Unexport unavailable metrics in Prometheus (#125492)

homeassistant/components/prometheus/__init__.py

@@ -76,6 +76,8 @@ from homeassistant.util.unit_conversion import TemperatureConverter
 _LOGGER = logging.getLogger(__name__)
 
 API_ENDPOINT = "/api/prometheus"
+IGNORED_STATES = frozenset({STATE_UNAVAILABLE, STATE_UNKNOWN})
+
 
 DOMAIN = "prometheus"
 CONF_FILTER = "filter"
@@ -211,14 +213,6 @@ class PrometheusMetrics:
         """Add/update a state in Prometheus."""
         entity_id = state.entity_id
         _LOGGER.debug("Handling state update for %s", entity_id)
-        domain, _ = hacore.split_entity_id(entity_id)
-
-        ignored_states = (STATE_UNAVAILABLE, STATE_UNKNOWN)
-
-        handler = f"_handle_{domain}"
-
-        if hasattr(self, handler) and state.state not in ignored_states:
-            getattr(self, handler)(state)
 
         labels = self._labels(state)
         state_change = self._metric(
@@ -231,7 +225,7 @@ class PrometheusMetrics:
             prometheus_client.Gauge,
             "Entity is available (not in the unavailable or unknown state)",
         )
-        entity_available.labels(**labels).set(float(state.state not in ignored_states))
+        entity_available.labels(**labels).set(float(state.state not in IGNORED_STATES))
 
         last_updated_time_seconds = self._metric(
             "last_updated_time_seconds",
@@ -240,6 +234,18 @@ class PrometheusMetrics:
         )
         last_updated_time_seconds.labels(**labels).set(state.last_updated.timestamp())
 
+        if state.state in IGNORED_STATES:
+            self._remove_labelsets(
+                entity_id,
+                None,
+                {state_change, entity_available, last_updated_time_seconds},
+            )
+        else:
+            domain, _ = hacore.split_entity_id(entity_id)
+            handler = f"_handle_{domain}"
+            if hasattr(self, handler) and state.state:
+                getattr(self, handler)(state)
+
     def handle_entity_registry_updated(
         self, event: Event[EventEntityRegistryUpdatedData]
     ) -> None:
@@ -266,10 +272,17 @@ class PrometheusMetrics:
             self._remove_labelsets(metrics_entity_id)
 
     def _remove_labelsets(
-        self, entity_id: str, friendly_name: str | None = None
+        self,
+        entity_id: str,
+        friendly_name: str | None = None,
+        ignored_metrics: set[MetricWrapperBase] | None = None,
     ) -> None:
-        """Remove labelsets matching the given entity id from all metrics."""
+        """Remove labelsets matching the given entity id from all non-ignored metrics."""
+        if ignored_metrics is None:
+            ignored_metrics = set()
         for metric in list(self._metrics.values()):
+            if metric in ignored_metrics:
+                continue
             for sample in cast(list[prometheus_client.Metric], metric.collect())[
                 0
             ].samples:
@@ -663,7 +676,7 @@ class PrometheusMetrics:
     def _sensor_override_component_metric(
         self, state: State, unit: str | None
     ) -> str | None:
-        """Get metric from override in component confioguration."""
+        """Get metric from override in component configuration."""
         return self._component_config.get(state.entity_id).get(CONF_OVERRIDE_METRIC)
 
     @staticmethod

tests/components/prometheus/test_init.py

@@ -74,6 +74,7 @@ from homeassistant.const import (
     STATE_OPEN,
     STATE_OPENING,
     STATE_UNAVAILABLE,
+    STATE_UNKNOWN,
     UnitOfEnergy,
     UnitOfTemperature,
 )
@@ -1666,13 +1667,15 @@ async def test_disabling_entity(
 
