Log component outages only once a minute

Right now it is almost once a second, which gets very spammy.

Signed-off-by: Sahas Subramanian <[email protected]>

Author: shsms

src/frequenz/sdk/timeseries/battery_pool/_component_metric_fetcher.py

@@ -74,6 +74,8 @@ class LatestMetricsFetcher(ComponentMetricFetcher, Generic[T], ABC):
 
     _receiver: Receiver[T]
     _max_waiting_time: float
+    _outage: bool
+    _last_log_time: datetime | None
 
     @classmethod
     async def async_new(
@@ -107,6 +109,8 @@ async def async_new(
         # pylint: disable=protected-access
         self._receiver = await self._subscribe()
         self._max_waiting_time = MAX_BATTERY_DATA_AGE_SEC
+        self._outage = False
+        self._last_log_time = None
         return self
 
     async def fetch_next(self) -> ComponentMetricsData | None:
@@ -120,15 +124,27 @@ async def fetch_next(self) -> ComponentMetricsData | None:
             data = await asyncio.wait_for(
                 self._receiver.receive(), self._max_waiting_time
             )
+            self._outage = False
+            self._last_log_time = None
 
         except ChannelClosedError:
             _logger.exception(
                 "Channel for component %d was closed.", self._component_id
             )
             return None
         except asyncio.TimeoutError:
-            # Next time wait infinitely until we receive any message.
-            _logger.debug("Component %d stopped sending data.", self._component_id)
+            now = datetime.now(tz=timezone.utc)
+            if (
+                not self._outage
+                or not self._last_log_time
+                or (now - self._last_log_time).total_seconds() > 60
+            ):
+                self._outage = True
+                self._last_log_time = now
+                _logger.warning(
+                    "Component %d stopped sending data.",
+                    self._component_id,
+                )
             return ComponentMetricsData(
                 self._component_id, datetime.now(tz=timezone.utc), {}
             )