syncThrobberAnimations takes a lot of time at startup when there are many tabs
Categories
(Firefox :: Tabbed Browser, defect, P3)
Tracking
()
| Tracking | Status | |
|---|---|---|
| firefox123 | --- | fixed |
People
(Reporter: sfoster, Assigned: sfoster)
References
Details
Attachments
(2 files)
I'm attaching a saved session with 1000 tabs. To use:
- Check the "Open previous windows and tabs" checkbox in about:preferences and quit the browser
- Copy in the attached sessionstore-1000-tabs.jsonlz4 file into the profile directory and rename it sessionstore.jsonlz4 (don't use a valuable profile, or back up your sessionstore-backups folder if you do.)
- Start the browser. It will probably hang for a few seconds while it reads the session file and creates all those tabs
I captured this profile starting up with this. It shows several seconds spent in syncThrobberAnimations. Which is odd because only the selected tab (and any pinned tabs) should be actively loading. The function tries to get the animations from each and every tab and then loops over them to sync all their startTimes. I think we filter for only the tabs that newTab.hasAttribute("busy").
|
Assignee | |
Comment 1•2 years ago
|
||
|
||
Updated•2 years ago
|
|
||
Comment 2•2 years ago
|
||
I'm curious, sfoster - did you notice any difference in the time to restore from that large session with this patch applied?
|
|
||
Comment 3•2 years ago
|
||
Nevermind, I see that this was answered in https://phabricator.services.mozilla.com/D196537#6520090.
|
||
Comment 4•2 years ago
|
||
The severity field is not set for this bug.
:dao, could you have a look please?
For more information, please visit BugBot documentation.
|
||
Updated•2 years ago
|
|
||
Comment 6•2 years ago
|
||
Backed out for causing multiple failures @ syncThrobberAnimations
Backout link: https://hg.mozilla.org/integration/autoland/rev/eb9a193de2b14421848c6db358fce5ed2e52b25a
|
|
Assignee | |
Comment 7•2 years ago
|
||
Ah, thanks I'll take a look and update the patch (and see if mach try auto would have caught this before I pushed.)
|
|
Assignee | |
Comment 9•2 years ago
•
|
||
To get some measurements of the before and after I used the STR from comment #0, and ran and then quit the browser to capture a startup profile:
`MOZ_PROFILER_STARTUP=1 MOZ_PROFILER_STARTUP_INTERVAL=10 MOZ_PROFILER_SHUTDOWN=/path/to/output/profile.json MOZ_PROFILER_STARTUP_ENTRIES=20000000 ./mach run`
And then loaded up the output file at https://profiler.firefox.com/, and filtered down the markers in the marker table view of the parent process with sessionrestore-.
We create a perf marker at sessionstore-windows-restored which is when the session data has been read and the windows created, but before any tabs are created. The sessionstore-one-or-no-tab-restored marker indicates the moment during startup when all the tabs have been restored. That is, the tab elements are created (most remain inactive and don't actually load anything until selected.) The time between the two moments is where this patch has an impact:
| session / patch | sessionstore-windows-restored (s) | sessionstore-one-or-no-tab-restored (s) | time to restore tabs (s) |
|---|---|---|---|
| 1000 tabs (m-c | 4.569 | 11.503 | 6.934 |
| 1000 tabs (m-c | 4.723 | 12.864 | 8.141 |
| 1000 tabs (m-c | 4.641 | 11.434 | 6.793 |
| 1000 tabs (m-c | 4.508 | 14.403 | 9.895 |
| 1000 tabs (m-c | 4.528 | 11.427 | 6.899 |
| 1000 tabs (patched) | 5.531 | 8.949 | 3.418 |
| 1000 tabs (patched) | 4.517 | 7.985 | 3.468 |
| 1000 tabs (patched) | 4.516 | 7.902 | 3.386 |
| 1000 tabs (patched) | 4.537 | 7.763 | 3.226 |
| 1000 tabs (patched) | 4.601 | 7.996 | 3.395 |
| 500 tabs (m-c) | 3.995 | 7.61 | 3.615 |
| 500 tabs (m-c) | 3.329 | 6.63 | 3.301 |
| 500 tabs (m-c) | 3.58 | 7.26 | 3.68 |
| 500 tabs (patched) | 4.038 | 6.726 | 2.688 |
| 500 tabs (patched) | 3.726 | 5.875 | 2.149 |
| 500 tabs (patched) | 3.343 | 5.93 | 2.587 |
| 100 tabs (m-c) | 2.689 | 6.836 | 4.147 |
| 100 tabs (m-c) | 2.476 | 4.677 | 2.201 |
| 100 tabs (m-c) | 2.551 | 4.781 | 2.23 |
| 100 tabs (m-c) | 2.663 | 4.958 | 2.295 |
| 100 tabs (patched) | 2.426 | 4.574 | 2.148 |
| 100 tabs (patched) | 2.386 | 4.265 | 1.879 |
| 100 tabs (patched) | 2.378 | 4.514 | 2.136 |
Here's that distilled to just the time we spent restoring the tabs with and without the patch:
| session / patch | median time to restore tabs (s) |
|---|---|
| 100 tabs (m-c) | 2.2625 |
| 100 tabs (patched) | 2.136 |
| 500 tabs (m-c) | 3.615 |
| 500 tabs (patched) | 2.587 |
| 1000 tabs (m-c) | 6.934 |
| 1000 tabs (patched) | 3.395 |
|
|
||
Comment 10•2 years ago
|
||
| bugherder | ||
Description
•