Closed Bug 1781782 Opened 2 years ago Closed 2 years ago

run xpcshell tests 'sequentially' when they fail frequently in parallel

Categories

(Testing :: XPCShell Harness, task)

Default
task

Tracking

(firefox105 fixed)

RESOLVED FIXED
105 Branch
Tracking Status
firefox105 --- fixed

People

(Reporter: jmaher, Assigned: jmaher)

References

Details

Attachments

(1 file)

xpcshell runs all tests in parallel, then any failures that happen will be retries sequentially. This results in 95% of the failures going away- I find that many of these tests are actually failing 25-100% of the time and often timing out.

The end result of this bug is to edit xpcshell.ini manifest files and add run-sequentially key to the tests which are most prone to retrying (yet not failing all the time)

to find these just run xpcshell tests without retry=True and it will report failures in parallel. Patterns emerge quickly.

These align with a bigQuery find showing which testnames failed once but not twice in xpcshell tasks.

One risk here is that some new tests will become high intermittent, effectively becoming disabled- I view this as we will be finding the real errors, not masking them.

Assignee: nobody → jmaher
Status: NEW → ASSIGNED
Pushed by jmaher@mozilla.com:
https://hg.mozilla.org/integration/autoland/rev/9fe9e9146b01
Run frequent xpcshell parallel failing tests sequential by default. r=ahal,necko-reviewers,extension-reviewers,application-update-reviewers,bytesized,rpl,kershaw
Status: ASSIGNED → RESOLVED
Closed: 2 years ago
Resolution: --- → FIXED
Target Milestone: --- → 105 Branch

After this bug landed, failed tests are running twice even if --verbose is set. That's not intended, is it?

Flags: needinfo?(jmaher)

this was intended. Is this causing a problem for you, or hiding problems?

Flags: needinfo?(jmaher)

It just seems pointless. I now get all the same information I got before, but twice. Which makes it more confusing to scroll back through the output to find what happened.

I spent some time looking at xpcshell jobs and I am not seeing this duplicated error. Maybe I am looking in the wrong place, can you give an example?

Regressions: 1790085
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: