Page MenuHomePhabricator

Increased Largest Contentful Paint for enwiki emulated mobile with Chrome 100
Closed, ResolvedPublic

Assigned To
Authored By
Mar 31 2022, 7:45 AM
Referenced Files
F35043325: Screenshot 2022-04-08 at 08.42.42.png
Apr 8 2022, 6:47 AM
F35043318: Screenshot 2022-04-07 at 16.23.11.png
Apr 8 2022, 6:47 AM
F35043330: Screenshot 2022-04-08 at 08.44.58.png
Apr 8 2022, 6:47 AM
F35042536: Screenshot 2022-04-07 at 16.19.57.png
Apr 7 2022, 2:20 PM
F35031483: Screenshot 2022-03-31 at 09.36.43.png
Mar 31 2022, 7:45 AM
F35031487: chrome99vs100.jpg
Mar 31 2022, 7:45 AM


Looking at the graphs for WebPageTest emulated mobile we can see that we have an increase for that metric:

Screenshot 2022-03-31 at 09.36.43.png (754×2 px, 250 KB)

If I just look at one URL I can see that it correlates to when Chrome 100 was released:

chrome99vs100.jpg (1×2 px, 246 KB)

Event Timeline

Looking into the WebPageReplay tests and the other test we run directly against Wikipedia and there there's no visible change. Let me look into if I get something useful out of WebPageTest.

I could also find a jump in responseEnd but only for emulated mobile:

Screenshot 2022-04-07 at 16.19.57.png (1×2 px, 434 KB)

As reported in T305572 I could see that Chrome makes multiple requests to optimizationguide-pa, In Browsertime/ those requests are blocked but since we haven't done any update to WebPageTest in a year or more, WebPageTest missed the Chrome flag to turn it off. I can also see that people reported that the requests happen even with the flag enabled so I just added that domain to blocking.

Screenshot 2022-04-07 at 16.23.11.png (1×2 px, 257 KB)

That fixed the late request in T305572 and more interesting it also fixes the response end:

Screenshot 2022-04-08 at 08.42.42.png (1×2 px, 245 KB)

It isn't super easy to see (since we had WPT go down for some time but also LCP is affected for Sweden and Facebook (hard to say about Obama):

Screenshot 2022-04-08 at 08.44.58.png (684×1 px, 259 KB)

Its interesting that this only happens on emulated mobile, so in best case this is not something that would affect real users. I'm gonna create a Chrome task and ask.

There's an upstream bug (that I do not have access to) and then there's some info in this thread:

If I understand correctly there was bug that was introduced (that hidden for us) and then they are working on trying to distribute the calls to the optimisation guide better.