1450732 - GC hangs in content process reported by BHR increased by ~60% on March. 9 2018

Status: RESOLVED DUPLICATE of bug 1447871

(Core :: JavaScript: GC, defect, P1)

Description

[Alex Thayer [:alexical] (she/her)]

See https://arewesmoothyet.com/?mode=track&trackedStat=All%20Hangs - scroll down to Content, and uncheck all of the non-GC categories in the legend to observe the change.

Comment 1

[Andrew Overholt [:overholt]]

sfink and I were looking at awsy graphs and this seems perhaps to correlate? https://treeherder.mozilla.org/perf.html#/graphs?timerange=5184000&series=mozilla-inbound,1643522,1,4&series=mozilla-inbound,1643532,0,4&series=mozilla-inbound,1643552,0,4&series=mozilla-inbound,1643542,0,4&series=mozilla-inbound,1643512,0,4&zoom=1520432239256.5598,1520809861688.0466,271535569.09454,363670400.55521417

Comment 2

[Alex Thayer [:alexical] (she/her)]

Yeah I think so - in bug 1449662 we're discussing what appears to be an off-by-one error somewhere in the pipeline of BHR. If we adjust for that, then March. 8 sounds right.

Comment 3

[Jan de Mooij [:jandem]]

Bug 1438211 landed around that time and introduced a memory leak that was fixed a few days ago by bug 1447871.

Comment 4

[Steve Fink [:sfink] [:s:]]

The perfherder link seems to show that it was only present March 8-10, then returned to normal. The BHR report seems to show it as persisting. Going off of the perfherder results, the one thing that seems relevant is that in https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=5521f4a9012d88d6aa399bc16b27ec0013976526&tochange=2cb86c6e003ddad85e7600bd84e03252b5269cd3 are changes to switch to different hashtable implementations in Gecko code. (But considering that the BHR problems persist, I'm not suggesting blaming that.)

Seemingly not relevant, but it's hard to ignore the massively overshadowing CC regression starting on 2018-01-31 and persisting. Is that one known?

Comment 5

[Steve Fink [:sfink] [:s:]]

(In reply to Jan de Mooij [:jandem] from comment #3)
> Bug 1438211 landed around that time and introduced a memory leak that was
> fixed a few days ago by bug 1447871.

Oh! Thanks, that seems very likely.

Comment 6

[Alex Thayer [:alexical] (she/her)]

(In reply to Steve Fink [:sfink] [:s:] from comment #4)
> Seemingly not relevant, but it's hard to ignore the massively overshadowing
> CC regression starting on 2018-01-31 and persisting. Is that one known?

Yeah - the CC regression is being tracked in bug 1450514.

(In reply to Jan de Mooij [:jandem] from comment #3)
> Bug 1438211 landed around that time and introduced a memory leak that was
> fixed a few days ago by bug 1447871.

That looks like the culprit. I'll close this out as a dupe for now and double check the graph in a few days.