Write core state to /run/supervisor & monitor (#2027)

* Write core state to /run/supervisor & monitor

* Add watchdog

* Add log if they getting started

* fix stale comment

* Fix script

* avoid output

* fix tests

* fix lint

* Update rootfs/etc/services.d/watchdog/run

Co-authored-by: Franck Nijhof <git@frenck.dev>

Co-authored-by: Franck Nijhof <git@frenck.dev>

rootfs/etc/services.d/supervisor/run

@@ -1,6 +1,6 @@
 #!/usr/bin/with-contenv bashio
 # ==============================================================================
-# Start Service service
+# Start Supervisor service
 # ==============================================================================
 export LD_PRELOAD="/usr/local/lib/libjemalloc.so.2"
 

rootfs/etc/services.d/watchdog/finish

@@ -0,0 +1,8 @@
+#!/usr/bin/execlineb -S1
+# ==============================================================================
+# Take down the S6 supervision tree when Watchdog fails
+# ==============================================================================
+if { s6-test ${1} -ne 0 }
+if { s6-test ${1} -ne 256 }
+
+s6-svscanctl -t /var/run/s6/services

rootfs/etc/services.d/watchdog/run

@@ -0,0 +1,34 @@
+#!/usr/bin/with-contenv bashio
+# ==============================================================================
+# Start Watchdog service
+# ==============================================================================
+declare failed_count=0
+declare supervisor_state
+
+bashio::log.info "Start local supervisor watchdog..."
+
+while [[ failed_count -lt 2 ]];
+do
+    sleep 300
+    supervisor_state="$(cat /run/supervisor)"
+
+    if [[ "${supervisor_state}" = "running"  ]]; then
+
+        # Check API
+        if bashio::supervisor.ping; then
+            failed_count=0
+        else
+            bashio::log.warning "Maybe found an issue on API healthy"
+            ((failed_count++))
+        fi
+
+    elif [[  "close stopping" = *"${supervisor_state}"* ]]; then
+        bashio::log.warning "Maybe found an issue on shutdown"
+        ((failed_count++))
+    else
+        failed_count=0
+    fi
+
+done
+
+basio::exit.nok "Watchdog detect issue with Supervisor - take container down!"

supervisor/const.py

@@ -32,6 +32,7 @@ FILE_HASSIO_UPDATER = Path(SUPERVISOR_DATA, "updater.json")
 MACHINE_ID = Path("/etc/machine-id")
 SOCKET_DBUS = Path("/run/dbus/system_bus_socket")
 SOCKET_DOCKER = Path("/run/docker.sock")
+RUN_SUPERVISOR_STATE = Path("/run/supervisor")
 
 DOCKER_NETWORK = "hassio"
 DOCKER_NETWORK_MASK = ip_network("172.30.32.0/23")

supervisor/core.py

@@ -2,10 +2,17 @@
 import asyncio
 from contextlib import suppress
 import logging
+from typing import Optional
 
 import async_timeout
 
-from .const import SOCKET_DBUS, SUPERVISED_SUPPORTED_OS, AddonStartup, CoreState
+from .const import (
+    RUN_SUPERVISOR_STATE,
+    SOCKET_DBUS,
+    SUPERVISED_SUPPORTED_OS,
+    AddonStartup,
+    CoreState,
+)
 from .coresys import CoreSys, CoreSysAttributes
 from .exceptions import (
     DockerAPIError,
@@ -23,12 +30,31 @@ class Core(CoreSysAttributes):
     def __init__(self, coresys: CoreSys):
         """Initialize Supervisor object."""
         self.coresys: CoreSys = coresys
-        self.state: CoreState = CoreState.INITIALIZE
         self.healthy: bool = True
         self.supported: bool = True
 
+        self._state: Optional[CoreState] = None
+
+    @property
+    def state(self) -> CoreState:
+        """Return state of the core."""
+        return self._state
+
+    @state.setter
+    def state(self, new_state: CoreState) -> None:
+        """Set core into new state."""
+        try:
+            RUN_SUPERVISOR_STATE.write_text(new_state.value)
+        except OSError as err:
+            _LOGGER.warning("Can't update supervisor state %s: %s", new_state, err)
+        finally:
+            self._state = new_state
+
     async def connect(self):
         """Connect Supervisor container."""
+        self.state = CoreState.INITIALIZE
+
+        # Load information from container
         await self.sys_supervisor.load()
 
         # If host docker is supported?

tests/conftest.py

@@ -90,11 +90,15 @@ async def coresys(loop, docker, dbus, network_manager, aiohttp_client) -> CoreSy
     ):
         coresys_obj = await initialize_coresys()
 
+    # Mock save json
     coresys_obj.ingress.save_data = MagicMock()
-    coresys_obj.arch._default_arch = "amd64"
 
+    # Mock test client
+    coresys_obj.arch._default_arch = "amd64"
     coresys_obj._machine = "qemux86-64"
     coresys_obj._machine_id = uuid4()
+
+    # Mock host communication
     coresys_obj._dbus = dbus
     coresys_obj._dbus.network = network_manager
 

tests/test_core_state.py

@@ -0,0 +1,33 @@
+"""Testing handling with CoreState."""
+from pathlib import Path
+from tempfile import TemporaryDirectory
+from unittest.mock import patch
+
+import pytest
+
+from supervisor.const import CoreState
+
+# pylint: disable=redefined-outer-name
+
+
+@pytest.fixture
+def run_dir():
+    """Fixture to inject hassio env."""
+    with patch(
+        "supervisor.core.RUN_SUPERVISOR_STATE"
+    ) as mock_run, TemporaryDirectory() as tmp_run:
+        tmp_state = Path(tmp_run, "supervisor")
+        mock_run.write_text = tmp_state.write_text
+        yield tmp_state
+
+
+def test_write_state(run_dir, coresys):
+    """Test write corestate to /run/supervisor."""
+
+    coresys.core.state = CoreState.RUNNING
+
+    assert run_dir.read_text() == CoreState.RUNNING.value
+
+    coresys.core.state = CoreState.SHUTDOWN
+
+    assert run_dir.read_text() == CoreState.SHUTDOWN.value