test(mcp): diagnose startup stalls (do not merge)

.github/scripts/recover-mcp-chromium-traces.mjs

@@ -0,0 +1,62 @@
+#!/usr/bin/env node
+
+import crypto from 'node:crypto';
+import fs from 'node:fs';
+import path from 'node:path';
+
+const traceRoot = path.resolve('test-results/mcp-chromium-traces');
+const retryDeadline = Date.now() + 10_000;
+
+async function recoverTrace(source) {
+  const target = path.join(path.dirname(source), `recovered-${crypto.randomUUID()}.pftrace`);
+  while (true) {
+    let size;
+    try {
+      size = (await fs.promises.stat(source)).size;
+    } catch (error) {
+      if (error.code === 'ENOENT')
+        return;
+      throw error;
+    }
+
+    if (size) {
+      try {
+        await fs.promises.rename(source, target);
+        console.log(`Recovered Chromium startup trace: ${target}`);
+        return;
+      } catch (error) {
+        if (error.code === 'ENOENT')
+          return;
+        if (error.code !== 'EACCES' && error.code !== 'EPERM')
+          throw error;
+      }
+    }
+
+    if (Date.now() >= retryDeadline) {
+      console.warn(`Chromium startup trace has no recoverable data: ${source}`);
+      return;
+    }
+    await new Promise(resolve => setTimeout(resolve, 200));
+  }
+}
+
+let traceDirectories;
+try {
+  traceDirectories = await fs.promises.readdir(traceRoot, { withFileTypes: true });
+} catch (error) {
+  if (error.code === 'ENOENT')
+    process.exit(0);
+  throw error;
+}
+
+const pending = [];
+for (const directory of traceDirectories) {
+  if (!directory.isDirectory())
+    continue;
+  const traceDirectory = path.join(traceRoot, directory.name);
+  for (const file of await fs.promises.readdir(traceDirectory)) {
+    if (!file.endsWith('.pftrace'))
+      pending.push(recoverTrace(path.join(traceDirectory, file)));
+  }
+}
+await Promise.all(pending);

.github/scripts/run-mcp-startup-diagnostic.sh

@@ -0,0 +1,64 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+project="$1"
+
+run_tests() {
+  set +e
+  "$@"
+  local test_status=$?
+  node .github/scripts/recover-mcp-chromium-traces.mjs
+  local recovery_status=$?
+  if [[ "$test_status" != "0" ]]; then
+    return "$test_status"
+  fi
+  return "$recovery_status"
+}
+
+if [[ "$(uname)" != "Linux" ]]; then
+  run_tests npm run test-mcp -- --project="$project" --workers=1 --timeout=60000
+  exit $?
+fi
+
+health_dir="mcp-runner-health"
+health_file="$health_dir/runner-health.tsv"
+mkdir -p "$health_dir"
+
+monitor_health() {
+  printf 'timestamp\tload1\tmem_available_kb\tswap_free_kb\tdisk_available_kb\tprocesses\tnode_rss_kb\tchromium_rss_kb\txvfb_rss_kb\n'
+  while true; do
+    local timestamp load1 mem_available swap_free disk_available processes node_rss chromium_rss xvfb_rss
+    timestamp="$(date -u +%Y-%m-%dT%H:%M:%SZ)"
+    load1="$(awk '{ print $1 }' /proc/loadavg)"
+    mem_available="$(awk '/^MemAvailable:/ { print $2 }' /proc/meminfo)"
+    swap_free="$(awk '/^SwapFree:/ { print $2 }' /proc/meminfo)"
+    disk_available="$(df -Pk . | awk 'NR == 2 { print $4 }')"
+    processes="$(ps -e --no-headers | wc -l)"
+    node_rss="$(ps -eo rss=,comm= | awk '$2 == "node" { sum += $1 } END { print sum + 0 }')"
+    chromium_rss="$(ps -eo rss=,comm= | awk '$2 ~ /^(chrome|chromium)$/ { sum += $1 } END { print sum + 0 }')"
+    xvfb_rss="$(ps -eo rss=,comm= | awk '$2 == "Xvfb" { sum += $1 } END { print sum + 0 }')"
+    printf '%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s\n' \
+      "$timestamp" "$load1" "$mem_available" "$swap_free" "$disk_available" \
+      "$processes" "$node_rss" "$chromium_rss" "$xvfb_rss"
+    sleep 15
+  done
+}
+
+monitor_health > "$health_file" &
+monitor_pid=$!
+cleanup() {
+  kill "$monitor_pid" 2>/dev/null || true
+  wait "$monitor_pid" 2>/dev/null || true
+}
+trap cleanup EXIT
+
+set +e
+run_tests timeout --signal=INT --kill-after=30s 35m npm run test-mcp -- --project="$project" --workers=1 --timeout=60000
+status=$?
+set -e
+
+if [[ "$status" == "124" ]]; then
+  echo "::error::Ubuntu MCP diagnostic exceeded 35 minutes; stopping before the hosted runner is lost."
+fi
+exit "$status"

