Merge remote-tracking branch 'origin/main' into feat_reputation

Author: isaac1227

app/deployer.py

@@ -668,12 +668,6 @@ def run_frontend(self):
                     "/var/run/docker.sock not found, please check if Docker is running and Docker Compose is installed."
                 )
 
-        try:
-            subprocess.check_call(["nvidia-smi"])
-            self.gpu_available = True
-        except Exception:
-            logging.info("No GPU available for the frontend, nodes will be deploy in CPU mode")
-
         network_name = f"{os.environ['USER']}_nebula-net-base"
 
         # Create the Docker network
@@ -684,7 +678,6 @@ def run_frontend(self):
         environment = {
             "NEBULA_CONTROLLER_NAME": os.environ["USER"],
             "NEBULA_PRODUCTION": self.production,
-            "NEBULA_GPU_AVAILABLE": self.gpu_available,
             "NEBULA_ADVANCED_ANALYTICS": self.advanced_analytics,
             "NEBULA_FRONTEND_LOG": "/nebula/app/logs/frontend.log",
             "NEBULA_LOGS_DIR": "/nebula/app/logs/",
@@ -756,6 +749,12 @@ def run_controller(self):
                 )
 
         network_name = f"{os.environ['USER']}_nebula-net-base"
+        
+        try:
+            subprocess.check_call(["nvidia-smi"])
+            self.gpu_available = True
+        except Exception:
+            logging.info("No GPU available for the frontend, nodes will be deploy in CPU mode")
 
         # Create the Docker network
         base = DockerUtils.create_docker_network(network_name)
@@ -790,6 +789,11 @@ def run_controller(self):
             ],
             extra_hosts={"host.docker.internal": "host-gateway"},
             port_bindings={self.controller_port: self.controller_port},
+            device_requests=[{
+                "Driver": "nvidia",
+                "Count": -1,
+                "Capabilities": [["gpu"]],
+            }] if self.gpu_available else None,
         )
 
         networking_config = client.api.create_networking_config({

nebula/addons/gps/nebulagps.py

@@ -19,16 +19,17 @@ def __init__(self, config, addr, update_interval: float = 5.0, verbose=False):
         self._config = config
         self._addr = addr
         self.update_interval = update_interval  # Frequency
-        self.running = False
         self._node_locations = {}  # Dictionary for storing node locations
         self._broadcast_socket = None
         self._nodes_location_lock = Locker("nodes_location_lock", async_lock=True)
         self._verbose = verbose
+        self._running = asyncio.Event()
+        self._background_tasks = []  # Track background tasks
 
     async def start(self):
         """Starts the GPS service, sending and receiving locations."""
         logging.info("Starting NebulaGPS service...")
-        self.running = True
+        self._running.set()
 
         # Create broadcast socket
         self._broadcast_socket = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
@@ -38,20 +39,39 @@ async def start(self):
         self._broadcast_socket.bind(("", self.BROADCAST_PORT))
 
         # Start sending and receiving tasks
-        asyncio.create_task(self._send_location_loop())
-        asyncio.create_task(self._receive_location_loop())
-        asyncio.create_task(self._notify_geolocs())
+        self._background_tasks = [
+            asyncio.create_task(self._send_location_loop(), name="NebulaGPS_send_location"),
+            asyncio.create_task(self._receive_location_loop(), name="NebulaGPS_receive_location"),
+            asyncio.create_task(self._notify_geolocs(), name="NebulaGPS_notify_geolocs"),
+        ]
 
     async def stop(self):
         """Stops the GPS service."""
-        logging.info("Stopping NebulaGPS service...")
-        self.running = False
+        logging.info("🛑  Stopping NebulaGPS service...")
+        self._running.clear()
+        logging.info("🛑  NebulaGPS _running event cleared")
+
+        # Cancel all background tasks
+        if self._background_tasks:
+            logging.info(f"🛑  Cancelling {len(self._background_tasks)} background tasks...")
+            for task in self._background_tasks:
+                if not task.done():
+                    task.cancel()
+                    try:
+                        await task
+                    except asyncio.CancelledError:
+                        pass
+            self._background_tasks.clear()
+            logging.info("🛑  All background tasks cancelled")
+
         if self._broadcast_socket:
             self._broadcast_socket.close()
             self._broadcast_socket = None
+            logging.info("🛑  NebulaGPS broadcast socket closed")
+        logging.info("✅  NebulaGPS service stopped successfully")
 
     async def is_running(self):
-        return self.running
+        return self._running.is_set()
 
     async def get_geoloc(self):
         latitude = self._config.participant["mobility_args"]["latitude"]
@@ -64,7 +84,18 @@ async def calculate_distance(self, self_lat, self_long, other_lat, other_long):
 
