Add num open fds sensor to systemmonitor (#152441)

Co-authored-by: G Johansson <goran.johansson@shiftit.se>

homeassistant/components/systemmonitor/__init__.py

@@ -50,7 +50,10 @@ async def async_setup_entry(
     _LOGGER.debug("disk arguments to be added: %s", disk_arguments)
 
     coordinator: SystemMonitorCoordinator = SystemMonitorCoordinator(
-        hass, entry, psutil_wrapper, disk_arguments
+        hass,
+        entry,
+        psutil_wrapper,
+        disk_arguments,
     )
     await coordinator.async_config_entry_first_refresh()
     entry.runtime_data = SystemMonitorData(coordinator, psutil_wrapper)

homeassistant/components/systemmonitor/coordinator.py

@@ -8,7 +8,7 @@ import logging
 import os
 from typing import TYPE_CHECKING, Any, NamedTuple
 
-from psutil import Process
+from psutil import AccessDenied, NoSuchProcess, Process
 from psutil._common import sdiskusage, shwtemp, snetio, snicaddr, sswap
 import psutil_home_assistant as ha_psutil
 
@@ -40,6 +40,7 @@ class SensorData:
     boot_time: datetime
     processes: list[Process]
     temperatures: dict[str, list[shwtemp]]
+    process_fds: dict[str, int]
 
     def as_dict(self) -> dict[str, Any]:
         """Return as dict."""
@@ -66,6 +67,7 @@ class SensorData:
             "boot_time": str(self.boot_time),
             "processes": str(self.processes),
             "temperatures": temperatures,
+            "process_fds": str(self.process_fds),
         }
 
 
@@ -161,6 +163,7 @@ class SystemMonitorCoordinator(TimestampDataUpdateCoordinator[SensorData]):
             boot_time=_data["boot_time"],
             processes=_data["processes"],
             temperatures=_data["temperatures"],
+            process_fds=_data["process_fds"],
         )
 
     def update_data(self) -> dict[str, Any]:
@@ -233,6 +236,28 @@ class SystemMonitorCoordinator(TimestampDataUpdateCoordinator[SensorData]):
                         )
                     continue
 
+        # Collect file descriptor counts only for selected processes
+        process_fds: dict[str, int] = {}
+        for proc in selected_processes:
+            try:
+                process_name = proc.name()
+                # Our sensors are a per-process name aggregation. Not ideal, but the only
+                # way to do it without user specifying PIDs which are not static.
+                process_fds[process_name] = (
+                    process_fds.get(process_name, 0) + proc.num_fds()
+                )
+            except (NoSuchProcess, AccessDenied):
+                _LOGGER.warning(
+                    "Failed to get file descriptor count for process %s: access denied or process not found",
+                    proc.pid,
+                )
+            except OSError as err:
+                _LOGGER.warning(
+                    "OS error getting file descriptor count for process %s: %s",
+                    proc.pid,
+                    err,
+                )
+
         temps: dict[str, list[shwtemp]] = {}
         if self.update_subscribers[("temperatures", "")] or self._initial_update:
             try:
@@ -250,4 +275,5 @@ class SystemMonitorCoordinator(TimestampDataUpdateCoordinator[SensorData]):
             "boot_time": self.boot_time,
             "processes": selected_processes,
             "temperatures": temps,
+            "process_fds": process_fds,
         }

homeassistant/components/systemmonitor/sensor.py

@@ -37,7 +37,8 @@ from homeassistant.helpers.update_coordinator import CoordinatorEntity
 from homeassistant.util import slugify
 
 from . import SystemMonitorConfigEntry
-from .const import DOMAIN, NET_IO_TYPES
+from .binary_sensor import BINARY_SENSOR_DOMAIN
+from .const import CONF_PROCESS, DOMAIN, NET_IO_TYPES
 from .coordinator import SystemMonitorCoordinator
 from .util import get_all_disk_mounts, get_all_network_interfaces, read_cpu_temperature
 