.github/workflows/infra.yml

@@ -13,6 +13,7 @@ on:
 jobs:
   doc-and-lint:
     name: "docs & lint"
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     runs-on: ubuntu-24.04
     steps:
     - uses: actions/checkout@v6
@@ -35,6 +36,7 @@ jobs:
       continue-on-error: true
   lint-snippets:
     name: "Lint snippets"
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v6

.github/workflows/tests_components.yml

@@ -25,6 +25,7 @@ env:
 jobs:
   test_components:
     name: ${{ matrix.os }} - Node.js ${{ matrix.node-version }}
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     strategy:
       fail-fast: false
       matrix:

.github/workflows/tests_extension.yml

@@ -35,6 +35,7 @@ env:
 jobs:
   test_extension:
     name: ${{ matrix.os }}
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     strategy:
       fail-fast: false
       matrix:

.github/workflows/tests_mcp.yml

@@ -1,4 +1,4 @@
-name: MCP
+name: MCP startup diagnostic (temporary)
 
 on:
   push:
@@ -30,24 +30,20 @@ env:
 
 jobs:
   test_mcp:
-    name: ${{ matrix.os }} - ${{ matrix.project }}
+    name: DIAGNOSTIC - ${{ matrix.os }} - ${{ matrix.project }} - sample ${{ matrix.sample }}
     # Used for authentication of flakiness upload
     environment: ${{ github.event_name == 'push' && 'allow-uploading-flakiness-results' || null }}
     strategy:
       fail-fast: false
+      # Avoid losing an entire platform batch if the hosted runner fleet is under pressure.
+      max-parallel: 5
       matrix:
-        os: [ubuntu-latest, macos-latest]
-        project: [chrome, chromium, firefox, webkit]
-        include:
-          - os: windows-latest
-            project: chrome
-          - os: windows-latest
-            project: chromium
-          - os: windows-latest
-            project: firefox
-          - os: windows-latest
-            project: webkit
-          - os: windows-latest
+        # Temporary Perfetto follow-up: ten fresh runners for each affected configuration.
+        os: [windows-latest, ubuntu-latest]
+        project: [chrome, msedge]
+        sample: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
+        exclude:
+          - os: ubuntu-latest
             project: msedge
     runs-on: ${{ matrix.os }}
     permissions:
@@ -58,8 +54,30 @@ jobs:
     - uses: ./.github/actions/run-test
       with:
         node-version: "22"
-        command: npm run test-mcp -- --project=${{ matrix.project }}
-        bot-name: "mcp-${{ matrix.os }}-${{ matrix.project }}"
+        # Keep the full suite and serial execution so setup/order/cleanup effects remain intact.
+        command: bash .github/scripts/run-mcp-startup-diagnostic.sh ${{ matrix.project }}
+        bot-name: "mcp-startup-diagnostic-${{ matrix.os }}-${{ matrix.project }}-sample-${{ matrix.sample }}"
         flakiness-client-id: ${{ secrets.AZURE_FLAKINESS_DASHBOARD_CLIENT_ID }}
         flakiness-tenant-id: ${{ secrets.AZURE_FLAKINESS_DASHBOARD_TENANT_ID }}
         flakiness-subscription-id: ${{ secrets.AZURE_FLAKINESS_DASHBOARD_SUBSCRIPTION_ID }}
+    - name: Upload raw MCP startup timelines
+      if: always()
+      uses: actions/upload-artifact@v4
+      with:
+        name: "mcp-startup-timelines-${{ matrix.os }}-${{ matrix.project }}-sample-${{ matrix.sample }}"
+        path: test-results/mcp-startup-timelines/*.jsonl
+        if-no-files-found: error
+    - name: Upload retained Chromium startup traces
+      if: always()
+      uses: actions/upload-artifact@v4
+      with:
+        name: "mcp-chromium-traces-${{ matrix.os }}-${{ matrix.project }}-sample-${{ matrix.sample }}"
+        path: test-results/mcp-chromium-traces/**/*.pftrace
+        if-no-files-found: warn
+    - name: Upload Ubuntu runner health
+      if: always() && matrix.os == 'ubuntu-latest'
+      uses: actions/upload-artifact@v4
+      with:
+        name: "mcp-runner-health-${{ matrix.os }}-${{ matrix.project }}-sample-${{ matrix.sample }}"
+        path: mcp-runner-health/runner-health.tsv
+        if-no-files-found: error

