etcd: fix paginated list missing events with parallel operations #34091
Conversation
This file contains hidden or 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
Currently, the etcd paginatedList implementation is affected by a bug that can lead to missing events of both upsertions and deletions that are performed between the end of the first Get call and the last Get call. Indeed, the tracked revision is incorrectly updated after every Get call, and etcd always returns the current revision, regardless of whether WithRev is actually requesting a specific revision. In turn, this leads to subsequent Get calls targeting different revisions, hence missing all the events happened in between for the already processed chunks of the prefix, as the watch operation is eventually started from the latest retrieved revision. Let's address this by consistently using the same revision during the entire paginatedList execution, to ensure that we correctly list all entries at that revision, and subsequently start watching events from the next one. Additionally, let's update the associated unit test to additionally cover the parallel operations case and prevent future regressions in this respect. This bug affected both Cilium running in KVStore mode and clustermesh, although the race condition window is sufficiently short to trigger its occurrence only rarely and in high scale environments. Fixes: e33b9f9 ("kvstore: add support for paginated lists in etcd.ListAndWatch") Signed-off-by: Marco Iorio <marco.iorio@isovalent.com>
|
/test |
bimmlerd
approved these changes
Jul 30, 2024
|
@joestringer this is the PR mentioned in the community meeting, maybe you could use your maintainer's priviledges to approve? |
joestringer
approved these changes
Aug 2, 2024
|
The changes in this PR have been introduced after we replaced gocheck with built-in go test in #32261 and are thus a tad annoying to backport as-is to 1.14 and 1.15, which still use gocheck, so I'm marking this as |
13 tasks
This was referenced Aug 12, 2024
3 tasks
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.
Currently, the etcd paginatedList implementation is affected by a bug that can lead to missing events of both upsertions and deletions that are performed between the end of the first Get call and the last Get call. Indeed, the tracked revision is incorrectly updated after every Get call, and etcd always returns the current revision, regardless of whether WithRev is actually requesting a specific revision. In turn, this leads to subsequent Get calls targeting different revisions, hence missing all the events happened in between for the already processed chunks of the prefix, as the watch operation is eventually started from the latest retrieved revision.
Let's address this by consistently using the same revision during the entire paginatedList execution, to ensure that we correctly list all entries at that revision, and subsequently start watching events from the next one. Additionally, let's update the associated unit test to additionally cover the parallel operations case and prevent future regressions in this respect.
This bug affected both Cilium running in KVStore mode and clustermesh, although the race condition window is sufficiently short to trigger its occurrence only rarely and in high scale environments.
Marked for backport to all affected versions, as IMO it fits into the
category, considering the Cilium then only recovers after a restart of the agent (although that may again trigger the same problem).