Add fallback scanner for Matter nodes

Author: marcelveldt

matter_server/server/device_controller.py

@@ -72,6 +72,7 @@
 NODE_PING_TIMEOUT = 10
 NODE_PING_TIMEOUT_BATTERY_POWERED = 60
 NODE_MDNS_BACKOFF = 60
+FALLBACK_NODE_SCANNER_INTERVAL = 1800
 
 MDNS_TYPE_OPERATIONAL_NODE = "_matter._tcp.local."
 MDNS_TYPE_COMMISSIONABLE_NODE = "_matterc._udp.local."
@@ -113,6 +114,7 @@ def __init__(
         self._aiobrowser: AsyncServiceBrowser | None = None
         self._aiozc: AsyncZeroconf | None = None
         self._sdk_executor = ThreadPoolExecutor(max_workers=1)
+        self._fallback_node_scanner_timer: asyncio.TimerHandle | None = None
 
     async def initialize(self) -> None:
         """Async initialize of controller."""
@@ -174,20 +176,24 @@ async def start(self) -> None:
             services,
             handlers=[self._on_mdns_service_state_change],
         )
+        # set-up fallback node scanner
+        asyncio.create_task(self._fallback_node_scanner())
 
     async def stop(self) -> None:
         """Handle logic on server stop."""
         if self.chip_controller is None:
             raise RuntimeError("Device Controller not initialized.")
-        # unsubscribe all node subscriptions
-        for sub in self._subscriptions.values():
-            await self._call_sdk(sub.Shutdown)
-        self._subscriptions = {}
-        # shutdown (and cleanup) mdns browser
+        # shutdown (and cleanup) mdns browser and fallback node scanner
         if self._aiobrowser:
             await self._aiobrowser.async_cancel()
+        if self._fallback_node_scanner_timer:
+            self._fallback_node_scanner_timer.cancel()
         if self._aiozc:
             await self._aiozc.async_close()
+        # unsubscribe all node subscriptions
+        for sub in self._subscriptions.values():
+            await self._call_sdk(sub.Shutdown)
+        self._subscriptions = {}
         # shutdown the sdk device controller
         await self._call_sdk(self.chip_controller.Shutdown)
         LOGGER.debug("Stopped.")
@@ -1250,3 +1256,30 @@ async def _node_offline(self, node_id: int) -> None:
         node.available = False
         self.server.signal_event(EventType.NODE_UPDATED, node)
         LOGGER.info("Marked node %s as offline", node_id)
+
+    async def _fallback_node_scanner(self) -> None:
+        """Scan for operational nodes in the background that are missed by mdns."""
+        # This code could/should be removed in the future and is added to have a fallback
+        # to discover operational nodes that got somehow missed by zeroconf.
+        # the issue in zeroconf is being investigated and in the meanwhile we have this fallback.
+        for node_id, node in self._nodes.items():
+            if node.available:
+                continue
+            now = time.time()
+            last_seen = self._node_last_seen.get(node_id, 0)
+            if now - last_seen < FALLBACK_NODE_SCANNER_INTERVAL:
+                continue
+            if await self.ping_node(node_id, attempts=3):
+                LOGGER.info("Node %s discovered using fallback ping", node_id)
+                await self._setup_node(node_id)
+
+        def reschedule_self() -> None:
+            self._fallback_node_scanner_timer = None
+            asyncio.create_task(self._fallback_node_scanner())
+
+        # reschedule task to run at next interval
+        if TYPE_CHECKING:
+            assert self.server.loop
+        self._fallback_node_scanner_timer = self.server.loop.call_later(
+            FALLBACK_NODE_SCANNER_INTERVAL, reschedule_self
+        )