1677569 - CYCLE_COLLECTOR_MAX_PAUSE regression on May 11, 2020

Status: RESOLVED INCOMPLETE

(Core :: Cycle Collector, defect)

Description

[Markus Stange [:mstange]]

We seem to have had a regression in CYCLE_COLLECTOR_MAX_PAUSE during the 78 cycle.

Nightly: https://mzl.la/3lEiOn3
Beta: https://mzl.la/3lEiOn3

:chutten divined the following push log range from the build IDs:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=f44e64a61ed1&tochange=61a83cc0b74b

Olli, does anything stand out from that log?

Comment 1

[Andrew McCreight [:mccr8]]

This is probably a dupe of bug 1658571, which was a regression from bug 1618292, which landed on 5-11. Though it looks like the values didn't quite go down to their previous values when that landed. They did come back down later?

Comment 2

[Markus Stange [:mstange]]

That matches perfectly.

Yes, it does look like the fix didn't bring the numbers back down to the original level, though it did make a noticeable impact - there's a clear decrease on September 24th.

Comment 3

[Markus Stange [:mstange]]

It's possible that the remaining piece is bug 1667316.

Comment 4

[Andrew McCreight [:mccr8]]

This also shows up in the GHOST_WINDOWS telemetry, as you'd expect.

Comment 5

[Andrew McCreight [:mccr8]]

In the 95th percentile, that is. Like this max pause, it hasn't gone down entirely to what it was before.

Comment 6

[Andrew McCreight [:mccr8]]

It looks like a lot of this was fixed, and we're probably not going to make much progress on it this much later.