compute: handle future epochs in channel adapter #32017
Merged
+35
−13
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
Previously the compute command channel adapter assumed that it is impossible to receive responses with epoch values greater than the channel adapter's current epoch. This assumption is wrong because the Timely cluster-side of workers learns about new connection epochs by observing commands from worker 0, which can be ahead of other workers' epochs. This commit removes the assumption and instead makes the channel adapter handle responses with future epochs by stashing them until sufficiently many client reconnects have been observed.
teskje
commented
Mar 27, 2025
Comment on lines
+773
to
+793
| // Serve this connection until we see any of the channels disconnect. | ||
| loop { | ||
| let (resp, resp_epoch) = match stashed_response.take() { | ||
| Some(stashed) => stashed, | ||
| None => match serve_rx_channels() { | ||
| Ok(response) => response, | ||
| Err(()) => break, | ||
| }, | ||
| }; | ||
|
|
||
| if resp_epoch < epoch { | ||
| // Response for a previous connection; discard it. | ||
| continue; | ||
| } else if resp_epoch > epoch { | ||
| // Response for a future connection; stash it and reconnect. | ||
| stashed_response = Some((resp, resp_epoch)); | ||
| break; | ||
| } else { | ||
| // Response for the current connection; forward it. | ||
| if response_tx.send(resp).is_err() { | ||
| break; |
There was a problem hiding this comment.
This would all be much nicer if we could have a crossbeam_channel::Receiver wrapper with a builtin stash that you could push_front to. Unfortunately crossbeam_channel::select doesn't support other types so I had to move the stash outside and ended up with this somewhat convoluted control flow.
|
TFTR! |
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.
Previously the compute command channel adapter assumed that it is impossible to receive responses with epoch values greater than the channel adapter's current epoch. This assumption is wrong because the Timely cluster-side of workers learns about new connection epochs by observing commands from worker 0, which can be ahead of other workers' epochs.
This PR removes the assumption and instead makes the channel adapter handle responses with future epochs by stashing them until sufficiently many client reconnects have been observed.
Motivation
Fixes https://github.com/MaterializeInc/database-issues/issues/9124
Tips for reviewer
I've tried to write a regression test using mzcompose and toxiproxy, spawning a multi-process cluster and resetting its controller connection in a loop. Didn't manage to reproduce the race that leads to the "epoch from the future" panic though.
Checklist
$T ⇔ Proto$Tmapping (possibly in a backwards-incompatible way), then it is tagged with aT-protolabel.