Open Bug 1878425 Opened 1 year ago Updated 1 year ago

test-macosx1100-64-shippable-qr/opt-reftest-* tasks occasionally somehow run in a mode where pixel-rounding seems to be different, which results in tons of reftest failures

Categories

(Core :: Layout, task)

task

Tracking

()

People

(Reporter: dholbert, Unassigned)

References

Details

Attachments

(3 files)

Some of the time, it appears our test-macosx1100-64-shippable-qr/opt-reftest-* tasks end up with a huge number of unexpected reftest failures (and some unexpected reftest passes), where the common theme among the failures/passes seems to be pixel-scaling and/or pixel rounding.

The massive number of failures also ends up causing the logs to exceed the treeherder log-parser size limit, so they don't show up on treeherder in a way that makes it easy to visualize the failures.

This happened on 3 distinct reftest runs today on mozilla-release:
https://firefoxci.taskcluster-artifacts.net/X9u7fC_5QfWNC5KvDSd51g/0/public/logs/live_backing.log
https://firefoxci.taskcluster-artifacts.net/cOhrtf5QSbCAq6GYurVwuw/0/public/logs/live_backing.log
https://firefoxci.taskcluster-artifacts.net/Li7SCFDKTby_a0i2yNfVrQ/0/public/logs/live_backing.log

I'm not seeing any other recent instances, but let's see if it continues to happen vs. if it was a one-off...

(Note, sheriffs filed bug 1878387 and bug 1878389 for the three failure-logs linked in comment 0 here.)

Blocks: 1878387, 1878389

I attached the gzipped logs that I linked in comment 0, for archival in case they disappear from taskcluster.

Depends on: 1878707
Duplicate of this bug: 1878707
No longer depends on: 1878707
No longer blocks: 1878389
Duplicate of this bug: 1878389
No longer blocks: 1878387
Duplicate of this bug: 1878387
Duplicate of this bug: 1881355
Duplicate of this bug: 1883062

FWIW, we accumulated a bunch of mistaken annotations in the past ~day that for logs like this one with a single reftest unexpected-pass on macosx1015, for a specific test that started passing due to a bug fix that landed recently (hooray). That expected-fail annotation was removed in Bug 1863421 comment 11 to reflect reality, so the unexpected passes should stop now.

Those unexpected passes were unrelated to the actual bug here -- so, that spike in reports here shouldn't be taken as any indication that this bug is suddenly getting more frequent than it previously was.

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

Attachment

General

Created:
Updated:
Size: