| Peter | |
| Jan 24 2022, 10:19 AM |
| F36897892: Screenshot 2023-03-08 at 13.09.27.png | |
| Mar 8 2023, 12:13 PM |
| F36897891: Screenshot 2023-03-08 at 13.09.10.png | |
| Mar 8 2023, 12:13 PM |
| F35147968: Screenshot 2022-05-17 at 12.10.00.png | |
| May 17 2022, 10:11 AM |
| F34946569: firstvisualchange-and-firstcontentfulpaint-webpagetest.png | |
| Feb 9 2022, 12:04 PM |
| F34946573: chrome-96->97-traffic.png | |
| Feb 9 2022, 12:04 PM |
| F34946575: FCP-chrome-users-fastest.png | |
| Feb 9 2022, 12:04 PM |
| F34946574: FCP-chrome-users-median.png | |
| Feb 9 2022, 12:04 PM |
| F34946570: difference-between-firstvisual-change-and-fcp-webpagetest.png | |
| Feb 9 2022, 12:04 PM |
There's been so many alerts since T298945 so I missed the alerts firing for upgrading to 97 but only when we run emulated mobile:
Could easily verify that TTFB are stable:
Looking at the tests without WebPageReplay using emulated Chrome there's the same pattern:
See https://bugs.chromium.org/p/chromium/issues/detail?id=1291502
Looking at FCP vs FVC:
Looking at the filmstrip it seems FVC is reported correct and FCP is wrong:
Let me check if something else rolled out at the same time.
And also on desktop. I haven't been able to see anything on our RUM metric. I could also see that the discrepancy between first paint/fcp and first visual change also happens with WebPageTest:
Let me create an Chrome issue and see if I can get some help understanding what's going on.
Change 759485 had a related patch set uploaded (by Phedenskog; author: Phedenskog):
[performance/synthetic-monitoring-tests@master] Disable Chrome PaintHoldingCrossOrigin feature.
Change 759485 merged by jenkins-bot:
[performance/synthetic-monitoring-tests@master] Disable Chrome PaintHoldingCrossOrigin feature.
I pushed a change for desktop and webpagereplay tests, lets see what the metrics looks like tomorrow. I also added it to tests on a Mac mini, the difference there is smaller but maybe it makes a difference.
Change 759572 had a related patch set uploaded (by Phedenskog; author: Phedenskog):
[performance/synthetic-monitoring-tests@master] Also disable PaintHoldingCrossOrigin on enwiki tests.
Change 759572 merged by jenkins-bot:
[performance/synthetic-monitoring-tests@master] Also disable PaintHoldingCrossOrigin on enwiki tests.
Change 761046 had a related patch set uploaded (by Phedenskog; author: Phedenskog):
[performance/synthetic-monitoring-tests@master] Try to disable on PaintHoldingCrossOrigin on WebPageTest.
Change 761046 merged by jenkins-bot:
[performance/synthetic-monitoring-tests@master] Try to disable on PaintHoldingCrossOrigin on WebPageTest.
I updated the Chrome upstream bug but want to add the screenshots here too. It's the same pattern on WebPageTest:
And I also think I see the same pattern for real users. When Chrome 97 rolled out, our fastest and median FCP for Chrome became faster. The only thing is that in synthetic we can see that the metric become faster but the page become slower.
Change 792447 had a related patch set uploaded (by Phedenskog; author: Phedenskog):
[performance/synthetic-monitoring-tests@master] Enable PaintHoldingCrossOrigin for investigating LCP changes.
Change 792447 merged by jenkins-bot:
[performance/synthetic-monitoring-tests@master] Enable PaintHoldingCrossOrigin for investigating LCP changes.
Here's what happens with Largest Contentful Paint. Lets run it one more iteration and then revert the change.
Change 792640 had a related patch set uploaded (by Phedenskog; author: Phedenskog):
[performance/synthetic-monitoring-tests@master] Disable paint holding again.
Change 792640 merged by jenkins-bot:
[performance/synthetic-monitoring-tests@master] Disable paint holding again.
Change 805126 had a related patch set uploaded (by Phedenskog; author: Phedenskog):
[performance/synthetic-monitoring-tests@master] Enable paint holding again.
Change 805126 merged by jenkins-bot:
[performance/synthetic-monitoring-tests@master] Enable paint holding again.
I've been starting to setup the bare metal server in T311983 and this bug is problematic (it always have but its even more now): If we run with default settings we still have the discrepancy between RUM metrics and visual metrics. If I turn off the change (disable-features=PaintHoldingCrossOrigin) it looks like vector-2022 introduced a new way of rendering, wherre sometimes Chrome chooses to render the search bar first. Compare these two:
I've pinged the Chrome issue again and asked when they are gonna release a fix.
