Soft restart supervisor (#2281)

* Add softrestart to supervisor

* decouble

* adjust logger

* make sure it need run

* Use job condition

* add more job running

rootfs/etc/services.d/supervisor/finish

@@ -1,5 +1,8 @@
-#!/usr/bin/execlineb -S0
+#!/usr/bin/execlineb -S1
 # ==============================================================================
 # Take down the S6 supervision tree when Supervisor fails
 # ==============================================================================
+if { s6-test ${1} -ne 100 }
+if { s6-test ${1} -ne 256 }
+
 redirfd -w 2 /dev/null s6-svscanctl -t /var/run/s6/services

supervisor/__main__.py

@@ -60,4 +60,4 @@ if __name__ == "__main__":
         loop.close()
 
     _LOGGER.info("Closing Supervisor")
-    sys.exit(0)
+    sys.exit(coresys.core.exit_code)

supervisor/api/__init__.py

@@ -251,6 +251,7 @@ class RestAPI(CoreSysAttributes):
                 web.get("/supervisor/logs", api_supervisor.logs),
                 web.post("/supervisor/update", api_supervisor.update),
                 web.post("/supervisor/reload", api_supervisor.reload),
+                web.post("/supervisor/restart", api_supervisor.restart),
                 web.post("/supervisor/options", api_supervisor.options),
                 web.post("/supervisor/repair", api_supervisor.repair),
             ]

supervisor/api/supervisor.py

@@ -195,6 +195,11 @@ class APISupervisor(CoreSysAttributes):
         """Try to repair the local setup / overlayfs."""
         return asyncio.shield(self.sys_core.repair())
 
+    @api_process
+    def restart(self, request: web.Request) -> Awaitable[None]:
+        """Soft restart Supervisor."""
+        return asyncio.shield(self.sys_supervisor.restart())
+
     @api_process_raw(CONTENT_TYPE_BINARY)
     def logs(self, request: web.Request) -> Awaitable[bytes]:
         """Return supervisor Docker logs."""

supervisor/core.py

@@ -26,6 +26,7 @@ class Core(CoreSysAttributes):
         """Initialize Supervisor object."""
         self.coresys: CoreSys = coresys
         self._state: Optional[CoreState] = None
+        self.exit_code: int = 0
 
     @property
     def state(self) -> CoreState:
@@ -257,7 +258,7 @@ class Core(CoreSysAttributes):
             _LOGGER.warning("Stage 2: Force Shutdown!")
 
         self.state = CoreState.CLOSE
-        _LOGGER.info("Supervisor is down")
+        _LOGGER.info("Supervisor is down - %d", self.exit_code)
         self.sys_loop.stop()
 
     async def shutdown(self):

supervisor/jobs/decorator.py

@@ -20,6 +20,7 @@ class JobCondition(str, Enum):
     HEALTHY = "healthy"
     INTERNET_SYSTEM = "internet_system"
     INTERNET_HOST = "internet_host"
+    RUNNING = "running"
 
 
 class Job:
@@ -83,6 +84,14 @@ class Job:
                 )
                 return False
 
+        if JobCondition.RUNNING in self.conditions:
+            if self._coresys.core.state != CoreState.RUNNING:
+                _LOGGER.warning(
+                    "'%s' blocked from execution, system is not running",
+                    self._method.__qualname__,
+                )
+                return False
+
         if JobCondition.FREE_SPACE in self.conditions:
             free_space = self._coresys.host.info.free_space
             if free_space < MINIMUM_FREE_SPACE_THRESHOLD:

supervisor/misc/tasks.py

@@ -125,6 +125,7 @@ class Tasks(CoreSysAttributes):
             JobCondition.HEALTHY,
             JobCondition.FREE_SPACE,
             JobCondition.INTERNET_HOST,
+            JobCondition.RUNNING,
         ]
     )
     async def _update_addons(self):
@@ -150,7 +151,13 @@ class Tasks(CoreSysAttributes):
             except AddonsError:
                 _LOGGER.error("Can't auto update Add-on %s", addon.slug)
 
