Closed Bug 2018450 Opened 7 months ago Closed 6 months ago

Spike in failures on linux/win ccov

Categories

(Firefox Build System :: Task Configuration, defect)

defect

Tracking

(firefox-esr140 unaffected, firefox148 unaffected, firefox149 wontfix, firefox150 wontfix, firefox151 fixed)

RESOLVED FIXED
151 Branch
Tracking Status
firefox-esr140 --- unaffected
firefox148 --- unaffected
firefox149 --- wontfix
firefox150 --- wontfix
firefox151 --- fixed

People

(Reporter: amarc, Assigned: jcristau)

References

(Regression)

Details

(Keywords: intermittent-failure, intermittent-testcase, regression)

hi Aryx, do you know what's going on here?

Flags: needinfo?(aryx.bugmail)
Duplicate of this bug: 2018449
Duplicate of this bug: 2018424
Duplicate of this bug: 2018421
Duplicate of this bug: 2018419
Duplicate of this bug: 2018418
Duplicate of this bug: 2018425

hi Andrew, could these be caused by some kind of worker issue?

Flags: needinfo?(aerickson)
Keywords: regression
Regressed by: 2016051
Duplicate of this bug: 2018417

The linux failures are on g-w cloud linux d2g hosts running the in-tree docker image gw-fxci-gcp-l1-2404-amd64-headless-googlecompute-2026-01-14.

Relops doesn't maintain the in-tree docker images. CC'ing :jcristau, who has been working on them a bit.

Flags: needinfo?(aerickson) → needinfo?(jcristau)

Are there specific failing tests (for each platform)? The linux worker image hasn't changed since mid-January.

Flags: needinfo?(jcristau)

The count of browser-chrome failures on Linux CCov per push jumped and the tests listed in the duplicate bugs fail either frequently or permanently, e.g. High frequency ccov browser/base/content/test/tabcrashed/browser_aboutRestartRequired_buildid_false-positive.js | single tracking bug (bug 2018417).

Bug 2018418 failed - during the setup function of the test - before only (?) on Try pushes by :gsvelto (example) but no change by Gabriele is part of the changelog.

Gabriele, do you know what started these failures?

Flags: needinfo?(aryx.bugmail) → needinfo?(gsvelto)
No longer regressed by: 2016051

I suspect it must have been because of a change in how these tasks are selected when running on try. I've got some pre-saved queries that I invoke with ./mach try fuzzy --preset <name> and they didn't include ccov tests, but didn't exclude them either. Recently the ccov tests are always running so I've been tweaking my queries to disable them explicitly unless I need them.

Flags: needinfo?(gsvelto)
Duplicate of this bug: 2020891
No longer depends on: 2021821
Duplicate of this bug: 2021821

I used the git revs to actually get a complete diff for all the changes of this merge to mozilla-central:
https://github.com/mozilla-firefox/firefox/compare/4a30b4e343d1..bfa2c2f2cb7c

My suspicion is that bug 2016051 actually caused this problem given that it touches quite a lot for CCOV builds and we only see these failures for CCOV builds.

Marco, can you please take a look?

Flags: needinfo?(mcastelluccio)

Bug 2016051 is adding a task at the end of all others, so it can't affect what previous tasks in the graph are doing.
The only regression it might have caused is how frequently we run the ccov tasks (see for example bug 2020718), but it can't affect the rate of failures.

Flags: needinfo?(mcastelluccio)
See Also: → 2020718

(In reply to Marco Castelluccio [:marco] from comment #33)

The only regression it might have caused is how frequently we run the ccov tasks (see for example bug 2020718), but it can't affect the rate of failures.

What does frequent here mean? We did not run those jobs at all before, so what actually calculates the frequency?

Note that especially the W-async-cf task (duped bug 2021821) is set to only run on try but not on any trunk branches. So as I read this is related to the frequency when the test is getting run, which jumped from never to always. And it's not related to the number of failures in such tasks - which are mostly expected due to underlying platform issues that need to get fixed.

OK, I thought the bug was about a higher rate of failures of a task that was already being scheduled in the past. The solution to bug 2020718 would probably fix this too then.

Depends on: 2020718
See Also: 2020718 →

Set release status flags based on info from the regressing bug 2016051

Marco, do you think it might make sense to backout the regressor temporarily until the underlying taskgraph fix has landed and a new release it out? Or isn't it possible due to some other automation / tools depending on this jobs now?

Flags: needinfo?(mcastelluccio)

Unfortunately we can't because we already switched other automation to rely on that.

Flags: needinfo?(mcastelluccio)

(In reply to Marco Castelluccio [:marco] from comment #40)

Unfortunately we can't because we already switched other automation to rely on that.

I thought so. Thanks for confirming. Let’s at least needinfo Andrew so he’s aware of this bug, as his taskgraph update work would likely help here significantly. Hopefully that can move forward quickly.

Flags: needinfo?(ahal)

Thanks, are you talking about https://github.com/taskcluster/taskgraph/pull/917 ? I'll try and get it ready for review

Flags: needinfo?(ahal)

(In reply to Andrew Halberstadt [:ahal] from comment #42)

Thanks, are you talking about https://github.com/taskcluster/taskgraph/pull/917 ? I'll try and get it ready for review

Thank you! Yes, that's at least what Marco was pointing at, which should as well help for this bug (see comment 33).

See Also: → 2016937

Set release status flags based on info from the regressing bug 2016051

looks like it's fixed

Amazing. Thanks a lot jcristau!

And yes, the number of jobs for CCOV builds drastically reduced:

https://treeherder.mozilla.org/jobs?repo=mozilla-central&searchStr=ccov&fromchange=81bd8e73cd80793f8d49b769880b6a954b8aa962&tochange=a5514d76e0e7d3f3b0a6afdee8b8bf8d29d2d129

I assume that with this change we still run all those jobs that we actually should run.

Marco, maybe you have the overview and could take a look?

Assignee: nobody → jcristau
Status: NEW → RESOLVED
Closed: 6 months ago
Flags: needinfo?(mcastelluccio)
Resolution: --- → FIXED
Target Milestone: --- → 151 Branch

We are looking good!

Flags: needinfo?(mcastelluccio)
You need to log in before you can comment on or make changes to this bug.