Skip to content

test: fix what the run of #3282 showed, and use the 64-core runners - #3284

Merged
Pavel Feldman (pavelfeldman) merged 1 commit into
microsoft:mainfrom
pavelfeldman:large-runners
Oct 21, 2026
Merged

Pavel Feldman (pavelfeldman) merged 1 commit into
microsoft:mainfrom
pavelfeldman:large-runners

Conversation

@pavelfeldman

Copy link
Copy Markdown
Member

Summary

  • The golden comparison is pixelmatch at threshold 0.2 again, as the matcher of @playwright/test compares; test: fix the ports and the helpers that the full matrix showed #3282 had switched it to ssim-cie94, which upstream uses only in comparePNGs() and utils.compare(). SS-069, SS-072, SS-082 and SS-025 on Windows Firefox failed by a few pixels.
  • The static HTTPS server of the Firefox policies test no longer looks up its host name: Python's HTTPServer does, and the reverse lookup hung on the macOS runner for the whole timeout of FF-003.
  • TR-039 counts the console messages with a listener and a future, as upstream does; expect_event() is an action of its own in the trace and flushed the console events the test asserts are not flushed yet.
  • Seven ports of upstream tests that call it.slow() get the slow mark (BC-059, DC-025, HA-058, HE-002, LC-008, SS-039, VI-031); DC-025 timed out on every try on Windows Edge, HE-002 on Windows Firefox.
  • CI runs Linux and Windows on the organization's 64-core runners with 16 workers, as upstream's tests_primary.yml; macOS runs two workers. The Build matrix runs the three browsers on Python 3.12 and Chromium on 3.10, 3.11, 3.13 and 3.14 and on macos-15-xlarge; the Stable matrix drops Edge on macOS.
  • The stacks of a test that hits its timeout show the frames of this repository only, and leave out the threads and tasks every run has.

…runners

- The golden comparison is pixelmatch at the threshold 0.2 again, as the
  matcher of @playwright/test compares; microsoft#3282 switched it to ssim-cie94,
  which upstream uses only in comparePNGs() and utils.compare(). SS-069,
  SS-072, SS-082 and SS-025 on Windows Firefox failed by a few pixels.
- The static HTTPS server of the Firefox policies test does not look up the
  name of its host: Python's HTTPServer does, and the reverse lookup hung
  on the macOS runner for the whole timeout of FF-003.
- TR-039 counts the console messages with a listener and a future, as
  upstream does. expect_event() is an action of its own in the trace, which
  flushed the console events that the test asserts are not flushed yet.
- The ports of the upstream tests that call it.slow() get the slow mark:
  BC-059, DC-025, HA-058, HE-002, LC-008, SS-039 and VI-031 lacked it.
  DC-025 timed out on every try on Windows Edge, HE-002 on Windows Firefox.
- CI runs Linux and Windows on the 64-core runners of the organization with
  16 workers, as upstream's tests_primary.yml does. macOS runs two workers,
  upstream's default of half the cores, where three ran the video and
  remote server tests into the timeout. The Stable job gets 120 minutes.
- The Build matrix runs the three browsers on Python 3.12, the most used
  version, and Chromium on 3.10, 3.11, 3.13 and 3.14 and on macos-15-xlarge.
  The Stable matrix drops Edge on macOS, which upstream does not test.
- The stacks of a test that hits its timeout show the frames of this
  repository only, the test and the client, and leave out the threads and
  tasks that every run has: the frames of pytest, asyncio and execnet around
  them were most of the dump and said nothing about the hang.
@pavelfeldman
Pavel Feldman (pavelfeldman) merged commit a65b9fc into microsoft:main Oct 21, 2026
23 of 29 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants