M8-R02: report DSMR MQTT source health

app/integrations/mqtt.py

@@ -248,6 +248,7 @@ class MqttManager:
         tls_enabled: bool,
         subscriptions: dict[str, Callable[[bytes], None]],
         base_client_id: str = "home-automation",
+        state_handler: Callable[[str], None] | None = None,
     ) -> bool:
         """Replace one source-owned client and its handlers.
 
@@ -259,6 +260,7 @@ class MqttManager:
         with self._source_lifecycle_lock:
             self._stop_source_client(source_id)
             if not host:
+                self._report_source_state(state_handler, "error", source_id)
                 return False
             self._next_source_generation += 1
             generation = self._next_source_generation
@@ -279,14 +281,18 @@ class MqttManager:
                     if not self._is_current_source_client(source_id, generation, connected_client):
                         return
                     if reason_code.is_failure:
+                        self._source_connected.discard(source_id)
                         logger.warning("DSMR MQTT connection refused for source_id=%s", source_id)
-                        return
-                    self._source_connected.add(source_id)
-                    for topic in captured_subscriptions:
-                        try:
-                            connected_client.subscribe(topic)
-                        except Exception:
-                            logger.exception("DSMR MQTT re-subscribe failed for source_id=%s", source_id)
+                        state = "error"
+                    else:
+                        self._source_connected.add(source_id)
+                        state = "online"
+                        for topic in captured_subscriptions:
+                            try:
+                                connected_client.subscribe(topic)
+                            except Exception:
+                                logger.exception("DSMR MQTT re-subscribe failed for source_id=%s", source_id)
+                self._report_source_state(state_handler, state, source_id)
 
             def _on_disconnect(
                 disconnected_client: mqtt.Client,
@@ -299,6 +305,7 @@ class MqttManager:
                     if not self._is_current_source_client(source_id, generation, disconnected_client):
                         return
                     self._source_connected.discard(source_id)
+                self._report_source_state(state_handler, "error", source_id)
 
             def _on_message(
                 message_client: mqtt.Client,
@@ -334,6 +341,7 @@ class MqttManager:
                     client.tls_set()
                 except Exception:
                     logger.exception("DSMR MQTT TLS setup failed for source_id=%s", source_id)
+                    self._report_source_state(state_handler, "error", source_id)
                     return False
             if username:
                 client.username_pw_set(username=username, password=password or None)
@@ -345,11 +353,13 @@ class MqttManager:
                 self._source_subscriptions[source_id] = captured_subscriptions
                 self._source_generations[source_id] = generation
                 self._source_states[source_id] = _SourceClientState(client, generation)
+            self._report_source_state(state_handler, "connecting", source_id)
             client.loop_start()
             try:
                 client.connect(host=host, port=port, keepalive=60)
             except Exception:
                 logger.exception("DSMR MQTT connect failed (source_id=%s, host=%s)", source_id, host)
+                self._report_source_state(state_handler, "error", source_id)
                 self._stop_source_client(source_id)
                 return False
             return True
@@ -541,6 +551,18 @@ class MqttManager:
             and self._source_clients.get(source_id) is client
         )
 
+    @staticmethod
+    def _report_source_state(
+        state_handler: Callable[[str], None] | None, state: str, source_id: int
+    ) -> None:
+        """Invoke an optional health callback without exposing connection credentials."""
+        if state_handler is None:
+            return
+        try:
+            state_handler(state)
+        except Exception:
+            logger.exception("DSMR MQTT source state update failed for source_id=%s", source_id)
+
 
 # ---------------------------------------------------------------------------
 # Module-level singleton — shared across lifespan and route handlers

app/services/dsmr_ingest.py

@@ -185,9 +185,11 @@ def apply_dsmr_subscription(settings: "Settings | None" = None) -> None:
         # Every source owns a distinct MQTT client, so equal topics from
         # different sources/brokers are dispatchable.  A telegram and tariff
         # topic on the *same* client would overwrite one handler, however.
+        rejected_source_ids: set[int] = set()
         for source_id, snapshot in list(desired.items()):
             if snapshot.tariff_topic and snapshot.topic == snapshot.tariff_topic:
                 logger.error("DSMR source_id=%s rejected: telegram/tariff topic collision", source_id)
+                rejected_source_ids.add(source_id)
                 desired.pop(source_id)
 
         with _subscription_lock:
@@ -203,6 +205,19 @@ def apply_dsmr_subscription(settings: "Settings | None" = None) -> None:
                 _subscription_tokens.pop(source_id, None)
                 set_current_tariff(source_id, None)
 