     async def _send_location_loop(self):
         """Send the geolocation periodically by broadcast."""
-        while self.running:
+        while await self.is_running():
+            # Check if learning cycle has finished
+            try:
+                from nebula.core.network.communications import CommunicationsManager
+
+                cm = CommunicationsManager.get_instance()
+                if cm.learning_finished():
+                    logging.info("GPS: Learning cycle finished, stopping location broadcast")
+                    break
+            except Exception:
+                pass  # If we can't get the communications manager, continue
+
             latitude, longitude = await self.get_geoloc()  # Obtener ubicación actual
             message = f"GPS-UPDATE {self._addr} {latitude} {longitude}"
             self._broadcast_socket.sendto(message.encode(), (self.BROADCAST_IP, self.BROADCAST_PORT))
@@ -74,7 +105,18 @@ async def _send_location_loop(self):
 
     async def _receive_location_loop(self):
         """Listens to and stores geolocations from other nodes."""
-        while self.running:
+        while await self.is_running():
+            # Check if learning cycle has finished
+            try:
+                from nebula.core.network.communications import CommunicationsManager
+
+                cm = CommunicationsManager.get_instance()
+                if cm.learning_finished():
+                    logging.info("GPS: Learning cycle finished, stopping location reception")
+                    break
+            except Exception:
+                pass  # If we can't get the communications manager, continue
+
             try:
                 data, addr = await asyncio.get_running_loop().run_in_executor(
                     None, self._broadcast_socket.recvfrom, 1024
@@ -91,7 +133,18 @@ async def _receive_location_loop(self):
                 logging.exception(f"Error receiving GPS update: {e}")
 
     async def _notify_geolocs(self):
-        while True:
+        while await self.is_running():
+            # Check if learning cycle has finished
+            try:
+                from nebula.core.network.communications import CommunicationsManager
+
+                cm = CommunicationsManager.get_instance()
+                if cm.learning_finished():
+                    logging.info("GPS: Learning cycle finished, stopping geolocation notifications")
+                    break
+            except Exception:
+                pass  # If we can't get the communications manager, continue
+
             await asyncio.sleep(self.update_interval)
             await self._nodes_location_lock.acquire_async()
             geolocs: dict = self._node_locations.copy()
@@ -102,7 +155,7 @@ async def _notify_geolocs(self):
                 for addr, (lat, long) in geolocs.items():
                     dist = await self.calculate_distance(self_lat, self_long, lat, long)
                     distances[addr] = (dist, (lat, long))
-                
+
                 self._config.update_nodes_distance(distances)
                 gpsevent = GPSEvent(distances)
                 asyncio.create_task(EventManager.get_instance().publish_addonevent(gpsevent))

nebula/addons/mobility.py

@@ -50,24 +50,28 @@ def __init__(self, config, verbose=False):
         """
         logging.info("Starting mobility module...")
         self.config = config
-        self.grace_time = self.config.participant["mobility_args"]["grace_time_mobility"]
-        self.period = self.config.participant["mobility_args"]["change_geo_interval"]
+        self._verbose = verbose
+        self._running = asyncio.Event()
+        self._nodes_distances = {}
+        self._nodes_distances_lock = Locker("nodes_distances_lock", async_lock=True)
+        self._mobility_task = None  # Track the background task
+
+        # Mobility configuration
         self.mobility = self.config.participant["mobility_args"]["mobility"]
         self.mobility_type = self.config.participant["mobility_args"]["mobility_type"]
-        self.radius_federation = float(self.config.participant["mobility_args"]["radius_federation"])
-        self.scheme_mobility = self.config.participant["mobility_args"]["scheme_mobility"]
-        self.round_frequency = int(self.config.participant["mobility_args"]["round_frequency"])
+        self.grace_time = self.config.participant["mobility_args"]["grace_time_mobility"]
+        self.period = self.config.participant["mobility_args"]["change_geo_interval"]
         # INFO: These values may change according to the needs of the federation
         self.max_distance_with_direct_connections = 150  # meters
         self.max_movement_random_strategy = 50  # meters
         self.max_movement_nearest_strategy = 50  # meters
         self.max_initiate_approximation = self.max_distance_with_direct_connections * 1.2
+        self.radius_federation = float(config.participant["mobility_args"]["radius_federation"])
+        self.scheme_mobility = config.participant["mobility_args"]["scheme_mobility"]
+        self.round_frequency = int(config.participant["mobility_args"]["round_frequency"])
         # Logging box with mobility information
         mobility_msg = f"Mobility: {self.mobility}\nMobility type: {self.mobility_type}\nRadius federation: {self.radius_federation}\nScheme mobility: {self.scheme_mobility}\nEach {self.round_frequency} rounds"
         print_msg_box(msg=mobility_msg, indent=2, title="Mobility information")
-        self._nodes_distances = {}
-        self._nodes_distances_lock = Locker("nodes_distances_lock", async_lock=True)
-        self._verbose = verbose
 
     @cached_property
     def cm(self):
@@ -103,8 +107,30 @@ async def start(self):
         """
         await EventManager.get_instance().subscribe_addonevent(GPSEvent, self.update_nodes_distances)
         await EventManager.get_instance().subscribe_addonevent(GPSEvent, self.update_nodes_distances)
-        task = asyncio.create_task(self.run_mobility())
-        return task
+        self._running.set()
+        self._mobility_task = asyncio.create_task(self.run_mobility(), name="Mobility_run_mobility")
+        return self._mobility_task
+
+    async def stop(self):
+        """
+        Stops the mobility module.
+        """
+        logging.info("Stopping Mobility module...")
+        self._running.clear()
+
+        # Cancel the background task
+        if self._mobility_task and not self._mobility_task.done():
+            logging.info("🛑  Cancelling Mobility background task...")
+            self._mobility_task.cancel()
+            try:
+                await self._mobility_task
+            except asyncio.CancelledError:
+                pass
+            self._mobility_task = None
+            logging.info("🛑  Mobility background task cancelled")
+
+    async def is_running(self):
+        return self._running.is_set()
 
     async def update_nodes_distances(self, gpsevent: GPSEvent):
         distances = await gpsevent.get_event_data()
@@ -138,7 +164,7 @@ async def run_mobility(self):
         if not self.mobility:
             return
         # await asyncio.sleep(self.grace_time)
-        while True:
+        while await self.is_running():
             await self.change_geo_location()
             await asyncio.sleep(self.period)
 

nebula/addons/networksimulation/nebulanetworksimulator.py

@@ -26,15 +26,15 @@ def __init__(self, changing_interval, interface, verbose=False):
         self._network_conditions = self.NETWORK_CONDITIONS.copy()
         self._network_conditions_lock = Locker("network_conditions_lock", async_lock=True)
         self._current_network_conditions = {}
-        self._running = False
+        self._running = asyncio.Event()
 
     @cached_property
     def cm(self):
         return CommunicationsManager.get_instance()
 
     async def start(self):
         logging.info("🌐  Nebula Network Simulator starting...")
-        self._running = True
+        self._running.set()
         grace_time = self.cm.config.participant["mobility_args"]["grace_time_mobility"]
         # if self._verbose: logging.info(f"Waiting {grace_time}s to start applying network conditions based on distances between devices")
         # await asyncio.sleep(grace_time)
@@ -43,37 +43,49 @@ async def start(self):
         )
 
     async def stop(self):
-        self._running = False
+        logging.info("🌐  Nebula Network Simulator stopping...")
+        self._running.clear()
+
+    async def is_running(self):
+        return self._running.is_set()
 
     async def _change_network_conditions_based_on_distances(self, gpsevent: GPSEvent):
         distances = await gpsevent.get_event_data()
-        await asyncio.sleep(self._refresh_interval)
-        if self._verbose:
-            logging.info("Refresh | conditions based on distances...")
-        try:
-            for addr, (distance, _) in distances.items():
-                if distance is None:
-                    # If the distance is not found, we skip the node
-                    continue
-                conditions = await self._calculate_network_conditions(distance)
-                # Only update the network conditions if they have changed
-                if addr not in self._current_network_conditions or self._current_network_conditions[addr] != conditions:
-                    addr_ip = addr.split(":")[0]
-                    self._set_network_condition_for_addr(
-                        self._node_interface, addr_ip, conditions["bandwidth"], conditions["delay"]
-                    )
-                    self._set_network_condition_for_multicast(
-                        self._node_interface, addr_ip, self.IP_MULTICAST, conditions["bandwidth"], conditions["delay"]
-                    )
-                    async with self._network_conditions_lock:
-                        self._current_network_conditions[addr] = conditions
-                else:
-                    if self._verbose:
-                        logging.info("network conditions havent changed since last time")
-        except KeyError:
-            logging.exception(f"📍  Connection {addr} not found")
-        except Exception:
-            logging.exception("📍  Error changing connections based on distance")
+        while await self.is_running():
+            if self._verbose:
+                logging.info("Refresh | conditions based on distances...")
+            try:
+                for addr, (distance, _) in distances.items():
+                    if distance is None:
+                        # If the distance is not found, we skip the node
+                        continue
+                    conditions = await self._calculate_network_conditions(distance)
+                    # Only update the network conditions if they have changed
+                    if (
+                        addr not in self._current_network_conditions
+                        or self._current_network_conditions[addr] != conditions
+                    ):
+                        addr_ip = addr.split(":")[0]
+                        self._set_network_condition_for_addr(
+                            self._node_interface, addr_ip, conditions["bandwidth"], conditions["delay"]
+                        )
+                        self._set_network_condition_for_multicast(
+                            self._node_interface,
+                            addr_ip,
+                            self.IP_MULTICAST,
+                            conditions["bandwidth"],
+                            conditions["delay"],
+                        )
+                        async with self._network_conditions_lock:
+                            self._current_network_conditions[addr] = conditions
+                    else:
+                        if self._verbose:
+                            logging.info("network conditions havent changed since last time")
+            except KeyError:
+                logging.exception(f"📍  Connection {addr} not found")
+            except Exception:
+                logging.exception("📍  Error changing connections based on distance")
+            await asyncio.sleep(self._refresh_interval)
 
     async def set_thresholds(self, thresholds: dict):
         async with self._network_conditions_lock: