Capture exception if image is missing on run (#4621)

* Retry run if image missing and handle fixup

* Fix lint and run error test

* Remove retry and just capture exception

supervisor/docker/addon.py

@@ -501,24 +501,16 @@ class DockerAddon(DockerInterface):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
         # Security check
         if not self.addon.protected:
             _LOGGER.warning("%s running with disabled protected mode!", self.addon.name)
 
-        # Cleanup
-        await self.stop()
-
         # Don't set a hostname if no separate UTS namespace is used
         hostname = None if self.uts_mode else self.addon.hostname
 
         # Create & Run container
         try:
-            docker_container = await self.sys_run_in_executor(
-                self.sys_docker.run,
-                self.image,
+            await self._run(
                 tag=str(self.addon.version),
                 name=self.name,
                 hostname=hostname,
@@ -549,7 +541,6 @@ class DockerAddon(DockerInterface):
             )
             raise
 
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Docker add-on %s with version %s", self.image, self.version
         )

supervisor/docker/audio.py

@@ -92,16 +92,7 @@ class DockerAudio(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=str(self.sys_plugins.audio.version),
             init=False,
             ipv4=self.sys_docker.network.audio,
@@ -118,8 +109,6 @@ class DockerAudio(DockerInterface, CoreSysAttributes):
             },
             mounts=self.mounts,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Audio %s with version %s - %s",
             self.image,

supervisor/docker/cli.py

@@ -33,16 +33,7 @@ class DockerCli(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             entrypoint=["/init"],
             tag=str(self.sys_plugins.cli.version),
             init=False,
@@ -60,8 +51,6 @@ class DockerCli(DockerInterface, CoreSysAttributes):
                 ENV_TOKEN: self.sys_plugins.cli.supervisor_token,
             },
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting CLI %s with version %s - %s",
             self.image,

supervisor/docker/dns.py

@@ -35,16 +35,7 @@ class DockerDNS(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=str(self.sys_plugins.dns.version),
             init=False,
             dns=False,
@@ -65,8 +56,6 @@ class DockerDNS(DockerInterface, CoreSysAttributes):
             ],
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting DNS %s with version %s - %s",
             self.image,

supervisor/docker/homeassistant.py

@@ -152,16 +152,7 @@ class DockerHomeAssistant(DockerInterface):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=(self.sys_homeassistant.version),
             name=self.name,
             hostname=self.name,
@@ -186,8 +177,6 @@ class DockerHomeAssistant(DockerInterface):
             tmpfs={"/tmp": ""},
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Home Assistant %s with version %s", self.image, self.version
         )

supervisor/docker/interface.py

@@ -377,6 +377,27 @@ class DockerInterface(JobGroup):
         """Run Docker image."""
         raise NotImplementedError()
 
+    async def _run(self, **kwargs) -> None:
+        """Run Docker image with retry inf necessary."""
+        if await self.is_running():
+            return
+
+        # Cleanup
+        await self.stop()
+
+        # Create & Run container
+        try:
+            docker_container = await self.sys_run_in_executor(
+                self.sys_docker.run, self.image, **kwargs
+            )
+        except DockerNotFound as err:
+            # If image is missing, capture the exception as this shouldn't happen
+            capture_exception(err)
+            raise
+
+        # Store metadata
+        self._meta = docker_container.attrs
+
     @Job(
         name="docker_interface_stop",
         limit=JobExecutionLimit.GROUP_ONCE,

supervisor/docker/multicast.py

@@ -38,16 +38,7 @@ class DockerMulticast(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=str(self.sys_plugins.multicast.version),
             init=False,
             name=self.name,
@@ -59,8 +50,6 @@ class DockerMulticast(DockerInterface, CoreSysAttributes):
             extra_hosts={"supervisor": self.sys_docker.network.supervisor},
             environment={ENV_TIME: self.sys_timezone},
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Multicast %s with version %s - Host", self.image, self.version
         )

supervisor/docker/observer.py

@@ -35,16 +35,7 @@ class DockerObserver(DockerInterface, CoreSysAttributes):
     )
     async def run(self) -> None:
         """Run Docker image."""
-        if await self.is_running():
-            return
-
-        # Cleanup
-        await self.stop()
-
-        # Create & Run container
-        docker_container = await self.sys_run_in_executor(
-            self.sys_docker.run,
-            self.image,
+        await self._run(
             tag=str(self.sys_plugins.observer.version),
             init=False,
             ipv4=self.sys_docker.network.observer,
@@ -63,8 +54,6 @@ class DockerObserver(DockerInterface, CoreSysAttributes):
             ports={"80/tcp": 4357},
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Observer %s with version %s - %s",
             self.image,

