All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=801128

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=a8c7b21d27e29b719592b5ba8d9533cedd861eab7698b37c64b94ad34ad38b87


Bot(s) for this bug's original alert(s):

linux-release

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14d3d18f040000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/14d3d18f040000

cc: Avoid main thread latency recovery during loading.
By khushalsagar@chromium.org · Tue Jan 09 19:25:03 2018
chromium @ 2f9cdf26e68f1fc58a978216b0e4d97eb660a2d3

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

Overall the change above has improved the loading times a lot (https://chromeperf.appspot.com/group_report?rev=528065), since it ensures we send main frames sooner than we were earlier. But that may not always be good since a generally lower priority task which was important for first paint is now delayed. That is not something we can identify on the cc thread.

I looked more deeply on the major one, Kakaku. And it looks like we are skipping an impl frame after the first main frame now. The main thread responds within the deadline and we are able to submit a CF, but we don't receive a frame ack before the next impl frame. I think as a result it is entering into impl latency recovery mode (https://cs.chromium.org/chromium/src/cc/scheduler/scheduler.cc?sq=package:chromium&l=942). The change above clears the history from a previous navigation which can change the estimates used and affect this decision. But I can't think of any other heuristic we could be applying here to avoid this. I thought about restricting impl latency recovery on min number of samples as well, but we should have mainframe/activation data from 1 frame when this happens.

Closing this as WontFix. Brian, feel free to reopen if you get an idea on something better we could do here.

 Issue 801129  has been merged into this issue.