Fix flaky Playwright CI startup

Author: biru-codeastromer

.github/workflows/ci.yml

@@ -54,8 +54,6 @@ jobs:
 
   web-e2e:
     runs-on: ubuntu-latest
-    env:
-      WMG_E2E_PYTHON: python
     steps:
       - uses: actions/checkout@v4
 
@@ -83,6 +81,72 @@ jobs:
         working-directory: web
         run: npx playwright install --with-deps chromium
 
+      - name: Start API server
+        run: |
+          mkdir -p .tmp
+          WMG_DB_URL=sqlite:///./.tmp/e2e.db \
+          WMG_STORAGE_DIR=./.tmp/e2e-storage \
+          WMG_UPLOAD_TOKEN=e2e-token \
+          WMG_AUTO_MIGRATE=true \
+          WMG_ENABLE_METRICS=false \
+          WMG_SEED_DEMO_DATA=true \
+          python -m uvicorn worldmodel_server.main:app --host 127.0.0.1 --port 8100 > .tmp/e2e-api.log 2>&1 &
+          echo $! > .tmp/e2e-api.pid
+
+      - name: Start web server
+        run: |
+          mkdir -p .tmp
+          cd web
+          INTERNAL_API_BASE=http://127.0.0.1:8100 \
+          NEXT_PUBLIC_API_BASE=http://127.0.0.1:8100 \
+          npm run dev -- --hostname 127.0.0.1 --port 3100 > ../.tmp/e2e-web.log 2>&1 &
+          echo $! > ../.tmp/e2e-web.pid
+
+      - name: Wait for local services
+        run: |
+          python - <<'PY'
+          import sys
+          import time
+          from urllib.request import urlopen
+
+          targets = [
+              "http://127.0.0.1:8100/healthz",
+              "http://127.0.0.1:3100",
+          ]
+          deadline = time.time() + 120
+          pending = set(targets)
+
+          while pending and time.time() < deadline:
+              for target in list(pending):
+                  try:
+                      with urlopen(target, timeout=2) as response:
+                          if response.status < 500:
+                              pending.remove(target)
+                  except Exception:
+                      pass
+              if pending:
+                  time.sleep(2)
+
+          if pending:
+              raise SystemExit(f"Services did not start in time: {sorted(pending)}")
+          PY
+
       - name: Run Playwright smoke tests
         working-directory: web
+        env:
+          PLAYWRIGHT_DISABLE_WEBSERVER: "1"
         run: npm run test:e2e
+
+      - name: Dump server logs on failure
+        if: failure()
+        run: |
+          echo "===== API LOG ====="
+          cat .tmp/e2e-api.log || true
+          echo "===== WEB LOG ====="
+          cat .tmp/e2e-web.log || true
+
+      - name: Stop background services
+        if: always()
+        run: |
+          kill $(cat .tmp/e2e-api.pid) 2>/dev/null || true
+          kill $(cat .tmp/e2e-web.pid) 2>/dev/null || true

web/playwright.config.ts

@@ -4,6 +4,7 @@ import { defineConfig } from "@playwright/test";
 
 const repoRoot = path.resolve(__dirname, "..");
 const pythonExecutable = process.env.WMG_E2E_PYTHON ?? ".venv/bin/python";
+const useManagedWebServers = process.env.PLAYWRIGHT_DISABLE_WEBSERVER !== "1";
 
 export default defineConfig({
   testDir: "./tests/e2e",
@@ -15,32 +16,34 @@ export default defineConfig({
     baseURL: "http://127.0.0.1:3100",
     trace: "on-first-retry"
   },
-  webServer: [
-    {
-      command: `${pythonExecutable} -m uvicorn worldmodel_server.main:app --host 127.0.0.1 --port 8100`,
-      cwd: repoRoot,
-      env: {
-        WMG_DB_URL: "sqlite:///./.tmp/e2e.db",
-        WMG_STORAGE_DIR: "./.tmp/e2e-storage",
-        WMG_UPLOAD_TOKEN: "e2e-token",
-        WMG_AUTO_MIGRATE: "true",
-        WMG_ENABLE_METRICS: "false",
-        WMG_SEED_DEMO_DATA: "true"
-      },
-      url: "http://127.0.0.1:8100/healthz",
-      reuseExistingServer: !process.env.CI,
-      timeout: 120_000
-    },
-    {
-      command: "npm run dev -- --hostname 127.0.0.1 --port 3100",
-      cwd: path.resolve(repoRoot, "web"),
-      env: {
-        INTERNAL_API_BASE: "http://127.0.0.1:8100",
-        NEXT_PUBLIC_API_BASE: "http://127.0.0.1:8100"
-      },
-      url: "http://127.0.0.1:3100",
-      reuseExistingServer: !process.env.CI,
-      timeout: 120_000
-    }
-  ]
+  webServer: useManagedWebServers
+    ? [
+        {
+          command: `${pythonExecutable} -m uvicorn worldmodel_server.main:app --host 127.0.0.1 --port 8100`,
+          cwd: repoRoot,
+          env: {
+            WMG_DB_URL: "sqlite:///./.tmp/e2e.db",
+            WMG_STORAGE_DIR: "./.tmp/e2e-storage",
+            WMG_UPLOAD_TOKEN: "e2e-token",
+            WMG_AUTO_MIGRATE: "true",
+            WMG_ENABLE_METRICS: "false",
+            WMG_SEED_DEMO_DATA: "true"
+          },
+          url: "http://127.0.0.1:8100/healthz",
+          reuseExistingServer: !process.env.CI,
+          timeout: 120_000
+        },
+        {
+          command: "npm run dev -- --hostname 127.0.0.1 --port 3100",
+          cwd: path.resolve(repoRoot, "web"),
+          env: {
+            INTERNAL_API_BASE: "http://127.0.0.1:8100",
+            NEXT_PUBLIC_API_BASE: "http://127.0.0.1:8100"
+          },
+          url: "http://127.0.0.1:3100",
+          reuseExistingServer: !process.env.CI,
+          timeout: 120_000
+        }
+      ]
+    : undefined
 });