ci: retry E2E wrangler boot once with port cleanup

.github/actions/e2e-env/action.yml

@@ -87,34 +87,76 @@ runs:
               echo "✓ Database initialization complete"
 
         - name: Start Pages Functions server
+          # Retries the SERVER BOOT ONLY, once, on a fresh process — never on test
+          # failures (those are a separate step). A dead process or an unmet 180s
+          # budget on attempt 1 kills any leftover wrangler/workerd/esbuild holding
+          # :8788 and relaunches once before failing the job for real. See #625:
+          # no observed CI failure has actually been a slow boot (worst seen was a
+          # few seconds), so the budget is left at 180s rather than padded on
+          # speculation — the retry covers one-off boot hiccups a longer wait
+          # wouldn't fix anyway (e.g. a stuck port from a prior process).
           shell: bash
           run: |
-              ./frontend/node_modules/.bin/wrangler pages dev frontend/dist --port 8788 --persist-to .wrangler/state > /tmp/wrangler.log 2>&1 &
-              WRANGLER_PID=$!
-
-              echo "Waiting for server to start..."
-              i=1
-              while [ $i -le 90 ]; do
-                  if ! kill -0 $WRANGLER_PID 2>/dev/null; then
-                      echo "✗ Wrangler process exited unexpectedly"
-                      echo "Wrangler logs:"
-                      cat /tmp/wrangler.log
-                      exit 1
-                  fi
-                  if curl -s http://localhost:8788/ > /dev/null 2>&1; then
-                      echo "✓ Server is ready on port 8788!"
-                      exit 0
-                  fi
-                  if [ $((i % 20)) -eq 0 ]; then
-                      echo "--- Wrangler log (iteration $i) ---"
-                      tail -20 /tmp/wrangler.log 2>/dev/null
-                      echo "---"
-                  fi
-                  echo "Waiting for server... ($i/90)"
-                  sleep 2
-                  i=$((i + 1))
-              done
-              echo "✗ Server failed to start after 180 seconds"
-              echo "Wrangler logs:"
-              cat /tmp/wrangler.log
+              MAX_ITER=90    # 90 * 2s = 180s budget per attempt
+              LOG_EVERY=10   # surface wrangler.log every 10 iterations (20s) instead of 40s
+
+              start_wrangler() {
+                  ./frontend/node_modules/.bin/wrangler pages dev frontend/dist --port 8788 --persist-to .wrangler/state > /tmp/wrangler.log 2>&1 &
+                  WRANGLER_PID=$!
+              }
+
+              kill_wrangler() {
+                  kill "$WRANGLER_PID" 2>/dev/null || true
+                  # wrangler forks esbuild + workerd children that don't die with the
+                  # parent; free the port explicitly so the retry doesn't fail to bind.
+                  for pid in $(lsof -ti:8788 2>/dev/null); do
+                      kill -9 "$pid" 2>/dev/null || true
+                  done
+              }
+
+              wait_for_server() {
+                  attempt=$1
+                  echo "Waiting for server to start (attempt $attempt/2)..."
+                  i=1
+                  while [ $i -le $MAX_ITER ]; do
+                      if ! kill -0 $WRANGLER_PID 2>/dev/null; then
+                          echo "✗ Wrangler process exited unexpectedly (attempt $attempt)"
+                          echo "Wrangler logs:"
+                          cat /tmp/wrangler.log
+                          return 1
+                      fi
+                      if curl -s http://localhost:8788/ > /dev/null 2>&1; then
+                          echo "✓ Server is ready on port 8788! (attempt $attempt, $((i * 2))s)"
+                          return 0
+                      fi
+                      if [ $((i % LOG_EVERY)) -eq 0 ]; then
+                          echo "--- Wrangler log so far (attempt $attempt, iteration $i) ---"
+                          tail -20 /tmp/wrangler.log 2>/dev/null
+                          echo "---"
+                      fi
+                      echo "Waiting for server... ($i/$MAX_ITER, attempt $attempt)"
+                      sleep 2
+                      i=$((i + 1))
+                  done
+                  echo "✗ Server failed to start after $((MAX_ITER * 2)) seconds (attempt $attempt)"
+                  echo "Wrangler logs:"
+                  cat /tmp/wrangler.log
+                  return 1
+              }
+
+              start_wrangler
+              if wait_for_server 1; then
+                  exit 0
+              fi
+
+              echo "Startup attempt 1 failed — killing any leftover process and retrying boot once..."
+              kill_wrangler
+              sleep 2
+
+              start_wrangler
+              if wait_for_server 2; then
+                  exit 0
+              fi
+
+              echo "✗ Server failed to start after 2 attempts — genuine startup failure, not transient boot slowness"
               exit 1

Makefile

@@ -61,11 +61,25 @@ e2e-setup: build ## Init + seed an isolated local D1 for E2E (safe to re-run)
 	SEED_SQL=$$(node scripts/seed-e2e-admin.mjs --email "$(E2E_ADMIN_EMAIL)" --password "$(E2E_ADMIN_PASSWORD)"); \
 	$(WRANGLER) d1 execute settimes-production-db --local --persist-to $(E2E_STATE) --command="$$SEED_SQL"
 
-e2e-serve: ## Start wrangler for E2E in the background (writes pidfile)
-	$(WRANGLER) pages dev frontend/dist --port 8788 --persist-to $(E2E_STATE) > $(E2E_STATE)/wrangler.log 2>&1 & \
-	echo $$! > $(E2E_PID); \
-	i=0; until curl -s -o /dev/null http://localhost:8788/ || [ $$i -ge 45 ]; do i=$$((i+1)); sleep 2; done; \
-	curl -s -o /dev/null http://localhost:8788/ || { echo "wrangler failed to start; log:"; tail -20 $(E2E_STATE)/wrangler.log; exit 1; }
+e2e-serve: ## Start wrangler for E2E in the background (writes pidfile); retries boot once, never test failures (mirrors e2e-env action.yml, #625)
+	try_boot() { \
+		$(WRANGLER) pages dev frontend/dist --port 8788 --persist-to $(E2E_STATE) > $(E2E_STATE)/wrangler.log 2>&1 & \
+		echo $$! > $(E2E_PID); \
+		i=1; \
+		while [ $$i -le 90 ]; do \
+			kill -0 "$$(cat $(E2E_PID))" 2>/dev/null || { echo "wrangler process exited unexpectedly (attempt $$1)"; tail -20 $(E2E_STATE)/wrangler.log; return 1; }; \
+			curl -s -o /dev/null http://localhost:8788/ && { echo "server ready on port 8788 (attempt $$1, $$((i*2))s)"; return 0; }; \
+			[ $$((i % 10)) -eq 0 ] && { echo "--- wrangler log so far (attempt $$1, iteration $$i) ---"; tail -20 $(E2E_STATE)/wrangler.log; }; \
+			i=$$((i+1)); sleep 2; \
+		done; \
+		echo "wrangler failed to start after 180s (attempt $$1)"; tail -20 $(E2E_STATE)/wrangler.log; return 1; \
+	}; \
+	try_boot 1 && exit 0; \
+	kill "$$(cat $(E2E_PID))" 2>/dev/null || true; \
+	for pid in $$(lsof -ti:8788 2>/dev/null); do kill -9 "$$pid" 2>/dev/null || true; done; \
+	sleep 2; \
+	try_boot 2 && exit 0; \
+	echo "wrangler failed to start after 2 attempts"; exit 1
 
 e2e-run: ## Run Playwright (server must be up; credentials required even for --list)
 	ADMIN_EMAIL="$(E2E_ADMIN_EMAIL)" ADMIN_PASSWORD="$(E2E_ADMIN_PASSWORD)" npx playwright test $(SPEC) --reporter=list