 
 @pytest.mark.parametrize("namespace", [""])
-async def test_entity_becomes_unavailable_with_export(
+@pytest.mark.parametrize("unavailable_state", [STATE_UNAVAILABLE, STATE_UNKNOWN])
+async def test_entity_becomes_unavailable(
     hass: HomeAssistant,
     entity_registry: er.EntityRegistry,
     client: ClientSessionGenerator,
     sensor_entities: dict[str, er.RegistryEntry],
+    unavailable_state: str,
 ) -> None:
-    """Test an entity that becomes unavailable is still exported."""
+    """Test an entity that becomes unavailable/unknown is no longer exported."""
     data = {**sensor_entities}
 
     await hass.async_block_till_done()
@@ -1699,6 +1702,20 @@ async def test_entity_becomes_unavailable_with_export(
         entity="sensor.outside_temperature",
     ).withValue(1).assert_in_metrics(body)
 
+    EntityMetric(
+        metric_name="last_updated_time_seconds",
+        domain="sensor",
+        friendly_name="Outside Temperature",
+        entity="sensor.outside_temperature",
+    ).assert_in_metrics(body)
+
+    EntityMetric(
+        metric_name="battery_level_percent",
+        domain="sensor",
+        friendly_name="Outside Temperature",
+        entity="sensor.outside_temperature",
+    ).withValue(12.0).assert_in_metrics(body)
+
     EntityMetric(
         metric_name="sensor_humidity_percent",
         domain="sensor",
@@ -1720,21 +1737,28 @@ async def test_entity_becomes_unavailable_with_export(
         entity="sensor.outside_humidity",
     ).withValue(1).assert_in_metrics(body)
 
-    # Make sensor_1 unavailable.
+    # Make sensor_1 unavailable/unknown.
     set_state_with_entry(
-        hass, data["sensor_1"], STATE_UNAVAILABLE, data["sensor_1_attributes"]
+        hass, data["sensor_1"], unavailable_state, data["sensor_1_attributes"]
     )
 
     await hass.async_block_till_done()
     body = await generate_latest_metrics(client)
 
-    # Check that only the availability changed on sensor_1.
+    # Check that the availability changed on sensor_1 and the metric with the value is gone.
     EntityMetric(
         metric_name="sensor_temperature_celsius",
         domain="sensor",
         friendly_name="Outside Temperature",
         entity="sensor.outside_temperature",
-    ).withValue(15.6).assert_in_metrics(body)
+    ).assert_not_in_metrics(body)
+
+    EntityMetric(
+        metric_name="battery_level_percent",
+        domain="sensor",
+        friendly_name="Outside Temperature",
+        entity="sensor.outside_temperature",
+    ).assert_not_in_metrics(body)
 
     EntityMetric(
         metric_name="state_change_total",
@@ -1750,6 +1774,13 @@ async def test_entity_becomes_unavailable_with_export(
         entity="sensor.outside_temperature",
     ).withValue(0.0).assert_in_metrics(body)
 
+    EntityMetric(
+        metric_name="last_updated_time_seconds",
+        domain="sensor",
+        friendly_name="Outside Temperature",
+        entity="sensor.outside_temperature",
+    ).assert_in_metrics(body)
+
     # The other sensor should be unchanged.
     EntityMetric(
         metric_name="sensor_humidity_percent",
@@ -1772,8 +1803,8 @@ async def test_entity_becomes_unavailable_with_export(
         entity="sensor.outside_humidity",
     ).withValue(1).assert_in_metrics(body)
 
-    # Bring sensor_1 back and check that it is correct.
-    set_state_with_entry(hass, data["sensor_1"], 200.0, data["sensor_1_attributes"])
+    # Bring sensor_1 back and check that it returned.
+    set_state_with_entry(hass, data["sensor_1"], 201.0, data["sensor_1_attributes"])
 
     await hass.async_block_till_done()
     body = await generate_latest_metrics(client)
@@ -1783,7 +1814,14 @@ async def test_entity_becomes_unavailable_with_export(
         domain="sensor",
         friendly_name="Outside Temperature",
         entity="sensor.outside_temperature",
-    ).withValue(200.0).assert_in_metrics(body)
+    ).withValue(201.0).assert_in_metrics(body)
+
+    EntityMetric(
+        metric_name="battery_level_percent",
+        domain="sensor",
+        friendly_name="Outside Temperature",
+        entity="sensor.outside_temperature",
+    ).withValue(12.0).assert_in_metrics(body)
 
     EntityMetric(
         metric_name="state_change_total",