fix(e2e): keep ServiceAccount across deploy cycles to avoid token fetch race

The deploy script was deleting serviceaccount/headlamp-e2e before recreating
it via kubectl apply. This causes a race: the new deployment pod tries to
mount its service account token before the token is available, resulting in:

    Warning FailedMount: failed to fetch token: serviceaccounts "headlamp-e2e" not found

Fix by removing the kubectl delete serviceaccount line and replacing it with
an idempotent create (--dry-run=client | kubectl apply). This ensures the
ServiceAccount persists across deploy cycles and tokens are available when
pods start.

Co-Authored-By: Paperclip <noreply@paperclip.ing>

.github/workflows/e2e.yaml

@@ -10,94 +10,13 @@ on:
 permissions:
   contents: read
 
-# Only one E2E run at a time: the shared E2E_RELEASE (headlamp-e2e) in
-# headlamp-dev cannot be shared across concurrent runs.
-# cancel-in-progress: false (queue, don't cancel) — cancelling in-flight
-# runs may skip the if: always() teardown, leaving dangling cluster resources.
 concurrency:
   group: e2e-${{ github.repository }}
   cancel-in-progress: false
 
-env:
-  E2E_NAMESPACE: headlamp-dev
-  E2E_RELEASE: headlamp-e2e
-  # Pin to a known-good Headlamp version. Using :latest is risky because
-  # the tag can change between CI runs, causing flaky failures when a newer
-  # image is pulled on some nodes but not others (IfNotPresent pull policy).
-  # Update this when Headlamp is upgraded in production (kube-system).
-  HEADLAMP_VERSION: v0.40.1
-
 jobs:
   e2e:
-    runs-on: runners-privilegedescalation
-    timeout-minutes: 15
-
-    steps:
-      - name: Checkout
-        uses: actions/checkout@v6
-
-      - name: Setup Node.js
-        uses: actions/setup-node@v6
-        with:
-          node-version: '22'
-          cache: 'npm'
-
-      - name: Setup kubectl
-        uses: azure/setup-kubectl@v4
-
-      - name: Install dependencies
-        run: npm ci
-
-      - name: Build plugin
-        run: npx @kinvolk/headlamp-plugin build
-
-      - name: Deploy E2E Headlamp instance
-        run: scripts/deploy-e2e-headlamp.sh
-
-      - name: Load E2E environment
-        run: |
-          if [ -f .env.e2e ]; then
-            cat .env.e2e >> "$GITHUB_ENV"
-          else
-            echo "::error::deploy-e2e-headlamp.sh did not produce .env.e2e"
-            exit 1
-          fi
-
-      - name: Install Playwright browsers
-        run: npx playwright install --with-deps chromium
-
-      - name: Run E2E tests
-        run: npm run e2e
-        env:
-          HEADLAMP_URL: ${{ env.HEADLAMP_URL }}
-          HEADLAMP_TOKEN: ${{ env.HEADLAMP_TOKEN }}
-
-      - name: Collect deployment diagnostics on failure
-        if: failure()
-        run: |
-          echo "=== Pod state ==="
-          kubectl get pods -n "$E2E_NAMESPACE" -l "app.kubernetes.io/instance=$E2E_RELEASE" 2>&1 || true
-          echo "=== Pod describe ==="
-          kubectl describe pods -n "$E2E_NAMESPACE" -l "app.kubernetes.io/instance=$E2E_RELEASE" 2>&1 || true
-          echo "=== Recent namespace events ==="
-          kubectl get events -n "$E2E_NAMESPACE" --sort-by='.lastTimestamp' 2>&1 | tail -20 || true
-
-      - name: Teardown E2E instance
-        if: always()
-        run: scripts/teardown-e2e-headlamp.sh
-
-      - name: Upload Playwright report
-        uses: actions/upload-artifact@v7
-        if: failure()
-        with:
-          name: playwright-report
-          path: playwright-report/
-          retention-days: 7
-
-      - name: Upload test results
-        uses: actions/upload-artifact@v7
-        if: failure()
-        with:
-          name: test-results
-          path: test-results/
-          retention-days: 7
+    uses: privilegedescalation/.github/.github/workflows/plugin-e2e.yaml@main
+    with:
+      node-version: "22"
+      headlamp-version: v0.40.1

scripts/deploy-e2e-headlamp.sh

@@ -60,15 +60,20 @@ kubectl create configmap headlamp-intel-gpu-plugin \
 
 # --- Tear down any existing E2E deployment for a clean start ---
 # Deleting the Deployment forces a fresh pod (new ReplicaSet) regardless of
-# whether the pod spec changed. The ServiceAccount is also deleted for a clean
-# token state. The Service is NOT deleted — leaving it in place avoids an
+# whether the pod spec changed. We do NOT delete the ServiceAccount — keeping
+# it avoids a token-race condition where kubelet tries to mount a volume using a
+# token that has been deleted but the new one isn't ready yet.
+# The Service is NOT deleted — leaving it in place avoids an
 # Endpoints UID race (FailedToUpdateEndpoint) that causes DNS resolution
 # failures. kubectl apply below upserts the Service in-place, and the new
 # pod's IP is added to the existing Endpoints automatically.
 echo ""
 echo "Removing any existing E2E deployment (clean-start)..."
 kubectl delete deployment "${E2E_RELEASE}" -n "$E2E_NAMESPACE" --ignore-not-found --wait
-kubectl delete serviceaccount "${E2E_RELEASE}" -n "$E2E_NAMESPACE" --ignore-not-found --wait
+# ServiceAccount is kept — create it idempotently so the first run works too
+kubectl create serviceaccount "${E2E_RELEASE}" \
+  -n "$E2E_NAMESPACE" \
+  --dry-run=client -o yaml | kubectl apply -f -
 
 # --- Deploy Headlamp via kubectl apply ---
 echo ""