+        # A disabled source has no installed MQTT owner.  Persist that fact
+        # after invalidating its callback token, so a retained callback cannot
+        # revive an earlier online state while teardown is in progress.
+        for source_id in stale_source_ids:
+            _mark_disabled_source_inactive(source_id)
+
+        # A topic collision is a configuration error for an enabled source,
+        # not a disabled-state transition.  It must therefore replace any
+        # earlier online state even when this process started without a
+        # matching runtime subscription to tear down.
+        for source_id in rejected_source_ids:
+            _mark_rejected_source_error(source_id)
+
         for source_id in stale_source_ids:
             mqtt_manager.remove_source(source_id)
 
@@ -250,6 +265,9 @@ def apply_dsmr_subscription(settings: "Settings | None" = None) -> None:
                 tls_enabled=snapshot.tls_enabled,
                 subscriptions=handlers,
                 base_client_id=base_client_id,
+                state_handler=lambda state, captured=snapshot, captured_token=token: (
+                    handle_captured_source_state(captured, captured_token, state)
+                ),
             )
             if not applied:
                 with _subscription_lock:
@@ -293,6 +311,72 @@ def handle_captured_tariff_message(
         handle_tariff_message(payload_bytes, snapshot.source_id)
 
 
+def handle_captured_source_state(snapshot: DsmrSourceSnapshot, token: object, state: str) -> None:
+    """Persist one active generation's connection health in a short DB session."""
+    with _subscription_lock:
+        if _subscription_tokens.get(snapshot.source_id) is not token:
+            return
+        session_local = get_session_local()
+        session = session_local()
+        try:
+            source = session.get(MeterSource, snapshot.source_id)
+            if source is None or not source.enabled or source.kind != "dsmr_mqtt":
+                return
+            source.status = state
+            source.last_error = "MQTT connection failed." if state == "error" else None
+            source.updated_at = datetime.now(timezone.utc)
+            session.commit()
+        except Exception:
+            session.rollback()
+            logger.exception("DSMR source health update failed for source_id=%s", snapshot.source_id)
+        finally:
+            session.close()
+
+
+def _mark_disabled_source_inactive(source_id: int) -> None:
+    """Clear an obsolete online health state for a disabled DSMR source.
+
+    The caller has already invalidated the source's generation token.  This
+    helper deliberately opens its own short session so reconcile never shares
+    a callback-thread transaction.  Deleted sources simply have no row left
+    to update.
+    """
+    session_local = get_session_local()
+    session = session_local()
+    try:
+        source = session.get(MeterSource, source_id)
+        if source is None or source.enabled or source.kind != "dsmr_mqtt":
+            return
+        source.status = "unknown"
+        source.last_error = None
+        source.updated_at = datetime.now(timezone.utc)
+        session.commit()
+    except Exception:
+        session.rollback()
+        logger.exception("DSMR disabled source health update failed for source_id=%s", source_id)
+    finally:
+        session.close()
+
+
+def _mark_rejected_source_error(source_id: int) -> None:
+    """Persist a non-sensitive error for an enabled source rejected by reconcile."""
+    session_local = get_session_local()
+    session = session_local()
+    try:
+        source = session.get(MeterSource, source_id)
+        if source is None or not source.enabled or source.kind != "dsmr_mqtt":
+            return
+        source.status = "error"
+        source.last_error = "DSMR source configuration invalid."
+        source.updated_at = datetime.now(timezone.utc)
+        session.commit()
+    except Exception:
+        session.rollback()
+        logger.exception("DSMR rejected source health update failed for source_id=%s", source_id)
+    finally:
+        session.close()
+
+
 def _handle_message_inner(payload_bytes: bytes, snapshot: DsmrSourceSnapshot) -> None:
     try:
         data = json.loads(payload_bytes)
@@ -331,7 +415,13 @@ def _handle_message_inner(payload_bytes: bytes, snapshot: DsmrSourceSnapshot) ->
                     payload=data,
                 )
             )
-            session.commit()
+        source = session.get(MeterSource, snapshot.source_id)
+        if source is not None and source.enabled and source.kind == "dsmr_mqtt":
+            source.status = "online"
+            source.last_seen_at = datetime.now(timezone.utc)
+            source.last_error = None
+            source.updated_at = datetime.now(timezone.utc)
+        session.commit()
     except sqlalchemy.exc.IntegrityError:
         session.rollback()
     except Exception: