1740011 - 61.16 - 4.75% netflix loadtime / netflix SpeedIndex + 2 more (Windows) regression on Mon October 25 2021

Status: RESOLVED WONTFIX

(Core :: Performance, defect)

Description

[Acasandrei Beatrice (needinfo me)]

Perfherder has detected a browsertime performance regression from push 5c29a315aa4aa1a19d5e2d65e5ba57df0c8501a2. As author of one of the patches included in that push, we need your help to address this regression.

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
61%	netflix loadtime	windows10-64-shippable-qr	cold webrender	1,247.33 -> 2,010.17
8%	netflix LastVisualChange	windows10-64-shippable-qr	cold webrender	1,865.50 -> 2,017.17
6%	netflix ContentfulSpeedIndex	windows10-64-shippable-qr	cold webrender	1,119.25 -> 1,186.83
5%	netflix SpeedIndex	windows10-64-shippable-qr	cold webrender	963.67 -> 1,009.42

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

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1736057

Comment 2

[Pascal Chevrel:pascalc (PTO until April 26)]

Denis, this looks like a bad perf regression on a top site, could you look at it? Thanks

Comment 3

[Denis Palmeiro [:denispal]]

I tried reproducing this locally, live and recorded, and could not reproduce. Looking at the profiles, it seems this is another case where we take longer to compile the script asynchronously (as expected because of the full parsing change) while the main thread is waiting to execute since there is no network delay present here that will allow the compile to occur in parallel with fetch requests.

Considering that we do win on several sites such as reddit, where there is a lot of JS execution due to react and our telemetry shows nice improvements for beta at the moment, I think this is a CI regression we may need to absorb which is not really representative of the live site. Up until today, beta telemetry is showing us a 5% improvement in FCP and page load time, and nearly a 20% improvement in JS execution time during page load.