supervisor/resolution/fixups/addon_execute_repair.py

@@ -0,0 +1,57 @@
+"""Helper to fix missing image for addon."""
+
+import logging
+
+from ...coresys import CoreSys
+from ..const import ContextType, IssueType, SuggestionType
+from .base import FixupBase
+
+_LOGGER: logging.Logger = logging.getLogger(__name__)
+
+
+def setup(coresys: CoreSys) -> FixupBase:
+    """Check setup function."""
+    return FixupAddonExecuteRepair(coresys)
+
+
+class FixupAddonExecuteRepair(FixupBase):
+    """Storage class for fixup."""
+
+    async def process_fixup(self, reference: str | None = None) -> None:
+        """Pull the addons image."""
+        addon = self.sys_addons.get(reference, local_only=True)
+        if not addon:
+            _LOGGER.info(
+                "Cannot repair addon %s as it is not installed, dismissing suggestion",
+                reference,
+            )
+            return
+
+        if await addon.instance.exists():
+            _LOGGER.info(
+                "Addon %s does not need repair, dismissing suggestion", reference
+            )
+            return
+
+        _LOGGER.info("Installing image for addon %s")
+        await addon.instance.install(addon.version)
+
+    @property
+    def suggestion(self) -> SuggestionType:
+        """Return a SuggestionType enum."""
+        return SuggestionType.EXECUTE_REPAIR
+
+    @property
+    def context(self) -> ContextType:
+        """Return a ContextType enum."""
+        return ContextType.ADDON
+
+    @property
+    def issues(self) -> list[IssueType]:
+        """Return a IssueType enum list."""
+        return [IssueType.MISSING_IMAGE]
+
+    @property
+    def auto(self) -> bool:
+        """Return if a fixup can be apply as auto fix."""
+        return True

tests/docker/test_addon.py

@@ -39,13 +39,6 @@ def fixture_addonsdata_user() -> dict[str, Data]:
         yield mock
 
 
-@pytest.fixture(name="os_environ")
-def fixture_os_environ():
-    """Mock os.environ."""
-    with patch("supervisor.config.os.environ") as mock:
-        yield mock
-
-
 def get_docker_addon(
     coresys: CoreSys, addonsdata_system: dict[str, Data], config_file: str
 ):
@@ -60,7 +53,7 @@ def get_docker_addon(
 
 
 def test_base_volumes_included(
-    coresys: CoreSys, addonsdata_system: dict[str, Data], os_environ
+    coresys: CoreSys, addonsdata_system: dict[str, Data], path_extern
 ):
     """Dev and data volumes always included."""
     docker_addon = get_docker_addon(
@@ -86,7 +79,7 @@ def test_base_volumes_included(
 
 
 def test_addon_map_folder_defaults(
-    coresys: CoreSys, addonsdata_system: dict[str, Data], os_environ
+    coresys: CoreSys, addonsdata_system: dict[str, Data], path_extern
 ):
     """Validate defaults for mapped folders in addons."""
     docker_addon = get_docker_addon(
@@ -143,7 +136,7 @@ def test_addon_map_folder_defaults(
 
 
 def test_journald_addon(
-    coresys: CoreSys, addonsdata_system: dict[str, Data], os_environ
+    coresys: CoreSys, addonsdata_system: dict[str, Data], path_extern
 ):
     """Validate volume for journald option."""
     docker_addon = get_docker_addon(
@@ -171,7 +164,7 @@ def test_journald_addon(
 
 
 def test_not_journald_addon(
-    coresys: CoreSys, addonsdata_system: dict[str, Data], os_environ
+    coresys: CoreSys, addonsdata_system: dict[str, Data], path_extern
 ):
     """Validate journald option defaults off."""
     docker_addon = get_docker_addon(
@@ -182,10 +175,7 @@ def test_not_journald_addon(
 
 
 async def test_addon_run_docker_error(
-    coresys: CoreSys,
-    addonsdata_system: dict[str, Data],
-    capture_exception: Mock,
-    os_environ,
+    coresys: CoreSys, addonsdata_system: dict[str, Data], path_extern
 ):
     """Test docker error when addon is run."""
     await coresys.dbus.timedate.connect(coresys.dbus.bus)
@@ -203,14 +193,13 @@ async def test_addon_run_docker_error(
         Issue(IssueType.MISSING_IMAGE, ContextType.ADDON, reference="test_addon")
         in coresys.resolution.issues
     )
-    capture_exception.assert_not_called()
 
 
 async def test_addon_run_add_host_error(
     coresys: CoreSys,
     addonsdata_system: dict[str, Data],
     capture_exception: Mock,
-    os_environ,
+    path_extern,
 ):
     """Test error adding host when addon is run."""
     await coresys.dbus.timedate.connect(coresys.dbus.bus)

tests/docker/test_interface.py

@@ -10,12 +10,18 @@ from docker.models.images import Image
 import pytest
 from requests import RequestException
 
