1900982 - Use arrays instead of objects in the tracer resources

Status: RESOLVED FIXED

(DevTools :: Debugger, enhancement)

Description

[Alexandre Poirot [:ochameau]]

The tracer currently uses objects to communicate stacks and frames.
This is a source of performance bottleneck as these JSON objects are ultimately stringified and represents larger packets than their equivalents in arrays.
This is also a significant difference between the tracer and the profiler. The profiler is using arrays in the uploaded profile.
Using arrays should help better integrate with the profiler and share more code between the devtools live tracer and the recorded tracer.

Comment 1

[Alexandre Poirot [:ochameau]]

Profile of mozilla-central, using objects and logging to the console:
https://share.firefox.dev/3XiwHNg

Profile with the patch, using arrays and also logging to the console:
https://share.firefox.dev/3Re4sf3

Against the following test:
data:text/html,<script>window.onclick=()=>{for(let i = 0; i < 10000; i++) a()}; function a(){}</script>

From the content process, flushTraces/emitTraces is significantly faster. From 25ms to 9ms.
This is small number on this test page, but flushTraces can be a much slower step on large websites.

Comment 2

[Alexandre Poirot [:ochameau]]

Attachment: Bug 1900982 - [devtools] Migrate JS tracer from Objects to Array as trace payloads.

Using arrays speeds up the serialization and transmission of traces between the server and the client
as it goes through JS Actors/IPC by serializing the JSON RDP packets.

This also allows to better align with the profiler data format.

Comment 3

[Alexandre Poirot [:ochameau]]

Attachment: Bug 1900982 - [devtools] Cover the backend performance of the JS tracer.

Comment 4

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/39e79beba8f9
[devtools] Cover the backend performance of the JS tracer. r=perftest-reviewers,devtools-reviewers,kshampur,nchevobbe
https://hg.mozilla.org/integration/autoland/rev/2cd0e475a654
[devtools] Migrate JS tracer from Objects to Array as trace payloads. r=devtools-reviewers,perftest-reviewers,kshampur,nchevobbe

Comment 5

[Sandor Molnar[:smolnar]]

https://hg.mozilla.org/mozilla-central/rev/39e79beba8f9
https://hg.mozilla.org/mozilla-central/rev/2cd0e475a654

Comment 6

[Nicolas Chevobbe [:nchevobbe]]

Perfherder has detected a devtools performance change from push 2cd0e475a6543ca9fbf0f71d934d5b81fbcd6f1b.

Regressions:

Ratio	Test	Platform	Options	Absolute values (old vs new)
19%	damp custom.jsdebugger.close.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender-sw	18.11 -> 21.51
18%	damp custom.jsdebugger.close.DAMP	macosx1015-64-shippable-qr	e10s fission stylo webrender-sw	14.03 -> 16.62
17%	damp custom.jsdebugger.close.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender	18.38 -> 21.51
15%	damp custom.jsdebugger.close.DAMP	macosx1015-64-shippable-qr	e10s fission stylo webrender	14.27 -> 16.48
11%	damp custom.netmonitor.close.DAMP	macosx1015-64-shippable-qr	e10s fission stylo webrender-sw	15.07 -> 16.80
11%	damp custom.netmonitor.close.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender-sw	18.29 -> 20.22
10%	damp custom.netmonitor.close.DAMP	macosx1015-64-shippable-qr	e10s fission stylo webrender	15.42 -> 17.00
9%	damp custom.netmonitor.close.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender	18.28 -> 19.96
7%	damp custom.jsdebugger.stepIn.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender-sw	2,680.95 -> 2,876.15
7%	damp custom.netmonitor.reload.DAMP	macosx1015-64-shippable-qr	e10s fission stylo webrender	238.02 -> 254.06
...	...	...	...	...
5%	damp custom.netmonitor.requestsFinished.DAMP	macosx1015-64-shippable-qr	e10s fission stylo webrender-sw	1,590.17 -> 1,675.53
5%	damp panelsInBackground.reload.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender-sw	413.86 -> 435.77
5%	damp custom.netmonitor.requestsFinished.DAMP	macosx1015-64-shippable-qr	e10s fission stylo webrender	1,601.87 -> 1,674.67
5%	damp browser-toolbox.debugger-ready.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender	638.73 -> 667.68
2%	damp custom.netmonitor.open.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender-sw	168.12 -> 171.75

Improvements:

Ratio	Test	Platform	Options	Absolute values (old vs new)
10%	damp simple.netmonitor.close.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender	6.45 -> 5.80
5%	damp custom.jsdebugger.pretty-print.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender	1,688.78 -> 1,609.10
2%	damp custom.jsdebugger.stepInNewSource.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender-sw	1,862.78 -> 1,824.10
2%	damp simple.netmonitor.open.DAMP	windows10-64-shippable-qr	e10s fission stylo webrender	154.43 -> 151.30

As author of one of the patches included in that push, we need your help to address this regression.
Details of the alert can be found in the alert summary, including links to graphs and comparisons for each of the affected tests. Please follow our guide to handling regression bugs and let us know your plans within 3 business days, or the patch(es) may be backed out in accordance with our regression policy.

If you need the profiling jobs you can trigger them yourself from treeherder job view or ask a sheriff to do that for you.

You can run these tests on try with ./mach try perf --alert 737

For more information on performance sheriffing please see our FAQ.