Adds the following tasks to mozilla-central:
mochitest-chrome with nova enabled
browser-chrome with nova enabled
browser-chrome with nova and vertical tabs enabled
various performance tests with nova enabled
Differential Revision: https://phabricator.services.mozilla.com/D303348
The WebCodecs camera variants (ve-*-cam) acquire frames from a fake camera
that is only available in Firefox, via the media.navigator.streams.fake family
of preferences. On Chrome and custom-car (Chromium) those preferences have no
effect, so the camera tests find no video input device and permafail. Remove
the camera variants from the chrome and custom-car subtest lists; they continue
to run on Firefox. The canvas (non-camera) variants are unaffected.
Differential Revision: https://phabricator.services.mozilla.com/D304619
Adds the following tasks to mozilla-central:
mochitest-chrome with nova enabled
browser-chrome with nova enabled
browser-chrome with nova and vertical tabs enabled
various performance tests with nova enabled
Differential Revision: https://phabricator.services.mozilla.com/D303348
This patch enables mitmproxy 12 on the Linux platform. Following up on
the work of Bug 1983694. Previously linux was hardcoded to mitmproxy 8
due to compatibility issues on ubuntu 18
Additionally linux 18 platform restriction for browsertime is removed
(browsertime raptor chrome/car/wasm). Also tp6-bench is disabled as the
stale recordings mixed in there were failing.
Additionally Bug 2031677 is also fulfilled
Differential Revision: https://phabricator.services.mozilla.com/D299175
A local-server browsertime test to measure navigation time when the
Speculation Rules API prefetches a link, so we can track the performance
of the upcoming prefetch feature.
Differential Revision: https://phabricator.services.mozilla.com/D296731
Route SP3+A55 tests to run on both Lambdatest and BitBar temporarily, to
get some longitudinal data comparison.
Additionally, when running SP3 on LT, disable accessibility services,
scrcpy, and WebRTC packages in an attempt to reduce variance.
Differential Revision: https://phabricator.services.mozilla.com/D291794
Make it easier to write a new performance test by adding documention on how to select a test framework as well as starter code.
Differential Revision: https://phabricator.services.mozilla.com/D289304
This patch temporarily disables and changes scheduling of some perftest
tasks on Windows NUC13 hardware while we are replacing some hardware
components. Here we will prioritize important tests and adjust
accordingly over the next few weeks.
Differential Revision: https://phabricator.services.mozilla.com/D299474
xperf-talos has been running on win11 25h2 in parallel with 24h2 since
Bug 2032529 landed on 2026-04-17. Drop the 24h2 xperf-talos tasks now
that 25h2 results are stable.
Differential Revision: https://phabricator.services.mozilla.com/D298580
This patch adds the `./mach perfdocs` command to mozperftest. It also sets up the logger required by perfdocs for producing reviewbot-compatible lint errors, and output these into a JSON file. A `--generate` option is added to enable re-generation of the documentation. At the same time, the existing lint task is changed to use this new command.
Because of a slight change in how perfdocs works, this patch also fixes a bug with the talos gatherer that had caused an existing failure to slip through.
Differential Revision: https://phabricator.services.mozilla.com/D297130
This reverts commit 867f4d29c8.
Revert "Bug 2035608 - Always run speedometer3 at the same port (I picked 62763), to ease before/after comparisons. r=perftest-reviewers,sparky"
This reverts commit 34f9ef67a6.
Add raptor test definitions for all codec/latency-mode combinations
using camera as the frame source with Firefox's fake camera device
(media.getusermedia.camera.fake.force). Linux skips H264 camera tests
since x264 is not available; Chrome and custom-car also omit full-res
H264, matching existing canvas test patterns.
Differential Revision: https://phabricator.services.mozilla.com/D291336
Pre-generate all VideoFrames before encoding tests begin so that frame
creation overhead does not affect benchmark measurements. Add I420 as a
selectable input pixel format alongside RGBX. Add raptor test entries
for I420 variants across all existing codec and latency-mode combinations.
Differential Revision: https://phabricator.services.mozilla.com/D290726
Previously tp6 essential tests were enabled on apple silicon machines
and are running on mozilla-central. This patch updates it to include
autoland now. This will also ensure it appears on AWFY in the same
consistent way as other platforms.
