Fix shutdown panics by separating completer context #401
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This failing test case exposes the issue in #400 100% of the time, which is caused by the `stopProducers()` call not actually waiting until the producers are fully shut down before proceeding with the remaining shutdown.
bgentry
force-pushed
the
bg-fix-completer-send-on-closed-channel-panic
branch
from
389aaf9
to
93532e8
Compare
bgentry
changed the title
bgentry
force-pushed
the
bg-fix-completer-send-on-closed-channel-panic
branch
from
93532e8
to
ca03eaa
Compare
bgentry
changed the title
Back in #258 / 702d5b2, the batch completer was added to improve throughput. As part of that refactor, it was turned into a startstop service that took a context on start. We took the care to ensure that the context provided to the completer was _not_ the `fetchCtx` (cancelled on `Stop()`) but instead was the raw user-provided `ctx`, specifically to make sure the completer could finish its work even after fetches were stopped. This worked well if the whole shutdown process was done with `Stop` / `StopAndCancel`, but it did not work if the user-provided context was itself cancelled outside of River. In that scenario, the completer would immediately begin shutting down upon cancellation, even without waiting for producers to finish sending it any final jobs that needed to be recorded. This went unnoticed until #379 / 0e57338 turned this scenario into a panic instead of a silent misbehavior, which is what was encountered in #400. To fix this situation, we need to use Go 1.21's new `context.WithoutCancel` API to fork the user-provided context so that we maintain whatever else is stored in there (i.e. so anything used by slog is still available) but we do not cancel this completer's context _ever_. The completer will manage its own shutdown when its `Stop()` is called as part of all of the other client services being stopped in parallel.
bgentry
force-pushed
the
bg-fix-completer-send-on-closed-channel-panic
branch
from
ca03eaa
to
9f4d1f1
Compare
bgentry
changed the title
Merged
brandur
added a commit
that referenced
this pull request
Jun 26, 2024
|
could this have been causing stuck jobs on graceful shutdown? |
|
@elee1766 I believe so, we should add that to the changelog. The core issue is the completer (marks completed jobs in the db) wasn’t properly waiting for jobs to finish executing due to a refactor. So I think it could have resulted in some jobs not being marked as finished when they had in fact executed and returned. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Back in #258 / 702d5b2, the batch completer was added to improve
throughput. As part of that refactor, it was turned into a startstop
service that took a context on start. We took the care to ensure that
the context provided to the completer was not the
fetchCtx(cancelled on
Stop()) but instead was the raw user-providedctx,specifically to make sure the completer could finish its work even after
fetches were stopped.
This worked well if the whole shutdown process was done with
Stop/StopAndCancel, but it did not work if the user-provided context wasitself cancelled outside of River. In that scenario, the completer would
immediately begin shutting down upon cancellation, even without waiting
for producers to finish sending it any final jobs that needed to be
recorded. This went unnoticed until #379 / 0e57338 turned this scenario
into a panic instead of a silent misbehavior, which is what was
encountered in #400.
To fix this situation, we need to use Go 1.21's new
context.WithoutCancelAPI to fork the user-provided context so that wemaintain whatever else is stored in there (i.e. so anything used by slog
is still available) but we do not cancel this completer's context
ever. The completer will manage its own shutdown when its
Stop()iscalled as part of all of the other client services being stopped in
parallel.
Fixes #400.