[Pavel Feldman (pavelfeldman)]

Summary

• The self-hosted Windows runner starts the test launcher at I/O priority Low; Node and Firefox storage threads inherit it and SQLite flushes stall until tests time out.
• Add a PowerShell helper to the run-test action that resets its own I/O priority to Normal, logs the change, and runs the test command. Covers the bash and pwsh paths.
• BC-076 waits for the simple.json response before checking request counts, fixing a fetch race.

Fixes #43704

[github-actions]

Test results for "tests 1"

1 flaky

⚠️ [installation tests] › playwright-test-plugin.spec.ts:48 › pnpm: @playwright/test plugin should work `@package-installations-ubuntu-latest`

23575 passed, 706 skipped

---

Merge workflow run.

[github-actions]

Test results for "MCP"

5189 passed, 614 skipped

---

Merge workflow run.

[github-actions]

Test results for "tests 2"

5 failed
❌ [android-page] › page/page-event-pageerror.spec.ts:20 › ER-001 should fire
❌ [android-page] › page/selectors-get-by.spec.ts:117 › SE-082 getByLabel should work with aria-labelledby
❌ [android-page] › page/selectors-get-by.spec.ts:122 › SE-083 getByLabel should prioritize aria-labelledby over native label
❌ [android-page] › page/selectors-get-by.spec.ts:140 › SE-086 getByLabel should prioritize aria-labelledby over aria-label
❌ [android-page] › page/selectors-role.spec.ts:436 › SE-060 should support name

11 flaky

⚠️ [chromium-page] › page/page-screenshot.spec.ts:534 › page screenshot › SS-039 should take fullPage screenshots during navigation `@tracing-chromium`
⚠️ [chromium-library] › library/beforeunload.spec.ts:20 › BU-001 should close browser with beforeunload page `@chromium-macos-15-large`
⚠️ [chromium-library] › library/browsercontext-basic.spec.ts:36 › BC-002 should be able to click across browser contexts `@chromium-macos-15-large`
⚠️ [chromium-library] › library/video.spec.ts:690 › screencast › VI-028 should capture full viewport `@chromium-ubuntu-22.04-node24`
⚠️ [chromium-library] › library/har.spec.ts:1174 › tracing.startHar › HA-062 should include pages `@chrome-macos-latest`
⚠️ [chromium-library] › library/screencast.spec.ts:28 › SC-001 screencast.start delivers frames via onFrame callback `@chrome-macos-latest`
⚠️ [chromium-library] › library/video.spec.ts:762 › screencast › VI-030 should work with video+trace `@chromium-ubuntu-22.04-arm-node22`
⚠️ [webkit-library] › library/heap.spec.ts:304 › LE-014 cycle handles `@webkit-playwright-x64-server2025-64-core`
⚠️ [webkit-library] › library/trace-viewer.spec.ts:2009 › TV-089 canvas disabled title `@webkit-macos-15-xlarge`
⚠️ [webkit-library] › library/tracing.spec.ts:565 › TR-023 should produce screencast frames crop `@webkit-macos-15-xlarge`
⚠️ [webkit-library] › library/page-clock.spec.ts:644 › CK-053 AbortSignal.timeout `@webkit-macos-15-large`

126935 passed, 4123 skipped

---

Merge workflow run.

[github-actions]

🟢 All five failures predate this PR

Hi, I'm the Playwright bot. I compared the latest CI report with the PR changes and refreshed CI test history.

Details

No failure appears caused by this PR. All five failures are in [android-page]. The same tests already fail with the same errors in the Android job on this PR's base commit, f9a841f, before these changes. They also fail in the earlier run for #43703.

Pre-existing failures

• Page-error stack: page/page-event-pageerror.spec.ts:20 › ER-001 should fire expects Uncaught Error and an <anonymous> stack frame; Android reports Error and an unnamed frame instead. The exact test failed in all 98 recorded runs outside this PR. This is a persistent failure, not an intermittent flake.
• ARIA labels and accessible names: page/selectors-get-by.spec.ts:117 › SE-082, :122 › SE-083, and :140 › SE-086 time out resolving getByLabel() with aria-labelledby. page/selectors-role.spec.ts:436 › SE-060 should support name receives an empty accessible name instead of Chosen name. Each exact test failed in 30 of 98 recorded runs outside this PR, including the base-commit run above with matching errors.

The Windows I/O-priority helper does not run in the Ubuntu Android job; its Linux test command is unchanged. The other change only waits for a response in the separate BC-076 browser-context reuse test. Neither change reaches these failing assertions.

Tests that passed on retry are not counted as failures here.

_{Triaged by the Playwright bot - agent run}