You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 429bedc
Browse filesBrowse the repository at this point in the historyBrowse files
feat: score rendered-test interactions by main-thread work to commit (#901)
* feat(stats-generator): score interaction trace by main-thread work to commit
* feat: record interaction work to commit alongside the Lighthouse breakdown
* Reverted ordering
---------
Co-authored-by: Alexander Karan <47707063+AlexanderKaran@users.noreply.github.com>
Copy file name to clipboardExpand all lines: packages/docs/src/benchmarks/lighthouse-client-side-rendered-tests/Description.mdx
+1-1Lines changed: 1 addition & 1 deletion
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,4 +1,4 @@
1
-
These CSR browser performance tests use Lighthouse to measure paint timings and a controlled navigation interaction.
1
+
These CSR browser performance tests use Lighthouse to measure paint timings and a controlled navigation interaction. The interaction is also scored as [work to commit](/methodology/#interaction-work-to-commit) from the same trace.
2
2
3
3
Methodology: [Client Side Rendered Tests (Lighthouse)](/methodology/#client-side-rendered-tests-lighthouse).
Copy file name to clipboardExpand all lines: packages/docs/src/benchmarks/lighthouse-client-side-rendered-tests/LighthouseFrameworkClientRenderedVersionCharts.astro
These SSR browser performance tests use Lighthouse to measure paint timings and a controlled navigation interaction.
1
+
These SSR browser performance tests use Lighthouse to measure paint timings and a controlled navigation interaction. The interaction is also scored as [work to commit](/methodology/#interaction-work-to-commit) from the same trace.
2
2
3
3
Methodology: [Server Side Rendered Tests (Lighthouse)](/methodology/#server-side-rendered-tests-lighthouse).
Copy file name to clipboardExpand all lines: packages/docs/src/benchmarks/lighthouse-server-side-rendered-tests/LighthouseFrameworkServerRenderedVersionCharts.astro
Copy file name to clipboardExpand all lines: packages/docs/src/content/docs/methodology.md
+54-7Lines changed: 54 additions & 7 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -44,7 +44,8 @@ Test use larger, 16-CPU runners, as described in their sections below.
44
44
disables its screen emulation, with no simulated CPU or network throttling.
45
45
Requests use a local connection to a production server. The Lighthouse flow
46
46
runs only the audits required for the published First Paint, First Contentful
47
-
Paint, and interaction-latency measurements.
47
+
Paint, and interaction-latency measurements, and records the interaction's
48
+
trace for work to commit.
48
49
-**Recorded environment:** Results include runner details and the measured
49
50
framework version; browser results also include the Chrome version used. Any benchmark that runs multiple itterations record mean, median, minimum, and maximum.
50
51
@@ -385,14 +386,18 @@ Current pinned dependencies are listed in each project's `package.json`:
385
386
- Interaction latency sums the input delay, processing duration, and
386
387
presentation delay from Lighthouse's INP breakdown. It measures this
387
388
controlled interaction, not page-lifetime INP.
389
+
- The same interaction is also scored as
390
+
[work to commit](/methodology/#interaction-work-to-commit), from the trace
391
+
Lighthouse records for it. Each comparison ranks frameworks by its own
392
+
metric, lowest first.
388
393
- CI builds once per framework and test, then keeps one production server running
389
394
for all five measurements. Each measurement launches a fresh Chrome process.
390
395
- Before all measurements, one HTTP request warms `/client-side-rendered`.
391
396
The response is fully consumed with a ten-second timeout. This warms the
392
397
server, not the measured browsers, and is excluded from results.
393
-
- Valid slow results are retained. Paint values and total interaction latency
394
-
must be finite and greater than zero; individual interaction components may
395
-
be zero.
398
+
- Valid slow results are retained. Paint values, total interaction latency, and
399
+
work to commit must be finite and greater than zero; individual interaction
400
+
components may be zero.
396
401
- Each framework uses its supported production routing and rendering controls.
397
402
- Next.js disables SSR for the measured table and detail content.
398
403
- Astro uses React client-only islands. Its detail link performs a full-document
@@ -414,15 +419,19 @@ Current pinned dependencies are listed in each project's `package.json`:
414
419
- Interaction latency sums the input delay, processing duration, and
415
420
presentation delay from Lighthouse's INP breakdown. It measures this
416
421
controlled interaction, not page-lifetime INP.
422
+
- The same interaction is also scored as
423
+
[work to commit](/methodology/#interaction-work-to-commit), from the trace
424
+
Lighthouse records for it. Each comparison ranks frameworks by its own
425
+
metric, lowest first.
417
426
- CI builds once per framework and test, then keeps one production server running
418
427
for all five measurements. Each measurement launches a fresh Chrome process.
419
428
- Before all measurements, HTTP requests warm `/server-side-rendered` and a
420
429
detail URL with a random UUID. Both responses are fully consumed under one
421
430
ten-second timeout. This warms the server, not the measured browsers, and is
422
431
excluded from results.
423
-
- Valid slow results are retained. Paint values and total interaction latency
424
-
must be finite and greater than zero; individual interaction components may
425
-
be zero.
432
+
- Valid slow results are retained. Paint values, total interaction latency, and
433
+
work to commit must be finite and greater than zero; individual interaction
434
+
components may be zero.
426
435
- Links use each framework's production navigation component or ordinary
427
436
anchors, allowing default prefetching. The table is loaded and rendered on
428
437
demand without caching or prerendering its output. Static surrounding shells
@@ -432,6 +441,44 @@ Current pinned dependencies are listed in each project's `package.json`:
432
441
- Astro's table and detail routes use `export const prerender = false` for
433
442
request-time rendering.
434
443
444
+
### Interaction Work to Commit
445
+
446
+
The client and server side rendered tests also score the benchmark's click from
447
+
the Chrome trace that Lighthouse records for the interaction step.
448
+
449
+
- Work to commit is the sum of three parts:
450
+
-**Input delay:** from the input's timestamp to the start of the first event
451
+
callback for the interaction.
452
+
-**Processing duration:** from the start of the first event callback to the
453
+
end of the last, across the interaction's pointer and click events.
454
+
-**Rendering work:** time the renderer main thread is busy, from the end of
455
+
processing until the interaction's frame commits. This includes framework
456
+
work scheduled by the event, style, layout, and paint. Overlapping tasks are
457
+
counted once.
458
+
- Input delay and processing duration match Lighthouse's INP breakdown for
459
+
client-routed navigations. Rendering work takes the place of presentation
460
+
delay.
461
+
- Idle main-thread time while waiting for the next frame is excluded. That wait
462
+
depends on where the click lands within the frame interval rather than on the
463
+
framework.
464
+
- Work after commit is not measured. This includes raster, draw, presentation,
465
+
and browser UI frames, such as the frame software-rendered headless Chrome
0 commit comments