pomerium / pomerium / 39712920246

54%

Build Type

push

github

Committed by web-flow

Commit Message

fix(databroker): don't report a stopped follower as unhealthy (#6878)

## Summary

Stopping a clustered databroker follower could report the
`databroker.cluster` health check as an error. A follower is stopped
whenever the clustered server is rebuilt, for example on a leader change
or a databroker config change. The result depended on which part of the
follower's sync loop noticed the stop first. If the request waiting on
the leader returned first, its gRPC `Canceled` error was reported as a
health failure instead of a normal shutdown. The next server's report
then overwrote it, so cluster health flapped briefly.

A stopped follower now always reports a normal shutdown. Real sync
failures are still reported as errors.

This also fixes the flaky `TestClusteredFollowerServer/sync`. The health
manager replays the last status to a newly registered provider, so that
subtest could start out holding the error left by an earlier subtest's
follower shutdown.

## Related issues

Follow-up to #6831 (ENG-4324).

## User Explanation

None.

## AI disclosure

Claude Code found the cause from the CI failure and wrote the fix and
test; I reviewed.

## Checklist

- [x] reference any related issues
- [x] updated unit tests
- [x] add appropriate label (`bug`)
- [x] disclosed AI usage (or wrote "none") per AI_POLICY.md
- [ ] ready for review

Coverage Stats

14 of 14 new or added lines in 1 file covered. (100.0%)

13 existing lines in 5 files now uncovered.

35065 of 65252 relevant lines covered (53.74%)

316.5 hits per line