@@ -125,6 +126,12 @@ def get_ip_address(
     return None
 
 
+def get_process_num_fds(entity: SystemMonitorSensor) -> int | None:
+    """Return the number of file descriptors opened by the process."""
+    process_fds = entity.coordinator.data.process_fds
+    return process_fds.get(entity.argument)
+
+
 @dataclass(frozen=True, kw_only=True)
 class SysMonitorSensorEntityDescription(SensorEntityDescription):
     """Describes System Monitor sensor entities."""
@@ -376,6 +383,16 @@ SENSOR_TYPES: dict[str, SysMonitorSensorEntityDescription] = {
         value_fn=lambda entity: entity.coordinator.data.swap.percent,
         add_to_update=lambda entity: ("swap", ""),
     ),
+    "process_num_fds": SysMonitorSensorEntityDescription(
+        key="process_num_fds",
+        translation_key="process_num_fds",
+        placeholder="process",
+        state_class=SensorStateClass.MEASUREMENT,
+        entity_registry_enabled_default=False,
+        mandatory_arg=True,
+        value_fn=get_process_num_fds,
+        add_to_update=lambda entity: ("processes", ""),
+    ),
 }
 
 
@@ -482,6 +499,38 @@ async def async_setup_entry(
             )
             continue
 
+        if _type == "process_num_fds":
+            # Create sensors for processes configured in binary_sensor section
+            processes = entry.options.get(BINARY_SENSOR_DOMAIN, {}).get(
+                CONF_PROCESS, []
+            )
+            _LOGGER.debug(
+                "Creating process_num_fds sensors for processes: %s", processes
+            )
+            for process in processes:
+                argument = process
+                is_enabled = check_legacy_resource(
+                    f"{_type}_{argument}", legacy_resources
+                )
+                unique_id = slugify(f"{_type}_{argument}")
+                loaded_resources.add(unique_id)
+                _LOGGER.debug(
+                    "Creating process_num_fds sensor: type=%s, process=%s, unique_id=%s, enabled=%s",
+                    _type,
+                    process,
+                    unique_id,
+                    is_enabled,
+                )
+                entities.append(
+                    SystemMonitorSensor(
+                        coordinator,
+                        sensor_description,
+                        entry.entry_id,
+                        argument,
+                        is_enabled,
+                    )
+                )
+            continue
     # Ensure legacy imported disk_* resources are loaded if they are not part
     # of mount points automatically discovered
     for resource in legacy_resources:

homeassistant/components/systemmonitor/strings.json

@@ -100,6 +100,9 @@
       },
       "swap_use_percent": {
         "name": "Swap usage"
+      },
+      "process_num_fds": {
+        "name": "Open file descriptors {process}"
       }
     }
   }

tests/components/systemmonitor/conftest.py

@@ -27,12 +27,20 @@ def mock_sys_platform() -> Generator[None]:
 class MockProcess(Process):
     """Mock a Process class."""
 
-    def __init__(self, name: str, ex: bool = False) -> None:
+    def __init__(
+        self,
+        name: str,
+        ex: bool = False,
+        num_fds: int | None = None,
+        raise_os_error: bool = False,
+    ) -> None:
         """Initialize the process."""
         super().__init__(1)
         self._name = name
         self._ex = ex
         self._create_time = 1708700400
+        self._num_fds = num_fds
+        self._raise_os_error = raise_os_error
 
     def name(self):
         """Return a name."""
@@ -40,6 +48,25 @@ class MockProcess(Process):
             raise NoSuchProcess(1, self._name)
         return self._name
 
+    def num_fds(self):
+        """Return the number of file descriptors opened by this process."""
+        if self._ex:
+            raise NoSuchProcess(1, self._name)
+
+        if self._raise_os_error:
+            raise OSError("Permission denied")
+
+        # Use explicit num_fds if provided, otherwise use defaults
+        if self._num_fds is not None:
+            return self._num_fds
+
+        # Return different values for different processes for testing
+        if self._name == "python3":
+            return 42
+        if self._name == "pip":
+            return 15
+        return 10
+
 
