I've seen the method taking lots of time for example with Facebook.

https://share.firefox.dev/3l1fJwV

I wonder if we end up having too many entries in mResourceEntries or in mUserEntries
and then sorting becomes very slow
https://searchfox.org/mozilla-central/rev/c938c7416c633639a5c8ce4412be586eefb48005/dom/performance/Performance.cpp#177,206

But, at least there is a clear bug in
https://searchfox.org/mozilla-central/rev/c938c7416c633639a5c8ce4412be586eefb48005/dom/performance/PerformancePaintTiming.cpp#38 and
https://searchfox.org/mozilla-central/rev/c938c7416c633639a5c8ce4412be586eefb48005/dom/performance/PerformanceResourceTiming.cpp#43
We end up calculating timestamp tons of times when sorting. We surely should
use some cached timestamp.

My guess is that bug 1518999 made this bad on Facebook.

(I wonder if this needs to tracked for FF84.)

This seems to be about resource entries.

Aha, mResourceTimingBufferSize=100000
I guess the web site has increased that number to so high.
But still, sorting shouldn't take much time.

Indeed, calculating the startTime every time slows down the sort.

The attached test file took about 120ms ~ 150ms for the GetEntriesByName call on my machine. With a patch that caches the startTime, it only takes ~7ms.

I'll get a patch up soon.

PerformanceResourceTiming recalculates the startTime every time when
StartTime() method is called, which slows down GetEntriesByName()
due to this method requires StartTime() to order the entries.

In addition to that, there's not need to recalculate it every time, since all
required timings are already set when the PerformanceTimingData
object is first created.

This patch calculates the startTime in the constructor and caches it.

Currently we store the raw StartTime as a TimeStamp object, and
convert it to DOMHighResTimeStamp when calling StartTime(). However,
there's no need to do the conversion every time.

Depends on D96295

FWIW, since I saw this again today, I tried to reproduce on Chrome, and yes, I got FB consuming basically 100% of a core.

That doesn't mean that Gecko's implementation shouldn't be optimized, comparison done during the sorting is implemented badly
(and perhaps some different data structure should be used to avoid need for any sorting?).

Backed out for marionette failures on test_refresh_firefox.py

backout: https://hg.mozilla.org/integration/autoland/rev/3fd9592e326d96fe5927843931ffd7829004b238

push: https://treeherder.mozilla.org/jobs?repo=autoland&revision=d7af96ca92b4533e81cc0ef24c27bd98676009f1&selectedTaskRun=DuUBA5JeTHeo2p61ZkfF7g.0

failure log: https://treeherder.mozilla.org/logviewer?job_id=321213513&repo=autoland&lineNumber=57305

[task 2020-11-09T17:47:44.349Z] 17:47:44 INFO - DEBUG: Completed blocker OS.File: flush pending requests, warn about unclosed files, shut down service. for phase xpcom-will-shutdown
[task 2020-11-09T17:47:44.350Z] 17:47:44 INFO - DEBUG: Finished phase xpcom-will-shutdown
[task 2020-11-09T17:47:44.986Z] 17:47:44 INFO - _RegisterApplication(), FAILED TO establish the default connection to the WindowServer, _CGSDefaultConnection() is NULL.
[task 2020-11-09T17:47:54.301Z] 17:47:54 INFO - Child process with id "2177" has been marked as detached because it is no longer in the managed process group. Keeping reference to the process id "2181" which is the new child process.
[task 2020-11-09T17:47:54.539Z] 17:47:54 INFO - TEST-UNEXPECTED-ERROR | browser/components/migration/tests/marionette/test_refresh_firefox.py TestFirefoxRefresh.testFxANoSync | KeyError: 'keyFetchToken'
[task 2020-11-09T17:47:54.539Z] 17:47:54 INFO - Traceback (most recent call last):
[task 2020-11-09T17:47:54.539Z] 17:47:54 INFO - File "/Users/cltbld/tasks/task_1604938831/build/venv/lib/python2.7/site-packages/marionette_harness/marionette_test/testcases.py", line 214, in run
[task 2020-11-09T17:47:54.539Z] 17:47:54 INFO - testMethod()
[task 2020-11-09T17:47:54.539Z] 17:47:54 INFO - File "/Users/cltbld/tasks/task_1604938831/build/tests/marionette/tests/browser/components/migration/tests/marionette/test_refresh_firefox.py", line 697, in testFxANoSync
[task 2020-11-09T17:47:54.539Z] 17:47:54 INFO - self.checkFxA()
[task 2020-11-09T17:47:54.540Z] 17:47:54 INFO - File "/Users/cltbld/tasks/task_1604938831/build/tests/marionette/tests/browser/components/migration/tests/marionette/test_refresh_firefox.py", line 488, in checkFxA
[task 2020-11-09T17:47:54.540Z] 17:47:54 INFO - self.assertEqual(result["accountData"]["keyFetchToken"], "top-secret")
[task 2020-11-09T17:47:54.540Z] 17:47:54 INFO - TEST-INFO took 10381ms

https://hg.mozilla.org/mozilla-central/rev/9ccec3c16e42
https://hg.mozilla.org/mozilla-central/rev/5bb6a35c2193

Olli, let me know if you hit this again.