1079705 - ~10M AWSY regression on Sept 30

Status: RESOLVED FIXED

(Core :: Networking, defect)

Description

[Arpad Borsos [:Swatinem]]

AWSY shows a ~10M regression in the upper "explicit" measurements and in the "heap-unclassified" category.
The "resident" measurement is also up some, but they are very noisy.

Comment 1

[Andrew McCreight [:mccr8]]

[Tracking Requested - why for this release]: This is a large regression in memory usage.

Comment 2

[Nicholas Nethercote [inactive]]

Drilling down into the heap-unclassified line in the "Miscellaneous Measurements" graph, it's pretty clear that these are the relevant changes:

https://hg.mozilla.org/integration/mozilla-inbound/pushloghtml?fromchange=946948bfc6a3&tochange=6122db43cad3

So it is bug 1002855. mccr8 backed out a related patch, involving resource timings, in bug 1064706, that was showing up as leaking on e10s builds. He also said that he didn't understand why it was only manifesting on e10s runs; perhaps this is how it was manifesting.

Comment 3

[Valentin Gosu [:valentin] (he/him)]

I landed bug 1064706, and now we no longer hold refs to the channel in nsPerformanceTiming.
It will be interesting to see how it affects this regression.

Comment 4

[Nicholas Nethercote [inactive]]

> I landed bug 1064706, and now we no longer hold refs to the channel in
> nsPerformanceTiming.
> It will be interesting to see how it affects this regression.

I see a clear decrease in the "explicit" and "heap-unclassified" measurements. The size of the decrease is sufficiently close to the original increase that I'm happy to declare victory :)

Thank you for the fix.

Comment 5

[Andrew McCreight [:mccr8]]

I added annotations to AWSY for this.