 @pytest.fixture
 def mock_setup_entry() -> Generator[AsyncMock]:

tests/components/systemmonitor/snapshots/test_diagnostics.ambr

@@ -22,6 +22,7 @@
         }),
         'load': '(1, 2, 3)',
         'memory': 'VirtualMemory(total=104857600, available=41943040, percent=40.0, used=62914560, free=31457280)',
+        'process_fds': "{'python3': 42, 'pip': 15}",
         'processes': "[tests.components.systemmonitor.conftest.MockProcess(pid=1, name='python3', status='sleeping', started='2024-02-23 15:00:00'), tests.components.systemmonitor.conftest.MockProcess(pid=1, name='pip', status='sleeping', started='2024-02-23 15:00:00')]",
         'swap': 'sswap(total=104857600, used=62914560, free=41943040, percent=60.0, sin=1, sout=1)',
         'temperatures': dict({
@@ -79,6 +80,7 @@
         'io_counters': None,
         'load': '(1, 2, 3)',
         'memory': 'VirtualMemory(total=104857600, available=41943040, percent=40.0, used=62914560, free=31457280)',
+        'process_fds': "{'python3': 42, 'pip': 15}",
         'processes': "[tests.components.systemmonitor.conftest.MockProcess(pid=1, name='python3', status='sleeping', started='2024-02-23 15:00:00'), tests.components.systemmonitor.conftest.MockProcess(pid=1, name='pip', status='sleeping', started='2024-02-23 15:00:00')]",
         'swap': 'sswap(total=104857600, used=62914560, free=41943040, percent=60.0, sin=1, sout=1)',
         'temperatures': dict({

tests/components/systemmonitor/snapshots/test_sensor.ambr

@@ -114,16 +114,6 @@
 # name: test_sensor[System Monitor Last boot - state]
   '2024-02-24T15:00:00+00:00'
 # ---
-# name: test_sensor[System Monitor Load (15 min) - attributes]
-  ReadOnlyDict({
-    'friendly_name': 'System Monitor Load (15 min)',
-    'icon': 'mdi:cpu-64-bit',
-    'state_class': <SensorStateClass.MEASUREMENT: 'measurement'>,
-  })
-# ---
-# name: test_sensor[System Monitor Load (15 min) - state]
-  '3'
-# ---
 # name: test_sensor[System Monitor Load (1 min) - attributes]
   ReadOnlyDict({
     'friendly_name': 'System Monitor Load (1 min)',
@@ -134,6 +124,16 @@
 # name: test_sensor[System Monitor Load (1 min) - state]
   '1'
 # ---
+# name: test_sensor[System Monitor Load (15 min) - attributes]
+  ReadOnlyDict({
+    'friendly_name': 'System Monitor Load (15 min)',
+    'icon': 'mdi:cpu-64-bit',
+    'state_class': <SensorStateClass.MEASUREMENT: 'measurement'>,
+  })
+# ---
+# name: test_sensor[System Monitor Load (15 min) - state]
+  '3'
+# ---
 # name: test_sensor[System Monitor Load (5 min) - attributes]
   ReadOnlyDict({
     'friendly_name': 'System Monitor Load (5 min)',
@@ -264,6 +264,24 @@
 # name: test_sensor[System Monitor Network throughput out eth1 - state]
   'unknown'
 # ---
+# name: test_sensor[System Monitor Open file descriptors pip - attributes]
+  ReadOnlyDict({
+    'friendly_name': 'System Monitor Open file descriptors pip',
+    'state_class': <SensorStateClass.MEASUREMENT: 'measurement'>,
+  })
+# ---
+# name: test_sensor[System Monitor Open file descriptors pip - state]
+  '15'
+# ---
+# name: test_sensor[System Monitor Open file descriptors python3 - attributes]
+  ReadOnlyDict({
+    'friendly_name': 'System Monitor Open file descriptors python3',
+    'state_class': <SensorStateClass.MEASUREMENT: 'measurement'>,
+  })
+# ---
+# name: test_sensor[System Monitor Open file descriptors python3 - state]
+  '42'
+# ---
 # name: test_sensor[System Monitor Packets in eth0 - attributes]
   ReadOnlyDict({
     'friendly_name': 'System Monitor Packets in eth0',

tests/components/systemmonitor/test_sensor.py

@@ -18,6 +18,8 @@ from homeassistant.const import STATE_UNAVAILABLE, STATE_UNKNOWN
 from homeassistant.core import HomeAssistant
 from homeassistant.helpers import entity_registry as er
 
+from .conftest import MockProcess
+
 from tests.common import MockConfigEntry, async_fire_time_changed
 
 
@@ -420,6 +422,107 @@ async def test_cpu_percentage_is_zero_returns_unknown(
     assert cpu_sensor.state == "15"
 
 
+@pytest.mark.usefixtures("entity_registry_enabled_by_default")
+async def test_python3_num_fds(
+    hass: HomeAssistant,
+    mock_psutil: Mock,
+    mock_os: Mock,
+    entity_registry: er.EntityRegistry,
+    snapshot: SnapshotAssertion,
+    freezer: FrozenDateTimeFactory,
+) -> None:
+    """Test python3 open file descriptors sensor."""
+    mock_config_entry = MockConfigEntry(
+        title="System Monitor",
+        domain=DOMAIN,
+        data={},
+        options={
+            "binary_sensor": {"process": ["python3", "pip"]},
+            "resources": [
+                "disk_use_percent_/",
+                "disk_use_percent_/home/notexist/",
+                "memory_free_",
+                "network_out_eth0",
+                "process_num_fds_python3",
+            ],
+        },
+    )
+    mock_config_entry.add_to_hass(hass)
+    await hass.config_entries.async_setup(mock_config_entry.entry_id)
+    await hass.async_block_till_done()
+
+    num_fds_sensor = hass.states.get(
+        "sensor.system_monitor_open_file_descriptors_python3"
+    )
+    assert num_fds_sensor is not None
+    assert num_fds_sensor.state == "42"
+    assert num_fds_sensor.attributes == {
+        "state_class": "measurement",
+        "friendly_name": "System Monitor Open file descriptors python3",
+    }
+
+    _process = MockProcess("python3", num_fds=5)
+    assert _process.num_fds() == 5
+    mock_psutil.process_iter.return_value = [_process]
+
+    freezer.tick(timedelta(minutes=1))
+    async_fire_time_changed(hass)
+    await hass.async_block_till_done(wait_background_tasks=True)
+
+    num_fds_sensor = hass.states.get(
+        "sensor.system_monitor_open_file_descriptors_python3"
+    )
+    assert num_fds_sensor is not None
+    assert num_fds_sensor.state == "5"
+
+
+@pytest.mark.usefixtures("entity_registry_enabled_by_default")
+async def test_python3_num_fds_os_error(
+    hass: HomeAssistant,
+    mock_psutil: Mock,
+    mock_os: Mock,
+    freezer: FrozenDateTimeFactory,
+    caplog: pytest.LogCaptureFixture,
+) -> None:
+    """Test python3 open file descriptors sensor handles OSError gracefully."""
+    mock_config_entry = MockConfigEntry(
+        title="System Monitor",
+        domain=DOMAIN,
+        data={},
+        options={
+            "binary_sensor": {"process": ["python3", "pip"]},
+            "resources": [
+                "process_num_fds_python3",
+            ],
+        },
+    )
+    mock_config_entry.add_to_hass(hass)
+    await hass.config_entries.async_setup(mock_config_entry.entry_id)
+    await hass.async_block_till_done()
+
+    num_fds_sensor = hass.states.get(
+        "sensor.system_monitor_open_file_descriptors_python3"
+    )
+    assert num_fds_sensor is not None
+    assert num_fds_sensor.state == "42"
+
+    _process = MockProcess("python3", raise_os_error=True)
+    mock_psutil.process_iter.return_value = [_process]
+
+    freezer.tick(timedelta(minutes=1))
+    async_fire_time_changed(hass)
+    await hass.async_block_till_done(wait_background_tasks=True)
+
+    # Sensor should still exist but have no data (unavailable or previous value)
+    num_fds_sensor = hass.states.get(
+        "sensor.system_monitor_open_file_descriptors_python3"
+    )
+    assert num_fds_sensor is not None
+    assert num_fds_sensor.state == STATE_UNKNOWN
+    # Check that warning was logged
+    assert "OS error getting file descriptor count for process 1" in caplog.text
+
+
 async def test_remove_obsolete_entities(
     hass: HomeAssistant,
     mock_psutil: Mock,
@@ -440,7 +543,7 @@ async def test_remove_obsolete_entities(
                 mock_added_config_entry.entry_id
             )
         )
-        == 37
+        == 39
     )
 
     entity_registry.async_update_entity(
@@ -481,7 +584,7 @@ async def test_remove_obsolete_entities(
                 mock_added_config_entry.entry_id
             )
         )
-        == 38
+        == 40
     )
 
     assert (