Adding Markus to review the format of generated profiles.

In D270623#9394498, @florian wrote:

I opened https://github.com/firefox-devtools/profiler/issues/5655 to track ideas.

In D270623#9388796, @nchevobbe wrote:

I don't know how to test this, but it seems to be the straightforward fix.

Thanks for debugging this!

Looks good to me, thanks!

There are 2 reasons for the bucket size:

• the performance issue when showing too many nodes at once
• the time it takes to scroll to the end of what has been expended. If the bucket size was much larger, scrolling would be inconvenient, and you would probably need to collapse the bucket again to see something else, which makes comparing data from multiple arrays much harder (having multiple arrays of indexes is very common when inspecting profile JSON files).

In D270316#9371576, @tarek wrote:

new try https://treeherder.mozilla.org/jobs?repo=try&revision=e0034ff303304182eea29994863a202da0e26e74

Thanks for the review!

I tried to make "expand all" expand buckets too, and for large-ish JSON files that restored the existing behavior of using a bit more than a CPU core until an out of memory error occurred.

Thanks for working on this, I'm looking forward to seeing the impact on job run times!

Try run: https://treeherder.mozilla.org/jobs?repo=try&selectedTaskRun=NLWMhcLaRc2sfhQn0KzUew.0&revision=7e42453431caf18134433853d38c27138ff8403b

Looks reasonable overall. Several trivial comments.

Thanks for the review!

In D270516#9360514, @ochameau wrote:

Thanks for spending time on fixing these important limitations of the JSON Viewer!

For some reason "expand all" doesn't work?

There's a try push at https://treeherder.mozilla.org/jobs?repo=try&selectedTaskRun=X5sD9iN4QD-WhkkIv3lC_Q.0&revision=f4b76ce8910af59c5142f741222410cd2d42bda6
Here's the resource usage profile of the new job: https://share.firefox.dev/4hyQW1y
Jobs actually running daily should finish much faster, as they try to fetch artifacts from the previous run before re-generating the data for a given day.

I landed a similar change in D269625.

Try looks acceptably green: https://treeherder.mozilla.org/jobs?repo=try&revision=4b1c4f35d3a4befb1b70878d6a6218e44b43ef4b

This isn't strictly related to bug 1995627, but it made sense to cleanup while I was in this area, test markers were the last ones using time.monotonic and ignoring the "time" field of structured logging data. It's better to have a consistent source of information for all the timestamps we show in the same profile.

Example profile: https://share.firefox.dev/4ovvwo9

For long running tests that use a lot of CPU and typically finished last, for a portion of their run time they were running alone. Now that they run in parallel to other tests for their entire duration, they take longer.

Profile with the patch: https://share.firefox.dev/4ovvwo9 (test_backgroundtask_experiments.js starts at the begining of the job)
Profile without the patch: https://share.firefox.dev/4niH90G (test_backgroundtask_experiments.js started after 50s spent on other tests)

In D269574#9326540, @florian wrote:

http://queze.net/goinfre/tmp/tests/#date=2025-10-21&q=geocoding&view=list (click the row with the test name to expend it) shows a chart of the run times on autoland/mozilla-central yesterday. 0 timeouts.

http://queze.net/goinfre/tmp/tests/#date=2025-10-21&q=geocoding&view=list (click the row with the test name to expend it) shows a chart of the run times on autoland/mozilla-central yesterday. 0 timeouts.

Do you have a try push with https://bugzilla.mozilla.org/show_bug.cgi?id=1995666#c3 addressed? That's the first step before annotating individual tests.

Thanks for cleaning this up!