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
{{ message }}
Repository navigation
fix(roadmap): alternatives ADR cap test reads logs of other loggers #4062
tests/roadmap/test_roadmap_alternatives.py::test_the_decisions_are_cut_by_whole_entries_the_oldest_go_first_and_a_warning_says_so failed in one full local devops ci run on a loaded host. The test passes alone (166 passed in the file) and passes in the four GitHub shards.
Cause
The test compares the list of ALL warning records with the one expected warning. In the failing run, eight extra records came from the logger opentelemetry.sdk.trace.export with the text "Processor is already shutdown, ignoring call". A test that ran before it in the same xdist worker shut a span processor down and left the tracer state behind.
Expected
The test reads only the records of the logger devops_cli.roadmap.adr. The test that leaves the shut processor behind is found and its tracer state is reset.
Steps
Run uv run devops ci on a loaded host, or run the telemetry tests and then the file in one worker.
The test fails on PR #4077 (shard 2 of 4, run 42416842507): the log records hold 'Processor is already shutdown, ignoring call' lines next to the expected warning. #4077 does not touch this test or alternatives code. I re-ran the failed job.
PR #4077 filters the log records of three alternatives tests by logger name (the first half of Expected). The second half stays open: find the test that leaves the shut span processor and reset its tracer state. The shard of #4077 failed twice on this test, so the order of tests in the worker made it fail every time there.
Activity
dan-petty commentedon Nov 7, 2026
Intake placed this item to the backlog: intake puts every new item in the backlog. A person plans the milestones.
fixprefix of the title gives the type.(roadmap)scope of the title gives the scope.A flaky test under load causes CI noise but does not block a user-facing feature.It removes a recurring CI failure that blocks merge confidence.The fix is to add a logger-name filter in one test and reset tracer state in another.Both fix test pollution from telemetry state in xdist workers, but the candidate filters log records by logger name while #3532 stops a logfire span from writing to stdout.dan-petty commentedon Nov 7, 2026
The test fails on PR #4077 (shard 2 of 4, run 42416842507): the log records hold 'Processor is already shutdown, ignoring call' lines next to the expected warning. #4077 does not touch this test or alternatives code. I re-ran the failed job.
test(roadmap): alternatives log tests read records of their own logger
dan-petty commentedon Nov 7, 2026
PR #4077 filters the log records of three alternatives tests by logger name (the first half of Expected). The second half stays open: find the test that leaves the shut span processor and reset its tracer state. The shard of #4077 failed twice on this test, so the order of tests in the worker made it fail every time there.
fix(roadmap): retitle records a skip mark for an item it leaves (#4077)