-    @Job(conditions=[JobCondition.FREE_SPACE, JobCondition.INTERNET_HOST])
+    @Job(
+        conditions=[
+            JobCondition.FREE_SPACE,
+            JobCondition.INTERNET_HOST,
+            JobCondition.RUNNING,
+        ]
+    )
     async def _update_supervisor(self):
         """Check and run update of Supervisor Supervisor."""
         if not self.sys_supervisor.need_update:
@@ -231,6 +238,7 @@ class Tasks(CoreSysAttributes):
         finally:
             self._cache[HASS_WATCHDOG_API] = 0
 
+    @Job(conditions=JobCondition.RUNNING)
     async def _update_cli(self):
         """Check and run update of cli."""
         if not self.sys_plugins.cli.need_update:
@@ -241,6 +249,7 @@ class Tasks(CoreSysAttributes):
         )
         await self.sys_plugins.cli.update()
 
+    @Job(conditions=JobCondition.RUNNING)
     async def _update_dns(self):
         """Check and run update of CoreDNS plugin."""
         if not self.sys_plugins.dns.need_update:
@@ -252,6 +261,7 @@ class Tasks(CoreSysAttributes):
         )
         await self.sys_plugins.dns.update()
 
+    @Job(conditions=JobCondition.RUNNING)
     async def _update_audio(self):
         """Check and run update of PulseAudio plugin."""
         if not self.sys_plugins.audio.need_update:
@@ -263,6 +273,7 @@ class Tasks(CoreSysAttributes):
         )
         await self.sys_plugins.audio.update()
 
+    @Job(conditions=JobCondition.RUNNING)
     async def _update_observer(self):
         """Check and run update of Observer plugin."""
         if not self.sys_plugins.observer.need_update:
@@ -274,6 +285,7 @@ class Tasks(CoreSysAttributes):
         )
         await self.sys_plugins.observer.update()
 
+    @Job(conditions=JobCondition.RUNNING)
     async def _update_multicast(self):
         """Check and run update of multicast."""
         if not self.sys_plugins.multicast.need_update:

supervisor/supervisor.py

@@ -11,6 +11,8 @@ import aiohttp
 from aiohttp.client_exceptions import ClientError
 from packaging.version import parse as pkg_parse
 
+from supervisor.jobs.decorator import Job, JobCondition
+
 from .const import SUPERVISOR_VERSION, URL_HASSIO_APPARMOR
 from .coresys import CoreSys, CoreSysAttributes
 from .docker.stats import DockerStats
@@ -144,6 +146,12 @@ class Supervisor(CoreSysAttributes):
             await self.update_apparmor()
         self.sys_create_task(self.sys_core.stop())
 
+    @Job(conditions=[JobCondition.RUNNING])
+    async def restart(self) -> None:
+        """Restart Supervisor soft."""
+        self.sys_core.exit_code = 100
+        self.sys_create_task(self.sys_core.stop())
+
     @property
     def in_progress(self) -> bool:
         """Return True if a task is in progress."""

tests/jobs/test_job_decorator.py

@@ -181,3 +181,27 @@ async def test_exception_not_handle(coresys: CoreSys):
 
     with pytest.raises(JobException):
         assert await test.execute()
+
+
+async def test_running(coresys: CoreSys):
+    """Test the running decorator."""
+
+    class TestClass:
+        """Test class."""
+
+        def __init__(self, coresys: CoreSys):
+            """Initialize the test class."""
+            self.coresys = coresys
+
+        @Job(conditions=[JobCondition.RUNNING])
+        async def execute(self):
+            """Execute the class method."""
+            return True
+
+    test = TestClass(coresys)
+
+    coresys.core.state = CoreState.RUNNING
+    assert await test.execute()
+
+    coresys.core.state = CoreState.FREEZE
+    assert not await test.execute()