Differential Revision: https://phabricator.services.mozilla.com/D295191
Add windows11-*-25h2 entries to worker.py so the taskgraph can resolve
worker pools for 25h2 test platforms. Split windows11-24h2-tests into
two test-sets: windows11-24h2-tests retains suites still on 24h2 cloud
pools (reftest, crashtest, jsreftest, firefox-ui-functional) while
windows11-25h2-tests contains suites already migrated to 25h2 pools.
Create new 25h2 platform entries in test-platforms.yml for the migrated
suites and slim down 24h2 entries to the remaining suites plus perf.
Differential Revision: https://phabricator.services.mozilla.com/D287593
Part of the ubuntu 18.04 -> 24.04 migration. (Bug 1983694) Add a couple
restriction transforms as some tests still need to remain on 18.04 for
the time being, but a majority of tests can be removed.
Additionally, a minor fix is added to the browsertime network-bench
tests as the *-qr suffix in the config was preventing 24.04 tasks from
being picked up.
Differential Revision: https://phabricator.services.mozilla.com/D289425
Some tests were either perma failing or highly intermittent when
replayed with mitmproxy 12. This patch updates `amazon`, `google-docs`,
and `cnn` with new recordings
Additionally, `wikia` and `google-search` are pinned to mitmproxy 11 for
the time being.
Differential Revision: https://phabricator.services.mozilla.com/D288319
Previously we did not run pageload tests on Apple silicon as we had
limited capacity and visual metrics did not work. We have much more
capacity now and the visual metrics problem doesn't occur anymore.
For starters we can just enable tp6-essential and slowly add more.
This patch also builds on the mitmproxy 12 upgrade (and more recent
versions of mitmproxy have an arm64 mitmdump binary so we don't need to
use rosetta)
Differential Revision: https://phabricator.services.mozilla.com/D282075
This patch enables performance pageload tests to now use mitmproxy
12.2.1, only on the Mac/Windows/Android platform.
This will run on a subset of tests as the binary doesn't properly run on
all of the old recordings.
Differential Revision: https://phabricator.services.mozilla.com/D282044
Split the coarse-grained mLock into two fine-grained locks to reduce contention:
- mDBLock (RWLock): Protects the DNS cache (mRecordDB) with read-heavy optimizations
- mQueueLock (Mutex): Protects work queues and thread coordination
This reduces lock hold times by decoupling cache lookups from queue operations,
allowing concurrent cache reads while queue work proceeds independently.
Stress test confirms 89% contention reduction at real scale (166.27x → 17.62x ratio).
Differential Revision: https://phabricator.services.mozilla.com/D288329
Split the coarse-grained mLock into two fine-grained locks to reduce contention:
- mDBLock (RWLock): Protects the DNS cache (mRecordDB) with read-heavy optimizations
- mQueueLock (Mutex): Protects work queues and thread coordination
This reduces lock hold times by decoupling cache lookups from queue operations,
allowing concurrent cache reads while queue work proceeds independently.
Stress test confirms 89% contention reduction at real scale (166.27x → 17.62x ratio).
Differential Revision: https://phabricator.services.mozilla.com/D288329
Split the coarse-grained mLock into two fine-grained locks to reduce contention:
- mDBLock (RWLock): Protects the DNS cache (mRecordDB) with read-heavy optimizations
- mQueueLock (Mutex): Protects work queues and thread coordination
This reduces lock hold times by decoupling cache lookups from queue operations,
allowing concurrent cache reads while queue work proceeds independently.
Stress test confirms 89% contention reduction at real scale (166.27x → 17.62x ratio).
Differential Revision: https://phabricator.services.mozilla.com/D288329
Speedometer 2 currently runs as "speedometer" in CI. Couple of issues
- can cause false regression alerts
- naming confusion with speedometer3
this patch renames the test from speedometer to speedometer2 and
disables it on all production branches while keeping it available for
Try pushes as the spidermonkey team still finds it useful for JS
sensitive changes
Differential Revision: https://phabricator.services.mozilla.com/D286273
Dave Townsend