Repository navigation
Conversation
Give the MPS ordering check the same filesystem deadline used by other non-timeout Dragon tests so slow cleanup does not mask the initialization behavior under test. Signed-off-by: Trent Nelson <trentn@nvidia.com>
Signed-off-by: Trent Nelson <trentn@nvidia.com>
|
No actionable comments were generated in the recent review. Walkthrough
Merge Risk:
|
melo-gonzo
left a comment
There was a problem hiding this comment.
Thanks. I read this at 1327383. This is from code reading. I did not run the tests.
- The Dragon MPS case now gets the same 5-second
result_timeoutas the other real-worker tests. #183's PR run hit this flake: theconnected=Truecase returnedfailed, and the same test passed on the vetted copy of that commit. - The NOOA cancel case now waits up to 30 seconds for startup under a 60-second run deadline, so the deadline can no longer fire before the cancel. The
finallycancels and reaps the worker even when startup times out, and thatTimeoutErrorstill fails the test. The deadline case keeps its 8-second limit.
No changes needed.
Summary
The NOOA cancellation test can exhaust its startup allowance before reaching the cancellation check. Give startup its own bounded wait and always cancel and reap the worker if that wait fails, while retaining the separate deadline test and bounded cleanup assertion.
The Dragon MPS success test also inherits a 0.2-second result/close timeout. Use the same 5-second allowance as neighboring real-worker tests. A delayed-close check reproduces this timing sensitivity; it does not establish the precise cause of the observed CI failure.
Validation
Repository hygiene
Tests only; no product behavior, dependency, generated artifact or release changes.