when tests fail and there isn't a test path in the failure message, output the directory/manifest that is run
Categories
(Testing :: General, task, P3)
Tracking
(Not tracked)
People
(Reporter: jmaher, Unassigned)
References
(Blocks 1 open bug)
Details
often we have failures that are leaks, crashes, assertions, timeouts, infrastructure. Many times this a result of an issue where the browser failed something during the set of tests and we detect a failure on shutdown.
In order for ML test scheduling to properly annotate the failure, we should ensure these failures have a manifest or some related identifier in the error output.
Reporter | ||
Comment 1•5 years ago
|
||
taking this as an example:
https://treeherder.mozilla.org/#/jobs?repo=mozilla-central&selectedTaskRun=AAl6hTQhSIqbQCYKQNZEKw-0&revision=385f49adaf00d02fc8e04da1e0031e3477182d67&searchStr=mda2
this is what treeherder shows as a failure:
TEST-UNEXPECTED-FAIL | Last test finished | application terminated with exit code 11
Exceeded max 20 bug suggestions, most of which are likely false positives.
PROCESS-CRASH | Last test finished | application crashed [@ mozilla::(anonymous namespace)::RunWatchdog(void*)]
Exceeded max 20 bug suggestions, most of which are likely false positives.
Return code: 1
Got 1 unexpected crashes
# TBPL FAILURE #
The mochitest suite: mochitest-media ran with return status: FAILURE
and what I see in the errorsummary.log file:
{"action": "log", "message": "TEST-UNEXPECTED-FAIL | Last test finished | application terminated with exit code 11", "line": 2186, "level": "ERROR"}
I would start with a proposal that we should make:
ERROR - TEST-UNEXPECTED-FAIL | Last test finished | application terminated with exit code 11
look like:
ERROR - TEST-UNEXPECTED-FAIL | dom/media/test/ | Last test finished: application terminated with exit code 11
there are 2 things different here:
- directory name is in the second part of the line
- the original part
Last test finished
is now in the error message with a : separating it from the original error message.
For this specific failure we would need to edit:
https://searchfox.org/mozilla-central/source/testing/mochitest/runtests.py#2359
The caveat to this is we have a crash:
application crashed [@ mozilla::(anonymous namespace)::RunWatchdog(void*)]
that could get overlooked across other suites and manifests because we are pinning it to a specific directory (in this case dom/media/test/
)
Updated•5 years ago
|
Comment 2•5 years ago
|
||
How frequent are these kinds of failures (if you exclude intermittents)? Should it be a higher priority for us?
Reporter | ||
Comment 3•4 years ago
|
||
here is another case:
https://treeherder.mozilla.org/logviewer.html#/jobs?job_id=305636826&repo=autoland&lineNumber=2177
and a bit from the structured log:
..."source": "mochitest", "reason": "MOZ_RELEASE_ASSERT(domWindow)", "test": "remoteautomation.py", "minidump_path": "/tmp/tmpzcNdn3/1a536fd3-594a-9c78-d41c-bf405e0d4c1f.dmp", "time": 1591712678616, "action": "crash"}
ekyle would like to see something in the structured log to indicate:
"group":"dom/media/test/mochitest.ini" / "test":None
:ekyle want to add more, or does this cover our conversation?
Comment 4•4 years ago
|
||
My only request is to not fill the "test"
field if the failure can not be blamed on a test. Adding a "group"
field to the line is optional.
Adding "action":"group_start"
and "action":"group_end"
records would be a good way to mark tests as belonging to a group, AND provide a way to describe the failure of a group, but not a test.
Description
•