1375550 - test-linux64-qr/debug-reftest-e10s-5 runs for too long due to warning spew in webrender, with stylo build enabled

Status: RESOLVED DUPLICATE of bug 1375843

(Testing :: General, defect)

Description

[Geoff Brown [:gbrown]]

Starting around

https://treeherder.mozilla.org/#/jobs?repo=autoland&filter-searchStr=quantumrender%20reftest%20debug&tochange=471a163b37d092fc5bf7a56bcf5c5295f727b8d8&fromchange=2bcf04b9b40ba3ef45d03f22529b5503d920ddd2

test-linux64-qr/debug-reftest-e10s-5 jobs starting taking too long and frequently timing out.

I notice all linux64-qr debug reftests at the top of that range have frequent warnings: 

WARN:webrender::clip_scroll_node: Tried to scroll a non-scroll node.

...maybe related?


I'll try to find the regressing changeset.

Comment 1

[Geoff Brown [:gbrown]]

It seems like bug 1374824 caused this: https://treeherder.mozilla.org/#/jobs?repo=autoland&revision=471a163b37d092fc5bf7a56bcf5c5295f727b8d8 linux64-qr debug reftest logs are flooded with the "Tried to scroll" warning, and e10s R5 runs at least 15 minutes slower than in previous pushes, resulting in frequent timeouts.

Comment 2

[Gregory Szorc [:gps]]

If you search for "stylo" in logs (e.g. https://public-artifacts.taskcluster.net/O-9tLUtXQNCrfT3mFmJZ5w/0/public/logs/live_backing.log), it does appear Stylo is disabled. So, uh, I have no clue why this would be unless there is some interaction between WebRender and Stylo causing it. Redirecting needinfo.

Comment 3

[Manish Goregaokar [:manishearth]]

Don't really have an idea what could have caused this :/

Comment 4

[Kartikaya Gupta (email:[e-mail])]

I confirmed Joel's conclusion independently via retriggers in this range:

https://treeherder.mozilla.org/#/jobs?repo=autoland&filter-searchStr=qr%20reftest%20e10s&tochange=28585cf7da6fdc07ac775ea47ad3aa8fae406351&fromchange=1990807be52407bdba9d61d1883300185c8b9952&group_state=expanded

The R5 failures appeared on bug 1374824 landing, disappeared when the build patch was backed out, and re-appeared when it relanded. Quite mysterious!

Comment 5

[Kartikaya Gupta (email:[e-mail])]

err sorry I meant Geoff not Joel!

Comment 6

[Kartikaya Gupta (email:[e-mail])]

The patch in bug 1375843 should silence the warning. We should still investigate why building stylo triggered this change, as it might point to a deeper problem. Let's use this bug for that, I guess.

Comment 7

[Kartikaya Gupta (email:[e-mail])]

I also realized that in bug 1375843 I reproduced this problem *without* stylo being built, because it's not enabled in local builds by default. So most likely it's just a pre-existing timing thing where webrender sometimes triggers this warning and sometimes it doesn't, and in automation on R5 building stylo changed enough timing that it started triggering pretty consistently. There are other known (unexpected) side-effects from building stylo that :glandium is still chasing down in bug 1374824.

Comment 8

[Geoff Brown [:gbrown]]

Job times seem to be back to normal after kats' work in bug 1375843.