the comment on the migrations file for details.

CHANGELOG.md

@@ -6,16 +6,21 @@ Canonical reference for changes, improvements, and bugfixes for Boundary.
 
 ### Bug Fixes
 
-* scheduler: Removes a Postgres check constraint on the length of the controller id.
-  The constraint causes an error when running jobs with controllers configured
-  with a name of less than 10 character:
-  `controller_id_must_be_at_least_10_characters constraint failed`.
-  Since a controller's name is used as its private id, Boundary does not enforce
-  the normal length requirements on the `server_controller` table
+* scheduler: Fix regression causing controller names of less than 10 characters
+  to fail to register jobs
   ([PR](https://github.com/hashicorp/boundary/pull/2226)).
+* sessions: Fix an additional case from the changes in the 0.8.x series that
+  could result in sessions never moving from `canceling` state to terminated.
+  ([PR](https://github.com/hashicorp/boundary/pull/2229))
 
 ## 0.9.0 (2022/06/20)
 
+### Known Issues
+
+* If a controller's defined name in a configuration file is less than 10
+  characters, errors may be seen on startup related to registration of jobs.
+  This is a regression in this version and will be fixed in the next release.
+
 ### New and Improved
 
 * PKI Workers: This release introduces a new worker type `pki` which

internal/db/schema/migrations/oss/postgres/35/02_cleanup.up.sql

@@ -0,0 +1,74 @@
+begin;
+
+-- This adds additional cleanup to what went into 0.8.x. There could be a case
+-- where sessions can persist in canceled state even if there are no connections
+-- for that session (possibly due to cleanup) and will never be transitioned
+-- because connection state cannot be determined.
+--
+-- If they have previously been marked terminated in the warehouse we delete
+-- them; if not we mark them terminated so it's recorded in the warehouse and
+-- the cleanup job should eventually clear these out.
+with
+  canceling_sessions(session_id) as (
+    select session_id
+      from session_state
+     where state = 'canceling'
+       and end_time is null
+  ),
+  no_connection_canceling_sessions as (
+    select session_id
+      from canceling_sessions
+     where session_id not in
+             (
+               select session_id
+                 from session_connection
+             )
+  ),
+  has_termination_reason (session_id) as (
+        select public_id
+          from session
+         where public_id in
+                 (
+                   select session_id
+                     from no_connection_canceling_sessions
+                 )
+           and termination_reason is not null
+  ),
+  terminated_in_warehouse (session_id) as (
+        select session_id
+          from has_termination_reason
+         where session_id in
+                (
+                  select session_id
+                    from wh_session_accumulating_fact
+                   where session_terminated_date_key <> -1
+                )
+  ),
+  not_terminated_in_warehouse (session_id) as (
+        select session_id
+          from has_termination_reason
+         where session_id not in
+                 (
+                   select session_id
+                     from terminated_in_warehouse
+                 )
+  ),
+  delete_terminated_in_warehouse_from_session as (
+        delete from session
+         where public_id in
+                 (
+                   select session_id
+                     from terminated_in_warehouse
+                 )
+  )
+  update session
+    set
+      version = version + 1,
+      termination_reason = 'canceled'
+    where public_id in
+            (
+              select session_id
+              from not_terminated_in_warehouse
+            );
+
+commit;