fix(e2e): retain-on-failure traces + a globalTimeout under the 45m CI cap

playwright.config.ts

@@ -24,12 +24,31 @@ export default defineConfig({
 	forbidOnly: !!process.env.CI,
 	retries: process.env.CI ? 1 : 0,
 	workers: 1,
+	// The shared quality.yml Playwright job is `timeout-minutes: 45`, and a job
+	// cancelled by that cap produces NO verdict: Playwright never prints its
+	// tally, the `if: failure()` trace upload never fires, and the
+	// `if: always()` report upload does not run on a cancelled job either — the
+	// run you most need to read is the one that leaves nothing behind, and it
+	// still renders as "fail" in `gh pr checks` while carrying no information.
+	// Runs cancelled at ~45m16s have been observed in this fleet. Measured
+	// overhead before `Run Playwright tests` starts is 2.0-2.4 min and the
+	// uploads after it take seconds, so 38m keeps ~7 min of margin while
+	// guaranteeing both a tally and the artifacts that explain it.
+	globalTimeout: 38 * 60_000,
 	reporter: 'list',
 	outputDir: 'test-results',
 	use: {
 		baseURL: resolveBaseUrl(),
 		storageState: path.resolve(__dirname, 'tests/e2e/.auth/admin.json'),
-		trace: 'on-first-retry',
+		// `on-first-retry` writes a trace only when a retry actually happens, so
+		// the trace artifact is a function of `retries`. Off CI `retries` is 0
+		// above, so a local failure has never produced a trace at all; on CI it
+		// traces the SECOND attempt only, which means the failure that does not
+		// reproduce — the one actually worth a trace — leaves no record of the
+		// attempt that failed. `retain-on-failure` traces every attempt and
+		// keeps the ones that failed: strictly more informative, and
+		// independent of the retry count.
+		trace: 'retain-on-failure',
 		screenshot: 'only-on-failure',
 	},
 	projects: [

tests/e2e/playwright.config.ts

@@ -71,6 +71,17 @@ export default defineConfig({
 	forbidOnly: !!process.env.CI,
 	retries: process.env.CI ? 1 : 0,
 	workers: 1,
+	// The shared quality.yml Playwright job is `timeout-minutes: 45`, and a job
+	// cancelled by that cap produces NO verdict: Playwright never prints its
+	// tally, the `if: failure()` trace upload never fires, and the
+	// `if: always()` report upload does not run on a cancelled job either — the
+	// run you most need to read is the one that leaves nothing behind, and it
+	// still renders as "fail" in `gh pr checks` while carrying no information.
+	// Runs cancelled at ~45m16s have been observed in this fleet. Measured
+	// overhead before `Run Playwright tests` starts is 2.0-2.4 min and the
+	// uploads after it take seconds, so 38m keeps ~7 min of margin while
+	// guaranteeing both a tally and the artifacts that explain it.
+	globalTimeout: 38 * 60_000,
 	reporter: [
 		['html', { open: 'never', outputFolder: path.join(APP_ROOT, 'playwright-report') }],
 		['list'],
@@ -80,7 +91,15 @@ export default defineConfig({
 	use: {
 		baseURL: BASE_URL,
 		storageState: path.resolve(__dirname, '.auth', 'admin.json'),
-		trace: 'on-first-retry',
+		// `on-first-retry` writes a trace only when a retry actually happens, so
+		// the trace artifact is a function of `retries`. Off CI `retries` is 0
+		// above, so a local failure has never produced a trace at all; on CI it
+		// traces the SECOND attempt only, which means the failure that does not
+		// reproduce — the one actually worth a trace — leaves no record of the
+		// attempt that failed. `retain-on-failure` traces every attempt and
+		// keeps the ones that failed: strictly more informative, and
+		// independent of the retry count.
+		trace: 'retain-on-failure',
 		screenshot: 'only-on-failure',
 	},