moonchen opened a new issue, #13445:
URL: https://github.com/apache/trafficserver/issues/13445
The `h2origin` gold test (`tests/gold_tests/h2/h2origin.test.py`) is flaky
on CI: the "Wait for the squid.log to be written" run intermittently fails
because the log line does not appear within a 10-second readiness window. It
passes on re-run with no code change.
```
Run: Wait for the squid.log to be written: Failed
Starting TestRun 2-tr : Checking that Process is ready within 10.0 seconds
so we can start process: Default - Failed
Reason: Process failed to become ready in time: 10.0 seconds. Waited
10.000008344650269 seconds
Process: watcher: Passed
Time-Out : Process finishes within expected time - Passed
Reason: Returned value: 9.50020146369934 < 30.0
Process: Default: Skipped
Reason: Was not started
```
Seen on
https://ci.trafficserver.apache.org/job/Github_Builds/job/autest/40871/ (AuTest
0of4), where the other 65 test files in the shard passed.
The test intends a 30-second budget for the wait:
https://github.com/apache/trafficserver/blob/master/tests/gold_tests/h2/h2origin.test.py#L87-L98
`watcher.TimeOut` and `tr.TimeOut` are both set to 30, but the readiness
gate that `StartBefore` applies to the watcher is a separate 10-second default
that the test does not raise. The watcher was still waiting at 9.5s when the
readiness check expired at 10.0s, so the 30-second budget is never reached.
Squid log entries appear only when the log buffer is flushed, so on a loaded CI
worker the entry for the last transaction can take longer than 10 seconds to
show up.
Passing `ready=` with a longer timeout to `StartBefore`, or lowering the log
flush interval for this test, would make the wait honor the intended budget.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
