Stabilize frontend Playwright CI runner

.github/workflows/tests.yml

@@ -213,7 +213,9 @@ jobs:
             exit 1
           fi
           mkdir -p output/playwright
-          "${docker_cmd[@]}" run --rm --ipc=host --network host \
+          scripts/ci/runner-diagnostics.sh "before Playwright PR ${MATRIX_SUITE}/${MATRIX_PROJECT}" -- "${docker_cmd[@]}"
+          SYSTEMD_INHIBIT_REASON="Frontend Playwright PR ${MATRIX_SUITE}/${MATRIX_PROJECT}" \
+          scripts/ci/with-systemd-inhibit.sh "${docker_cmd[@]}" run --rm --ipc=host --network host \
             --volume "$PWD:/source:ro" \
             --volume "$PWD/output/playwright:/work/output/playwright" \
             --workdir /work \
@@ -254,6 +256,11 @@ jobs:
               fi
             '
 
+      - name: Runner diagnostics after Playwright failure
+        if: failure() || cancelled()
+        continue-on-error: true
+        run: scripts/ci/runner-diagnostics.sh "after Playwright PR ${{ matrix.suite }}/${{ matrix.project }}"
+
       - name: Upload Playwright report
         if: failure() || cancelled()
         continue-on-error: true
@@ -278,7 +285,7 @@ jobs:
     timeout-minutes: 60
     strategy:
       fail-fast: false
-      max-parallel: 4
+      max-parallel: 2
       matrix:
         browser: [chromium, webkit, firefox]
         device: [mobile, desktop, tablet]
@@ -297,6 +304,7 @@ jobs:
       PLAYWRIGHT_ARTIFACT_NAMESPACE: e2e-full-${{ matrix.browser }}-${{ matrix.device }}-${{ matrix.role }}
       PLAYWRIGHT_REPORTER_MODE: line-html
       PLAYWRIGHT_WORKERS: ${{ matrix.browser == 'firefox' && '1' || '2' }}
+      PLAYWRIGHT_VIDEO_MODE: off
     steps:
       - name: Repair self-hosted workspace permissions
         shell: bash
@@ -381,7 +389,9 @@ jobs:
             exit 1
           fi
           mkdir -p output/playwright
-          "${docker_cmd[@]}" run --rm --ipc=host --network host \
+          scripts/ci/runner-diagnostics.sh "before Playwright full ${MATRIX_BROWSER}/${MATRIX_DEVICE}/${MATRIX_ROLE}" -- "${docker_cmd[@]}"
+          SYSTEMD_INHIBIT_REASON="Frontend Playwright full ${MATRIX_BROWSER}/${MATRIX_DEVICE}/${MATRIX_ROLE}" \
+          scripts/ci/with-systemd-inhibit.sh "${docker_cmd[@]}" run --rm --ipc=host --network host \
             --volume "$PWD:/source:ro" \
             --volume "$PWD/output/playwright:/work/output/playwright" \
             --workdir /work \
@@ -390,6 +400,7 @@ jobs:
             --env PLAYWRIGHT_ARTIFACT_NAMESPACE="$PLAYWRIGHT_ARTIFACT_NAMESPACE" \
             --env PLAYWRIGHT_REPORTER_MODE="$PLAYWRIGHT_REPORTER_MODE" \
             --env PLAYWRIGHT_WORKERS="$PLAYWRIGHT_WORKERS" \
+            --env PLAYWRIGHT_VIDEO_MODE="$PLAYWRIGHT_VIDEO_MODE" \
             --env PLAYWRIGHT_DEV_PORT="$playwright_dev_port" \
             --env MATRIX_ROLE="$MATRIX_ROLE" \
             --env MATRIX_BROWSER="$MATRIX_BROWSER" \
@@ -417,6 +428,11 @@ jobs:
               npm run test:e2e:full:slice -- --role="$MATRIX_ROLE" --project="$MATRIX_BROWSER-$MATRIX_DEVICE"
             '
 
+      - name: Runner diagnostics after Playwright failure
+        if: failure() || cancelled()
+        continue-on-error: true
+        run: scripts/ci/runner-diagnostics.sh "after Playwright full ${{ matrix.browser }}/${{ matrix.device }}/${{ matrix.role }}"
+
       - name: Upload Playwright report
         if: failure() || cancelled()
         continue-on-error: true

README.md

@@ -137,6 +137,11 @@ PLAYWRIGHT_PARALLEL_WORKERS_WEBKIT=1
 
 The runner fails fast if the combined worker count exceeds 5.
 
+GitHub Actions keeps the full browser/device/role matrix stable by capping full-suite
+matrix parallelism at 2 jobs, wrapping Docker Playwright runs with `systemd-inhibit`
+when available, and setting `PLAYWRIGHT_VIDEO_MODE=off` for the full matrix. Traces
+and screenshots are still retained on failure.
+
 Artifacts and summaries:
 
