@testdriverai/runner 7.8.0-test.54 → 7.8.0-test.56

package/lib/automation.js

@@ -662,33 +662,68 @@ class Automation extends EventEmitter {
 
   async _captureScreenshot() {
     const sharp = require('sharp');
-    const tmpFile = path.join(os.tmpdir(), `td_screenshot_${Date.now()}.png`);
+    const maxAttempts = 3;
 
-    try {
-      // Capture screenshot via pyautogui → saves to temp file
-      // Python handles Retina downscale: if physical size differs from logical,
-      // the image is resized to logical dimensions before saving.
-      await runPyAutoGUI(
-        'img = pyautogui.screenshot()\n' +
-        'logical = pyautogui.size()\n' +
-        'if img.size[0] != logical[0] or img.size[1] != logical[1]:\n' +
-        '    from PIL import Image\n' +
-        '    img = img.resize((logical[0], logical[1]), Image.LANCZOS)\n' +
-        'img.save(sys.argv[1], format="PNG")',
-        [tmpFile],
-        20000
-      );
+    for (let attempt = 1; attempt <= maxAttempts; attempt++) {
+      const tmpFile = path.join(os.tmpdir(), `td_screenshot_${Date.now()}.png`);
 
-      // Read the PNG and re-encode with sharp (lossless, no compression)
-      const pngBuffer = fs.readFileSync(tmpFile);
-      const buffer = await sharp(pngBuffer)
-        .png({ compressionLevel: 0 })
-        .toBuffer();
+      try {
+        // Capture screenshot via pyautogui → saves to temp file
+        // Python handles Retina downscale: if physical size differs from logical,
+        // the image is resized to logical dimensions before saving.
+        await runPyAutoGUI(
+          'img = pyautogui.screenshot()\n' +
+          'logical = pyautogui.size()\n' +
+          'if img.size[0] != logical[0] or img.size[1] != logical[1]:\n' +
+          '    from PIL import Image\n' +
+          '    img = img.resize((logical[0], logical[1]), Image.LANCZOS)\n' +
+          'img.save(sys.argv[1], format="PNG")',
+          [tmpFile],
+          20000
+        );
 
-      return buffer.toString('base64');
-    } finally {
-      // Clean up temp file
-      try { fs.unlinkSync(tmpFile); } catch {}
+        // Read the PNG and re-encode with sharp (lossless, no compression)
+        const pngBuffer = fs.readFileSync(tmpFile);
+        const image = sharp(pngBuffer);
+
+        // Detect all-black screenshots (Xvfb/compositor issue)
+        if (IS_LINUX) {
+          const { channels } = await image.stats();
+          // channels[0..2] = R, G, B — check if max pixel value across all channels is near-zero
+          const maxPixel = Math.max(
+            channels[0]?.max ?? 0,
+            channels[1]?.max ?? 0,
+            channels[2]?.max ?? 0
+          );
+          if (maxPixel <= 1) {
+            console.warn(`[automation] Screenshot attempt ${attempt}/${maxAttempts}: image is all black (max pixel=${maxPixel})`);
+            if (attempt < maxAttempts) {
+              // Try to heal: poke the display to trigger a redraw
+              try {
+                await runPyAutoGUI(
+                  "import subprocess; " +
+                  "subprocess.run(['xdotool', 'key', '--clearmodifiers', 'super'], timeout=5); " +
+                  "subprocess.run(['xset', 's', 'off'], timeout=5); " +
+                  "subprocess.run(['xset', 's', 'noblank'], timeout=5); " +
+                  "subprocess.run(['xset', '-dpms'], timeout=5)",
+                  [],
+                  10000
+                );
+              } catch {}
+              // Wait for display to recover
+              await new Promise(r => setTimeout(r, 2000));
+              continue;
+            }
+            console.error('[automation] All screenshot attempts returned black — display may be broken');
+          }
+        }
+
+        const buffer = await image.png({ compressionLevel: 0 }).toBuffer();
+        return buffer.toString('base64');
+      } finally {
+        // Clean up temp file
+        try { fs.unlinkSync(tmpFile); } catch {}
+      }
     }
   }
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@testdriverai/runner",
-  "version": "7.8.0-test.54",
+  "version": "7.8.0-test.56",
   "description": "TestDriver Runner - Ably-based remote automation agent with Node.js automation",
   "main": "index.js",
   "bin": {

package/scripts-desktop/start-desktop.sh

@@ -175,7 +175,130 @@ sleep 1
 
 echo "[start-desktop] Desktop environment ready"
 
+# ─── Helper: restart Xvfb ────────────────────────────────────────────────────
+restart_xvfb() {
+  echo "[watchdog] Restarting Xvfb..."
+  killall Xvfb 2>/dev/null || true
+  sleep 1
+  rm -f /tmp/.X0-lock /tmp/.X11-unix/X0 2>/dev/null
+  Xvfb :0 -ac -screen 0 "${SCREEN_WIDTH}x${SCREEN_HEIGHT}x24" -retro -nolisten tcp &
+  XVFB_PID=$!
+  # Wait for Xvfb to be ready
+  for _w in $(seq 1 10); do
+    xdpyinfo -display :0 > /dev/null 2>&1 && break
+    sleep 1
+  done
+  if ! kill -0 $XVFB_PID 2>/dev/null; then
+    echo "[watchdog] ERROR: Xvfb failed to restart"
+    return 1
+  fi
+  # Re-disable screen blanking & DPMS on the fresh Xvfb
+  xset s off 2>/dev/null || true
+  xset s noblank 2>/dev/null || true
+  xset -dpms 2>/dev/null || true
+  echo "[watchdog] Xvfb restarted (PID: $XVFB_PID)"
+}
+
+# ─── Helper: restart xfce4 (mirrors E2B's defunct-process check) ─────────────
+restart_xfce4() {
+  echo "[watchdog] Restarting xfce4-session..."
+  killall xfce4-session 2>/dev/null || true
+  sleep 1
+  startxfce4 &
+  XFCE4_PID=$!
+  sleep 3
+  killall xfce4-power-manager 2>/dev/null || true
+  killall xfce4-screensaver 2>/dev/null || true
+  xfconf-query -c xfwm4 -p /general/use_compositing -s false 2>/dev/null || true
+  xset s off 2>/dev/null || true
+  xset s noblank 2>/dev/null || true
+  xset -dpms 2>/dev/null || true
+  echo "[watchdog] xfce4-session restarted (PID: $XFCE4_PID)"
+}
+
+# ─── Helper: restart x11vnc ──────────────────────────────────────────────────
+restart_x11vnc() {
+  echo "[watchdog] Restarting x11vnc..."
+  killall x11vnc 2>/dev/null || true
+  sleep 1
+  x11vnc -display :0 -forever -nopw -shared -rfbport 5900 \
+    -noxdamage -fixscreen V=2 \
+    -bg -o /dev/null 2>/dev/null || true
+  echo "[watchdog] x11vnc restarted"
+}
+
+# ─── Watchdog loop ────────────────────────────────────────────────────────────
+# Monitors Xvfb, xfce4-session, and x11vnc health every 10 seconds.
+# Restarts any component that has crashed or become defunct.
+# Also periodically re-disables screen blanking/compositor as belt-and-suspenders.
+WATCHDOG_INTERVAL=10
+BLANKING_RESET_COUNTER=0
+
+watchdog_loop() {
+  while true; do
+    sleep "$WATCHDOG_INTERVAL"
+
+    # ── Check Xvfb ──
+    if ! pgrep -x Xvfb > /dev/null 2>&1; then
+      echo "[watchdog] Xvfb not running! Recovering..."
+      restart_xvfb
+      # x11vnc and xfce need a running Xvfb, restart them too
+      restart_xfce4
+      restart_x11vnc
+      continue
+    fi
+
+    # Verify Xvfb is actually responding (not just a zombie process)
+    if ! xdpyinfo -display :0 > /dev/null 2>&1; then
+      echo "[watchdog] Xvfb process exists but display :0 is unresponsive! Recovering..."
+      restart_xvfb
+      restart_xfce4
+      restart_x11vnc
+      continue
+    fi
+
+    # ── Check xfce4-session (E2B pattern: detect <defunct> zombie) ──
+    XFCE_PID=$(pgrep -x xfce4-session | head -1)
+    if [ -z "$XFCE_PID" ]; then
+      echo "[watchdog] xfce4-session not running! Restarting..."
+      restart_xfce4
+    elif ps aux | grep "$XFCE_PID" | grep -v grep | head -1 | grep -q '<defunct>'; then
+      echo "[watchdog] xfce4-session is defunct (zombie)! Restarting..."
+      restart_xfce4
+    fi
+
+    # ── Check x11vnc ──
+    if ! pgrep -x x11vnc > /dev/null 2>&1; then
+      echo "[watchdog] x11vnc not running! Restarting..."
+      restart_x11vnc
+    fi
+
+    # ── Periodically re-disable screen blanking & compositor (every ~60s) ──
+    BLANKING_RESET_COUNTER=$((BLANKING_RESET_COUNTER + 1))
+    if [ "$BLANKING_RESET_COUNTER" -ge 6 ]; then
+      BLANKING_RESET_COUNTER=0
+      xset s off 2>/dev/null || true
+      xset s noblank 2>/dev/null || true
+      xset -dpms 2>/dev/null || true
+      xfconf-query -c xfwm4 -p /general/use_compositing -s false 2>/dev/null || true
+    fi
+
+    # ── Monitor /dev/shm usage ──
+    if [ -d /dev/shm ]; then
+      SHM_USAGE=$(df /dev/shm 2>/dev/null | awk 'NR==2 {print $5}' | tr -d '%')
+      if [ -n "$SHM_USAGE" ] && [ "$SHM_USAGE" -gt 90 ] 2>/dev/null; then
+        echo "[watchdog] WARNING: /dev/shm is ${SHM_USAGE}% full — X11 may fail to allocate pixmaps"
+      fi
+    fi
+  done
+}
+
+# Start watchdog in background
+watchdog_loop &
+WATCHDOG_PID=$!
+echo "[start-desktop] Watchdog started (PID: $WATCHDOG_PID)"
+
 # Keep the script running so E2B doesn't consider the sandbox stopped
 # Trap signals for clean shutdown
-trap "kill $XVFB_PID $NOVNC_PID 2>/dev/null; exit 0" SIGTERM SIGINT
+trap "kill $XVFB_PID $NOVNC_PID $WATCHDOG_PID 2>/dev/null; exit 0" SIGTERM SIGINT
 wait