Skip to content

test: fix two flaky integration tests#1324

Merged
bigbes merged 1 commit into
v3from
bigbes/fix-flaky-integration-tests
Jul 20, 2026
Merged

test: fix two flaky integration tests#1324
bigbes merged 1 commit into
v3from
bigbes/fix-flaky-integration-tests

Conversation

@bigbes

@bigbes bigbes commented Jul 17, 2026

Copy link
Copy Markdown
Collaborator

Two integration tests raced a deadline that a loaded CI runner could blow
past. Both were observed failing on unrelated PRs, and both re-ran green on
an unchanged tree, so neither was a real product failure.

Related to TNTP-8611

Both tests raced a deadline that a loaded CI runner could blow past, and
both failed in a way that pointed away from the actual cause.

check_restart discarded the result of wait_pid_files_changed, and
wait_event reports a timeout by returning False. Once the 5s wait expired
the test carried on regardless and asserted on a PID file that was not
written yet, which surfaces as a bare 'NOT RUNNING' == 'RUNNING' rather
than as the timeout it is. Assert the wait and give it 30s; a restart that
is merely slow no longer fails, and one that is genuinely stuck now names
wait_pid_files_changed instead of the status check ten lines below.

The cluster auth tests published to a live endpoint with ?timeout=0.1
while asserting on a server-side auth error, so the 100ms deadline raced
the server's reply and won often enough to fail: the assert then compared
against 'context deadline exceeded'. These are the only live-endpoint
callers still on 0.1 -- the rest of the cluster suite already uses
timeout=5 -- so this brings them in line. The unreachable-host cases keep
their short timeout, where a fast deadline is the point.
@bigbes
bigbes merged commit 60bd413 into v3 Jul 20, 2026
12 checks passed
@bigbes
bigbes deleted the bigbes/fix-flaky-integration-tests branch July 20, 2026 10:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants