Fixed watchdog test

libs/python/computer-server/test_connection.py

@@ -13,10 +13,16 @@ import argparse
 import sys
 
 
-async def test_connection(host="localhost", port=8000, keep_alive=False):
+async def test_connection(host="localhost", port=8000, keep_alive=False, container_name=None):
     """Test connection to the Computer Server."""
-    uri = f"ws://{host}:{port}/ws"
-    print(f"Connecting to {uri}...")
+    if container_name:
+        # Container mode: use WSS with container domain and port 8443
+        uri = f"wss://{container_name}.containers.cloud.trycua.com:8443/ws"
+        print(f"Connecting to container {container_name} at {uri}...")
+    else:
+        # Local mode: use WS with specified host and port
+        uri = f"ws://{host}:{port}/ws"
+        print(f"Connecting to local server at {uri}...")
 
     try:
         async with websockets.connect(uri) as websocket:
@@ -54,13 +60,23 @@ def parse_args():
     parser = argparse.ArgumentParser(description="Test connection to Computer Server")
     parser.add_argument("--host", default="localhost", help="Host address (default: localhost)")
     parser.add_argument("--port", type=int, default=8000, help="Port number (default: 8000)")
+    parser.add_argument("--container-name", help="Container name for cloud connection (uses WSS and port 8443)")
     parser.add_argument("--keep-alive", action="store_true", help="Keep connection alive")
     return parser.parse_args()
 
 
 async def main():
     args = parse_args()
-    success = await test_connection(args.host, args.port, args.keep_alive)
+    
+    # Convert hyphenated argument to underscore for function parameter
+    container_name = getattr(args, 'container_name', None)
+    
+    success = await test_connection(
+        host=args.host, 
+        port=args.port, 
+        keep_alive=args.keep_alive,
+        container_name=container_name
+    )
     return 0 if success else 1
 
 

tests/watchdog.py

@@ -49,7 +49,7 @@ async def computer():
         await computer.disconnect()
 
 
-@pytest.mark.asyncio
+@pytest.mark.asyncio(loop_scope="session")
 async def test_simple_server_ping(computer):
     """
     Simple test to verify server connectivity before running watchdog tests.
@@ -66,7 +66,7 @@ async def test_simple_server_ping(computer):
         pytest.fail(f"Basic server connectivity test failed: {e}")
 
 
-@pytest.mark.asyncio
+@pytest.mark.asyncio(loop_scope="session")
 async def test_watchdog_recovery_after_hanging_command(computer):
     """
     Test that the watchdog can recover the server after a hanging command.
@@ -80,15 +80,16 @@ async def test_watchdog_recovery_after_hanging_command(computer):
     print("Starting watchdog recovery test...")
     
     async def hanging_command():
-        """Execute a command that takes 5 minutes to complete."""
+        """Execute a command that sleeps forever to hang the server."""
         try:
-            print("Starting hanging command (sleep 300)...")
-            result = await computer.interface.run_command("sleep 300")
-            print(f"Hanging command completed: {result}")
-            return result
+            print("Starting hanging command (sleep infinity)...")
+            # Use a very long sleep that should never complete naturally
+            result = await computer.interface.run_command("sleep 999999")
+            print(f"Hanging command completed unexpectedly: {result}")
+            return True  # Should never reach here if watchdog works
         except Exception as e:
-            print(f"Hanging command failed (expected if watchdog restarts): {e}")
-            return None
+            print(f"Hanging command interrupted (expected if watchdog restarts): {e}")
+            return None  # Expected result when watchdog kills the process
     
     async def ping_server():
         """Ping the server every 30 seconds with echo commands."""
@@ -183,13 +184,22 @@ async def test_watchdog_recovery_after_hanging_command(computer):
             # Test passes if we had some successful pings, indicating recovery
             assert successful_pings > 0, f"No successful pings detected. Server may not have recovered."
             
+            # Check if hanging command was killed (indicating watchdog restart)
+            if hanging_result is None:
+                print("✅ SUCCESS: Hanging command was killed - watchdog restart detected")
+            elif hanging_result is True:
+                print("⚠️  WARNING: Hanging command completed naturally - watchdog may not have restarted")
+            
             # If we had failures followed by successes, that indicates watchdog recovery
             if failed_pings > 0 and successful_pings > 0:
                 print("✅ SUCCESS: Watchdog recovery detected - server became unresponsive then recovered")
+                # Additional check: hanging command should be None if watchdog worked
+                assert hanging_result is None, "Expected hanging command to be killed by watchdog restart"
             elif successful_pings > 0 and failed_pings == 0:
                 print("✅ SUCCESS: Server remained responsive throughout test")
             
             print(f"Test completed: {successful_pings} successful pings, {failed_pings} failed pings")
+            print(f"Hanging command result: {hanging_result} (None = killed by watchdog, True = completed naturally)")
         else:
             pytest.fail("Ping task did not complete - unable to assess server recovery")