1641740 - Investigate profiling tests timing issues

Status: RESOLVED DUPLICATE of bug 1595687

(Core :: Gecko Profiler, defect, P3)

Description

[Gerald Squelart (he/him) (not at Mozilla since 2022-09-15)]

There are some big differences in the time it takes to profile a couple of tests, it is unclear whether the amount of captured data is different, or there is bad interaction between the two consecutive tests.

Spawned from bug 1586105 comment 45:

I (Julien) looked at the measurements again from the latest error log linked above:

For the multi_frame test:

[task 2020-03-29T06:26:45.198Z] 06:26:45     INFO - Capture the profile data.
[task 2020-03-29T06:28:00.724Z] 06:28:00     INFO - Check if the captured pages are the ones with correct values we created.

=> 1m15 to capture

For the single_frame test:

[task 2020-03-29T06:28:45.526Z] 06:28:45     INFO - Capture the profile data.
[task 2020-03-29T06:31:37.333Z] 06:31:37     INFO - Check if the captured page is the one with correct values we created.

=> 2m52 to capture

From a recent log (https://treeherder.mozilla.org/logviewer.html#?job_id=302715554&repo=mozilla-central):
for multi_frame:

[task 2020-05-18T10:23:18.139Z] 10:23:18     INFO - Capture the profile data.
[task 2020-05-18T10:23:54.725Z] 10:23:54     INFO - Check if the captured pages are the ones with correct values we created.

=> 36s

for single frame:

[task 2020-05-18T10:24:17.125Z] 10:24:17     INFO - Capture the profile data.
[task 2020-05-18T10:25:20.236Z] 10:25:20     INFO - Check if the captured page is the one with correct values we created.

=> 1m03

So there's still a big difference, and I think we should look at it more.
Maybe we should 1/ look at the data, see if we have more data in the second case, and 2/ understand the difference with both cases.

Comment 1

[Gerald Squelart (he/him) (not at Mozilla since 2022-09-15)]

Oh, I had already opened bug 1595687 to investigate this issue! 😅