 - `output/playwright/ci-parallel-report/index.html`

playwright.config.ts

@@ -13,6 +13,9 @@ const htmlReportOutputFolder = path.join(artifactRoot, "report");
 const configuredWorkers = Number(process.env.PLAYWRIGHT_WORKERS || 2);
 const workers = Number.isFinite(configuredWorkers) && configuredWorkers > 0 ? configuredWorkers : 2;
 const reporterMode = (process.env.PLAYWRIGHT_REPORTER_MODE || "").trim();
+const configuredVideoMode = (process.env.PLAYWRIGHT_VIDEO_MODE || "retain-on-failure").trim();
+const allowedVideoModes = new Set(["off", "on", "retain-on-failure", "on-first-retry"]);
+const videoMode = allowedVideoModes.has(configuredVideoMode) ? configuredVideoMode : "retain-on-failure";
 const reporter =
   reporterMode === "line-html"
     ? [["line"], ["html", { open: "never", outputFolder: htmlReportOutputFolder }]]
@@ -59,7 +62,7 @@ export default defineConfig({
     baseURL,
     trace: "retain-on-failure",
     screenshot: "only-on-failure",
-    video: "retain-on-failure",
+    video: videoMode,
   },
   projects: [
     buildProject("chromium-mobile", "chromium", "Pixel 5"),

scripts/ci/runner-diagnostics.sh

@@ -0,0 +1,98 @@
+#!/usr/bin/env bash
+set -uo pipefail
+
+label="${1:-runner}"
+if [[ $# -gt 0 ]]; then
+  shift
+fi
+
+if [[ "${1:-}" == "--" ]]; then
+  shift
+fi
+
+docker_cmd=("$@")
+
+run_bounded() {
+  local seconds="$1"
+  shift
+
+  if command -v timeout >/dev/null 2>&1; then
+    timeout "${seconds}s" "$@" || true
+    return
+  fi
+
+  "$@" || true
+}
+
+run_silent_bounded() {
+  local seconds="$1"
+  shift
+
+  if command -v timeout >/dev/null 2>&1; then
+    timeout "${seconds}s" "$@" >/dev/null 2>&1
+    return
+  fi
+
+  "$@" >/dev/null 2>&1
+}
+
+resolve_docker_cmd() {
+  if [[ "${#docker_cmd[@]}" -gt 0 ]]; then
+    return 0
+  fi
+
+  if command -v docker >/dev/null 2>&1 && run_silent_bounded 10 docker info; then
+    docker_cmd=(docker)
+    return 0
+  fi
+
+  if command -v sudo >/dev/null 2>&1 && run_silent_bounded 10 sudo -n docker info; then
+    docker_cmd=(sudo docker)
+    return 0
+  fi
+
+  docker_cmd=()
+  return 1
+}
+
+echo "::group::Runner diagnostics (${label})"
+date --iso-8601=seconds 2>/dev/null || date
+hostname || true
+uname -a || true
+
+if command -v free >/dev/null 2>&1; then
+  echo
+  echo "Memory:"
+  free -h || true
+fi
+
+if command -v df >/dev/null 2>&1; then
+  echo
+  echo "Disk:"
+  df -h / /tmp "$PWD" 2>/dev/null || df -h / /tmp 2>/dev/null || true
+fi
+
+if command -v systemd-inhibit >/dev/null 2>&1; then
+  echo
+  echo "Systemd inhibitors:"
+  run_bounded 10 systemd-inhibit --list
+fi
+
+if resolve_docker_cmd; then
+  echo
+  echo "Docker info:"
+  run_bounded 20 "${docker_cmd[@]}" info
+
+  echo
+  echo "Docker disk usage:"
+  run_bounded 20 "${docker_cmd[@]}" system df
+
+  echo
+  echo "Recent Docker containers:"
+  run_bounded 20 "${docker_cmd[@]}" ps -a --no-trunc --format "table {{.ID}}\t{{.Status}}\t{{.Names}}" | head -n 20 || true
+else
+  echo
+  echo "Docker diagnostics unavailable to this runner user."
+fi
+
+echo "::endgroup::"

scripts/ci/with-systemd-inhibit.sh

@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if [[ "$#" -eq 0 ]]; then
+  echo "Usage: $0 <command> [args...]" >&2
+  exit 64
+fi
+
+reason="${SYSTEMD_INHIBIT_REASON:-GitHub Actions frontend E2E job is running}"
+
+can_use_systemd_inhibit() {
+  if ! command -v systemd-inhibit >/dev/null 2>&1; then
+    return 1
+  fi
+
+  if command -v timeout >/dev/null 2>&1; then
+    timeout 10s systemd-inhibit --what=sleep:shutdown --mode=block --who=frontend-ci --why="frontend-ci probe" true >/dev/null 2>&1
+    return
+  fi
+
+  systemd-inhibit --what=sleep:shutdown --mode=block --who=frontend-ci --why="frontend-ci probe" true >/dev/null 2>&1
+}
+
+if can_use_systemd_inhibit; then
+  exec systemd-inhibit --what=sleep:shutdown --mode=block --who=frontend-ci --why="$reason" "$@"
+fi
+
+exec "$@"

tests/unit/playwright-full-workflow.spec.js

@@ -28,4 +28,21 @@ describe("Playwright full E2E workflow grouping", () => {
       "output/playwright/test-lists/${{ matrix.browser }}-${{ matrix.device }}-${{ matrix.role }}.txt"
     );
   });
+
+  it("keeps full-suite runner pressure bounded and diagnosable", () => {
+    const source = workflowSource();
+
+    expect(source).toMatch(/e2e-full:[\s\S]*?max-parallel: 2/u);
+    expect(source).toMatch(/e2e-full:[\s\S]*?PLAYWRIGHT_VIDEO_MODE: off/u);
+    expect(source).toContain("scripts/ci/with-systemd-inhibit.sh");
+    expect(source).toContain("scripts/ci/runner-diagnostics.sh");
+  });
+
+  it("allows CI to reduce Playwright video artifact pressure", () => {
+    const source = readFileSync(join(root, "playwright.config.ts"), "utf8");
+
+    expect(source).toContain("PLAYWRIGHT_VIDEO_MODE");
+    expect(source).toContain('"retain-on-failure"');
+    expect(source).toContain('"on-first-retry"');
+  });
 });