.github/workflows/tests_primary.yml

@@ -32,6 +32,7 @@ env:
 jobs:
   test_linux:
     name: ${{ matrix.os }} (${{ matrix.browser }} - Node.js ${{ matrix.node-version }})
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     environment: ${{ github.event_name == 'push' && 'allow-uploading-flakiness-results' || null }}
     strategy:
       fail-fast: false
@@ -67,6 +68,7 @@ jobs:
 
   test_test_runner:
     name: Test Runner
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     environment: ${{ github.event_name == 'push' && 'allow-uploading-flakiness-results' || null }}
     strategy:
       fail-fast: false
@@ -142,6 +144,7 @@ jobs:
 
   test_web_components:
     name: Web Components - ${{ matrix.package }}
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     environment: ${{ github.event_name == 'push' && 'allow-uploading-flakiness-results' || null }}
     permissions:
       id-token: write   # This is required for OIDC login (azure/login) to succeed
@@ -165,6 +168,7 @@ jobs:
 
   test_vscode_extension:
     name: VSCode Extension
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     runs-on: ubuntu-latest
     steps:
     - uses: actions/checkout@v6
@@ -201,6 +205,7 @@ jobs:
 
   test_package_installations:
     name: "Installation Test ${{ matrix.os }}"
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     environment: ${{ github.event_name == 'push' && 'allow-uploading-flakiness-results' || null }}
     strategy:
       fail-fast: false
@@ -237,6 +242,7 @@ jobs:
 
   test_clock_frozen_time_linux:
     name: time library - ${{ matrix.clock }}
+    if: github.event_name != 'pull_request' || github.head_ref != 'skn0tt-diagnose-mcp-startup-stalls'
     environment: ${{ github.event_name == 'push' && 'allow-uploading-flakiness-results' || null }}
     permissions:
       id-token: write   # This is required for OIDC login (azure/login) to succeed

packages/playwright-core/src/server/browser.ts

@@ -18,6 +18,7 @@ import fs from 'fs';
 
 import { makeSocketPath } from '@utils/fileUtils';
 import { createGuid } from '@utils/crypto';
+import { startupTrace } from '@utils/startupTrace';
 import { BrowserContext, validateBrowserContextOptions } from './browserContext';
 import { Download } from './download';
 import { SdkObject } from './instrumentation';
@@ -231,6 +232,7 @@ export class BrowserServer {
       endpoint = this._pipeSocketPath;
     }
 
+    startupTrace('browser.registry-publication.start', { browserPid: this._browser.options.browserProcess.process?.pid, title, endpoint });
     const browserInfo: BrowserInfo = {
       guid: this._browser.guid,
       browserName: this._browser.options.browserType,
@@ -243,10 +245,12 @@ export class BrowserServer {
       workspaceDir: options.workspaceDir,
       metadata: options.metadata,
     });
+    startupTrace('browser.registry-publication.end', { browserPid: this._browser.options.browserProcess.process?.pid, title, endpoint });
     return { endpoint };
   }
 
   async stop() {
+    startupTrace('browser.server-teardown.start', { browserPid: this._browser.options.browserProcess.process?.pid });
     if (!this._browser.options.userDataDir)
       await serverRegistry.delete(this._browser.guid);
     if (this._pipeSocketPath && process.platform !== 'win32')
@@ -256,6 +260,7 @@ export class BrowserServer {
     this._pipeServer = undefined;
     this._wsServer = undefined;
     this._isStarted = false;
+    startupTrace('browser.server-teardown.end', { browserPid: this._browser.options.browserProcess.process?.pid });
   }
 
   private async _socketPath() {