amadan
main's own CI at 26fbd042 (s3-polling-336) failed on this test too, the worker subcase this time: TestAMoveOntoTheInstalledGoodBuildWaitsForASignedRollback/worker (5.02s), same 5 s waitFor. So main is red on it now, not only branches.
Flaky test: internal/rollout TestAMoveOntoTheInstalledGoodBuildWaitsForASignedRollback/socket_instance (about 1 in 20) open
Seen while building db-region-forward-2 (2026-10-22). The branch does not touch internal/rollout or anything its test binary links (it adds only a _test.go file to internal/edge), so the test binary is main's.
What fails: TestAMoveOntoTheInstalledGoodBuildWaitsForASignedRollback/socket_instance, at holdran_test.go:796: timed out waiting for the move onto v1. That is waitFor's fixed 5 s deadline (internal/rollout/fake_test.go:533) after the signed rollback's pointer pass.
Rates:
- CI: failed on two consecutive branch runs (276d9e0e and 5857db84, both in the gate's internal/rollout run, 5.0 s each), passed on the next two (753f47d8, d094a4a0). main's own run at 13e2f75f passed.
- Locally (DOCS-T1, amd64, niced):
go test ./internal/rollout/ -run TestAMoveOntoTheInstalledGoodBuildWaitsForASignedRollback -count=20failed once in 20 on the first try, 0 in 20 on the second; three plain runs passed; the whole package passed once under -race.
So it looks load-sensitive: the 5 s wait loses to a busy runner. Not investigated beyond that.