On a private instance, we are occasionally seeing a Start Render time for Chrome browser results that is lower than the reported TTFB time. WPT Server & Clients are both 2.11.
This should not be possible. Is there a bug here/anyone else seeing this issue?
Thank you,
James
Bump…anyone seeing this behavior, or have thoughts on it?
At least partially answering my own question.
After more focused debugging and testing across multiple clients, it appears that there are occasional race conditions with the code being tested that are exclusive to the Chrome browser. The test metrics and logs of FF, IE, Safari do not display the same results as Chrome.
So, if anyone does see this specific behavior, or Start Render times that are overly optimistic, take a look at the associated test run Console Log first…some very useful information in there.
Are you seeing it with Video capture enabled as well or with timeline disabled? I’m wondering if it is taking one of the paint messages in the timeline and tripping up over that. It shouldn’t be the case and if you have a case that you can reproduce on the public instance I’d love to take a look at it.
Thanks, Patrick. I just reviewed the last few days of our private instance results, and am not seeing the issue of Start Render time < TTFB for any tests.
Our test clients have not been changed (still on 2.11 pending some downtime to update all of them at once). The codebase in test still shows occasional race conditions.
I do note that the Chrome browser version was updated several days ago, and again today, however…