Fix data race of sub.seq during replay -> live tail cutover #43
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
Prevent a data race that leads to out-of-order events being delivered to jetstream subscribers during cutover from replay to live-tailing, partially resolving [bluesky-social#42](bluesky-social#42). `emitToSubscriber` reads and writes `sub.seq`, but it does not itself lock the subscriber's `sub.lk`. It can't, because its main caller, the server `Emit` function, takes that lock across its call. The other caller is the replay goroutine, which runs concurrently, and did not take the lock, leaving `sub.seq` exposed to races when both it and `Emit` ran concurrently (for ~0.5s during cutover). This change makes the replay goroutine take the lock before calling `emitToSubscriber`, protecting `sub.seq` from this race.
uniphil
commented
Feb 5, 2025
| @@ -215,6 +215,8 @@ func (s *Server) HandleSubscribe(c echo.Context) error { | |||
| go func() { | |||
| for { | |||
| lastSeq, err := s.Consumer.ReplayEvents(ctx, sub.compress, *sub.cursor, playbackRateLimit, func(ctx context.Context, timeUS int64, did, collection string, getEventBytes func() []byte) error { | |||
| sub.lk.Lock() | |||
| defer sub.lk.Unlock() | |||
| return emitToSubscriber(ctx, log, sub, timeUS, did, collection, true, getEventBytes) | |||
There was a problem hiding this comment.
hmm. emitToSubscriber with playback = true would do a blocking send on the outbox while holding the subscriber's lock. could this hold up the Emit semaphore when it tries to acquire the sub's lock in the overlap time of cutover?
|
closing for #45 |
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Outdated suggestions cannot be applied.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.
Prevent a data race that leads to out-of-order events being delivered during cutover from replay to live-tailing, partially resolving #42.
emitToSubscriberreads and writessub.seq, but does not itself lock the subscriber'ssub.lk. It can't, because its main caller, the serverEmitfunction, takes that lock across its call toemitToSubscriber.The other caller is the replay goroutine, which runs concurrently, and did not take the lock, leaving
sub.seqexposed to races when both it andEmitran concurrently (for ~0.5s during cutover).This change makes the replay goroutine take the lock before calling
emitToSubscriber, protectingsub.seqfrom this race.I've verified locally that jetstream still works and cuts over, and created a pair of repro scripts that (try to) trigger the data race: https://gist.github.com/uniphil/346acb62088022729394d2324bf2ad8a
With the changes from this PR, the local repro goes from almost always triggering the race to never in all tests so far -- see sample output in the linked gist. I'll verify prod again if this gets merged and deployed 🙂