You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Merge pull request kubernetes#61195 from grantr/use-race-free-fake-watcher
Automatic merge from submit-queue (batch tested with PRs 61195, 61479). If you want to cherry-pick this change to another branch, please follow the instructions <a href="https://github.com/kubernetes/community/blob/master/contributors/devel/cherry-picks.md">here</a>.
Use RaceFreeFakeWatcher in ObjectTracker to fix racy watch panics
**What this PR does / why we need it**:
The `FakeWatcher` added to `ObjectTracker` in kubernetes#57504 allows sends on the result channel after it's closed; for example calling `Stop()` then `Add(obj)` will cause a panic. In my experience this has led to flaky tests when informers and controllers are running.
Replacing `FakeWatcher` with `RaceFreeFakeWatcher` fixes the problem, since `RaceFreeFakeWatcher` ignores additional events that occur after the watcher is stopped. It also panics instead of blocking when the result channel is full, which seems like a more useful behavior in tests than blocking.
I removed the `FakeWatchBufferSize` constant since `RaceFreeFakeWatcher` doesn't take a buffer size argument anymore. This seems fine since the `DefaultChanSize` constant is close to the `FakeWatchBufferSize` value (100 vs 128).
**Special notes for your reviewer**:
I can provide a minimal repro of a flaky test caused by the earlier behavior if necessary.
**Release note**:
```release-note
Fix racy panics when using fake watches with ObjectTracker
```
0 commit comments