+from supervisor.addons import Addon
 from supervisor.const import BusEvent, CpuArch
 from supervisor.coresys import CoreSys
 from supervisor.docker.const import ContainerState
 from supervisor.docker.interface import DockerInterface
 from supervisor.docker.monitor import DockerContainerStateEvent
-from supervisor.exceptions import DockerAPIError, DockerError, DockerRequestError
+from supervisor.exceptions import (
+    DockerAPIError,
+    DockerError,
+    DockerNotFound,
+    DockerRequestError,
+)
 
 
 @pytest.fixture(autouse=True)
@@ -223,3 +229,21 @@ async def test_image_pull_fail(
         )
 
     capture_exception.assert_called_once_with(err)
+
+
+async def test_run_missing_image(
+    coresys: CoreSys,
+    install_addon_ssh: Addon,
+    container: MagicMock,
+    capture_exception: Mock,
+    path_extern,
+):
+    """Test run captures the exception when image is missing."""
+    coresys.docker.containers.create.side_effect = [NotFound("missing"), MagicMock()]
+    container.status = "stopped"
+    install_addon_ssh.data["image"] = "test_image"
+
+    with pytest.raises(DockerNotFound):
+        await install_addon_ssh.instance.run()
+
+    capture_exception.assert_called_once()

tests/resolution/fixup/test_addon_execute_repair.py

@@ -0,0 +1,73 @@
+"""Test fixup core execute repair."""
+
+from unittest.mock import MagicMock, patch
+
+from docker.errors import NotFound
+
+from supervisor.addons.addon import Addon
+from supervisor.coresys import CoreSys
+from supervisor.docker.addon import DockerAddon
+from supervisor.docker.interface import DockerInterface
+from supervisor.docker.manager import DockerAPI
+from supervisor.resolution.const import ContextType, IssueType, SuggestionType
+from supervisor.resolution.fixups.addon_execute_repair import FixupAddonExecuteRepair
+
+
+async def test_fixup(docker: DockerAPI, coresys: CoreSys, install_addon_ssh: Addon):
+    """Test fixup rebuilds addon's container."""
+    docker.images.get.side_effect = NotFound("missing")
+    install_addon_ssh.data["image"] = "test_image"
+
+    addon_execute_repair = FixupAddonExecuteRepair(coresys)
+    assert addon_execute_repair.auto is True
+
+    coresys.resolution.create_issue(
+        IssueType.MISSING_IMAGE,
+        ContextType.ADDON,
+        reference="local_ssh",
+        suggestions=[SuggestionType.EXECUTE_REPAIR],
+    )
+    with patch.object(DockerInterface, "install") as install:
+        await addon_execute_repair()
+        install.assert_called_once()
+
+    assert not coresys.resolution.issues
+    assert not coresys.resolution.suggestions
+
+
+async def test_fixup_no_addon(coresys: CoreSys):
+    """Test fixup dismisses if addon is missing."""
+    addon_execute_repair = FixupAddonExecuteRepair(coresys)
+    assert addon_execute_repair.auto is True
+
+    coresys.resolution.create_issue(
+        IssueType.MISSING_IMAGE,
+        ContextType.ADDON,
+        reference="local_ssh",
+        suggestions=[SuggestionType.EXECUTE_REPAIR],
+    )
+
+    with patch.object(DockerAddon, "install") as install:
+        await addon_execute_repair()
+        install.assert_not_called()
+
+
+async def test_fixup_image_exists(
+    docker: DockerAPI, coresys: CoreSys, install_addon_ssh: Addon
+):
+    """Test fixup dismisses if image exists."""
+    docker.images.get.return_value = MagicMock()
+
+    addon_execute_repair = FixupAddonExecuteRepair(coresys)
+    assert addon_execute_repair.auto is True
+
+    coresys.resolution.create_issue(
+        IssueType.MISSING_IMAGE,
+        ContextType.ADDON,
+        reference="local_ssh",
+        suggestions=[SuggestionType.EXECUTE_REPAIR],
+    )
+
+    with patch.object(DockerAddon, "install") as install:
+        await addon_execute_repair()
+        install.assert_not_called()