1709270 - 10.68 - 3.69% web-de dcf / google-search-restaurants LastVisualChange + 16 more (Android) regression on Mon April 19 2021

Status: NEW

(Testing :: Performance, defect, P5)

Description

[Florin Strugariu [:Bebe]]

Perfherder has detected a browsertime performance regression from push 8a4eff9d5f949a52dd7097395f4a124ceb9e39d4. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

Ratio	Suite	Test	Platform	Options	Absolute values (old vs new)
11%	web-de	dcf	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,373.42 -> 1,520.12
6%	youtube	fcp	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,003.06 -> 1,060.92
6%	google	ContentfulSpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	951.17 -> 1,005.42
6%	google	SpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	955.00 -> 1,009.42
6%	google	PerceptualSpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	952.08 -> 1,005.75
6%	google	FirstVisualChange	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	947.75 -> 1,001.08
5%	booking	FirstVisualChange	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,287.00 -> 1,354.75
5%	youtube	dcf	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,129.90 -> 1,185.62
5%	youtube	dcf	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,134.88 -> 1,190.00
5%	google-search-restaurants	FirstVisualChange	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,202.46 -> 1,259.50
...	...	...	...	...	...
4%	bing-search-restaurants	ContentfulSpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,298.12 -> 1,351.08
4%	google-search-restaurants	FirstVisualChange	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,206.92 -> 1,255.67
4%	booking	ContentfulSpeedIndex	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,445.25 -> 1,502.58
4%	google-search-restaurants	LastVisualChange	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,491.96 -> 1,550.08
4%	google-search-restaurants	LastVisualChange	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,490.25 -> 1,545.17

Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs and let us know your plans within 3 business days, or the offending patch(es) will be backed out in accordance with our regression policy.

For more information on performance sheriffing please see our FAQ.

Comment 1

[Emilio Cobos Álvarez (:emilio)]

Presumably you wanted to ni? glenn and set the regressing bug properly? Otherwise this won't ever get fixed.

Comment 3

[Glenn Watson [:gw]]

I'm not sure exactly how we assign regressions to bugs, but are we sure the correlation here is correct?

The reason I ask is that the patch referenced above was backed out for causing a visual drawing regression (see https://hg.mozilla.org/mozilla-central/rev/09bea3dd34bd0da18d56758701528ca7be22c0ea).

I then landed an updated / fixed patch which is https://hg.mozilla.org/mozilla-central/rev/6527552f2b28, and a subsequent fix for a regression in https://hg.mozilla.org/mozilla-central/rev/c56cf2509c45 just a couple of days ago.

Given the nature of the change, it seems unlikely to cause a visual or performance regression, but it is of course plausible there is something unexpected happening.

Are you able to check if the regression disappeared when the reported change was backed out, and also if it was re-introduced / fixed when either of the above patches landed?

Comment 4

[Florin Strugariu [:Bebe]]

I'm 100% that the flagged commit is the one causing the regression.

You are right that the commit was backed out was backedout and committed again but because of the noise and the small time between the backout and commit we might not see it in the graphs

I did some more re-triggers so I can point out the backout and re-commit see:
https://treeherder.mozilla.org/perfherder/graphs?highlightAlerts=1&highlightChangelogData=1&series=autoland,3386242,1,13&timerange=2592000&zoom=1618671446704,1619287386219,1421.4175730409188,1605.7065165092743

Comment 5

[Florin Strugariu [:Bebe]]

I can confirm that this patch is causing the regression:

Alert of initial commit:
https://treeherder.mozilla.org/perfherder/alerts?id=29850

Alert for backout:
https://treeherder.mozilla.org/perfherder/alerts?id=29979
== Change summary for alert #29979 (as of Thu, 06 May 2021 07:40:40 GMT) ==

Improvements:

Ratio	Suite	Test	Platform	Options	Absolute values (old vs new)
3%	google-search-restaurants	LastVisualChange	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,535.25 -> 1,495.58

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=29979

Alert for new commit after backout:
== Change summary for alert #29980 (as of Thu, 06 May 2021 07:42:47 GMT) ==

Regressions:

Ratio	Suite	Test	Platform	Options	Absolute values (old vs new)
2%	google-search-restaurants	LastVisualChange	android-hw-g5-7-0-arm7-api-16-shippable-qr	cold webrender	1,522.96 -> 1,550.75

For up to date results, see: https://treeherder.mozilla.org/perfherder/alerts?id=29980

Comment 6

[Glenn Watson [:gw]]

OK, thanks for confirming!

From what I understand, LastVisualChange refers to the latest time anything visual changed on this test, is that right? Do you know if there is any way to see what the reference screenshots are from CI before / after this patch, or do I need to get this hardware and build myself locally (sorry if I missed where there is a link to documentation on these tests, I'm not very familiar with them).

Would any change in these tests related to https://hg.mozilla.org/mozilla-central/rev/c56cf2509c45 (which landed 2 days ago) have shown up in perf alerts yet? (That fixes a rendering bug related to that patch - specifically, it fixes rounded corners drawing on iframes, which might result in a visual difference, but I'd need screenshots to confirm if that might be related).