Summary
The desktop Playwright e2e suite's slash-command-menu tests fail intermittently on the Linux e2e lane with expect(locator).toBeVisible() timing out at 10s (element(s) not found), and pass on every rerun. Observed at least four times across three unrelated PRs in the last three days, each time as the only failure in an otherwise green run.
Evidence (all failures are toBeVisible 10s timeouts; every rerun passed)
| Date (UTC) |
PR |
Failing test |
Run/job |
| 2026-08-18 ~16:27 |
#3174 (Windows sandbox — unrelated code) |
slash-command-menu-compacts-the-active-session |
run 32157698387 attempt 1 |
| 2026-08-18 ~10:0x |
#3174 |
slash-command-menu-compacts-the-active-session |
earlier attempt, rerun passed |
| 2026-08-19 ~11:03 |
#3240 (release scripts — unrelated code) |
slash-command-menu-dispatch…-ring-it-into-a-running-turn |
run 32244987684, job 96044046442 |
| 2026-08-20 ~01:50 |
#3265 (installer rollback — unrelated code) |
slash-command-menu-dispatch…-ring-it-into-a-running-turn |
run 32322199338, job 96286486651 |
Playwright artifacts (test-failed-1.png, error-context.md, trace.zip) are attached to each failing run under e2e/test-results/slash-command-menu-*.
Why this matters more since the Apache migration
gh run rerun now requires admin rights on the repository, so contributors cannot re-run a flaked lane; the only recovery is pushing an empty commit, which cancels any in-flight lanes for the ref (concurrency group) and re-triggers every bot review. A single flake therefore costs a full CI cycle plus review noise.
Suggested directions (for the owning team to judge)
- The two failing specs share a shape: open the slash-command menu, then assert a menu item/result is visible within 10s. Either the menu open animation/data load occasionally exceeds 10s on the shared runner, or the click lands before the menu is interactive. A
waitFor on the menu container before the item assertion, or a raised per-assertion timeout for these two specs, would likely close the gap.
- Alternatively mark the two specs
retries: 1 in the Playwright project config if the timing source is judged environmental.
No repository code owned by the reporting PRs touches this suite; filing so the desktop e2e owners can pick it up with the traces above.
Summary
The desktop Playwright e2e suite's slash-command-menu tests fail intermittently on the Linux
e2elane withexpect(locator).toBeVisible()timing out at 10s (element(s) not found), and pass on every rerun. Observed at least four times across three unrelated PRs in the last three days, each time as the only failure in an otherwise green run.Evidence (all failures are
toBeVisible10s timeouts; every rerun passed)slash-command-menu-compacts-the-active-sessionslash-command-menu-compacts-the-active-sessionslash-command-menu-dispatch…-ring-it-into-a-running-turnslash-command-menu-dispatch…-ring-it-into-a-running-turnPlaywright artifacts (
test-failed-1.png,error-context.md,trace.zip) are attached to each failing run undere2e/test-results/slash-command-menu-*.Why this matters more since the Apache migration
gh run rerunnow requires admin rights on the repository, so contributors cannot re-run a flaked lane; the only recovery is pushing an empty commit, which cancels any in-flight lanes for the ref (concurrency group) and re-triggers every bot review. A single flake therefore costs a full CI cycle plus review noise.Suggested directions (for the owning team to judge)
waitForon the menu container before the item assertion, or a raised per-assertion timeout for these two specs, would likely close the gap.retries: 1in the Playwright project config if the timing source is judged environmental.No repository code owned by the reporting PRs touches this suite; filing so the desktop e2e owners can pick it up with the traces above.