1724992 - Permanent macOS T-swr(s) Talos svgr software render PROCESS-CRASH | tscrollx | application crashed [unknown top frame] after Timeout waiting for test completion; killing browser... when Gecko 93 merges to Beta on 2021-09-06

Status: RESOLVED FIXED

(Core :: Graphics: WebRender, defect)

Description

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Tentatively from bug 1723665, will verify with a backout on Try but that takes 3h.

• central-as-beta simulation
• How to run these simulations
• Failure log

[task 2021-08-10T13:03:16.841Z] 13:03:16     INFO -  PID 1075 | Cycle 1(21): loaded http://127.0.0.1:49221/tests/scroll/reader.htm (next: http://127.0.0.1:49221/tests/scroll/tiled.html)
[task 2021-08-10T13:03:21.313Z] 13:03:21     INFO -  PID 1075 | Cycle 1(22): loaded http://127.0.0.1:49221/tests/scroll/reader.htm (next: http://127.0.0.1:49221/tests/scroll/tiled.html)
[task 2021-08-10T13:03:21.838Z] 13:03:21     INFO -  Timeout waiting for test completion; killing browser...
[task 2021-08-10T13:03:21.839Z] 13:03:21     INFO -  Killing psutil.Process(pid=1075, name='firefox', status='running', started='12:53:21') and writing a minidump file

Comment 1

[Glenn Watson [:gw]]

I've asked for this to be backed out for now, as it caused another regression too. I'll investigate today.

Comment 2

[Sebastian Hengst [:aryx] (needinfo me if it's about an intermittent or backout)]

Try push confirmed it's from bug 1723665. Thank you for investigating.

Comment 3

[Glenn Watson [:gw]]

I found what's causing the failure, but still need to investigate further the underlying cause.

The issue is that the talos test is timing out (600 seconds) under this configuration and being killed by the test runner.

There's a few possible causes:

• The patch severely regresses performance on this test. In theory it should have zero effect on performance, but it's possible something weird is going on. I'm running all the talos tests on other platforms with sw-wr / hw-wr and this patch to see if they have regressed at all.
• Might be hitting something related to the issues in this test (https://bugzilla.mozilla.org/show_bug.cgi?id=1645275)
• Might have been right on the edge of timing out before and this change subtly changes timing in a small way to go over the timeout.

Comment 4

[Jan Honza Odvarko [:Honza] (always need-info? me)]

The regression was reopened (patch backed out), so closing this regression.