ecamp / ecamp3 / 27863398884

43%

devel: 43%

Pull #10105

github

e2e: reduce flaky test failures (#9643)

The Playwright e2e suite produced inconsistent failures, with different
tests failing across runs (HTTP cache, dashboard filters, PDF downloads).
Two root causes were identified and addressed in playwright.config.ts:

1. expect.timeout was 8s. The suite asserts against the full stack
   (frontend -> reverse-proxy -> API -> DB), and data-loading assertions
   (e.g. the dashboard filter chips re-rendering after page.reload() once
   the category/status endpoints have loaded) regularly need more than 8s
   on a loaded stack. Raised to 15s so these assertions stop racing the
   data load while still failing reasonably fast.

2. workers was undefined. In CI each project (chromium/firefox/webkit/
   behavior-tests) runs as its own isolated job against its own backend,
   so the default is fine there. Locally, `npx playwright test` runs all
   projects at once against a single shared backend; on a many-core dev
   machine the default spawns many workers that all hammer that single
   stack (the heavy client-side PDF tests in particular saturate the API
   and starve other tests), which is the main source of the "different
   failures every run" flakiness. Cap local workers at 4 to keep the load
   on the shared stack bounded and deterministic regardless of machine
   size; CI behaviour is unchanged and a faster local run is still
   available via --workers.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Signed-off-by: BacLuc <lucius.bachmann@clubpage.ch>

840 of 4746 branches covered (17.7%)

4856 of 11182 relevant lines covered (43.43%)

983.78 hits per line