Fix docker ratelimit exception and tests

Bumps [actions/checkout](https://github.com/actions/checkout) from 5.0.0 to 5.0.1.
- [Release notes](https://github.com/actions/checkout/releases)
- [Changelog](https://github.com/actions/checkout/blob/main/CHANGELOG.md)
- [Commits](08c6903cd8...93cb6efe18)

---
updated-dependencies:
- dependency-name: actions/checkout
  dependency-version: 5.0.1
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

.github/workflows/builder.yml

@@ -53,7 +53,7 @@ jobs:
       requirements: ${{ steps.requirements.outputs.changed }}
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
         with:
           fetch-depth: 0
 
@@ -92,7 +92,7 @@ jobs:
         arch: ${{ fromJson(needs.init.outputs.architectures) }}
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
         with:
           fetch-depth: 0
 
@@ -178,7 +178,7 @@ jobs:
     steps:
       - name: Checkout the repository
         if: needs.init.outputs.publish == 'true'
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
 
       - name: Initialize git
         if: needs.init.outputs.publish == 'true'
@@ -203,7 +203,7 @@ jobs:
     timeout-minutes: 60
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
 
       # home-assistant/builder doesn't support sha pinning
       - name: Build the Supervisor

.github/workflows/ci.yaml

@@ -26,7 +26,7 @@ jobs:
     name: Prepare Python dependencies
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python
         id: python
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
@@ -68,7 +68,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         id: python
@@ -111,7 +111,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         id: python
@@ -154,7 +154,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Register hadolint problem matcher
         run: |
           echo "::add-matcher::.github/workflows/matchers/hadolint.json"
@@ -169,7 +169,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         id: python
@@ -213,7 +213,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         id: python
@@ -257,7 +257,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         id: python
@@ -293,7 +293,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         id: python
@@ -339,7 +339,7 @@ jobs:
     name: Run tests Python ${{ needs.prepare.outputs.python-version }}
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         id: python
@@ -398,7 +398,7 @@ jobs:
     needs: ["pytest", "prepare"]
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
         uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
         id: python

.github/workflows/release-drafter.yml

@@ -11,7 +11,7 @@ jobs:
     name: Release Drafter
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
         with:
           fetch-depth: 0
 

.github/workflows/sentry.yaml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Sentry Release
         uses: getsentry/action-release@128c5058bbbe93c8e02147fe0a9c713f166259a6 # v3.4.0
         env:

.github/workflows/update_frontend.yml

@@ -14,7 +14,7 @@ jobs:
       latest_version: ${{ steps.latest_frontend_version.outputs.latest_tag }}
     steps:
       - name: Checkout code
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Get latest frontend release
         id: latest_frontend_version
         uses: abatilo/release-info-action@32cb932219f1cee3fc4f4a298fd65ead5d35b661 # v1.3.3
@@ -49,7 +49,7 @@ jobs:
     if: needs.check-version.outputs.skip != 'true'
     steps:
       - name: Checkout code
-        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8 # v5.0.0
+        uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd # v5.0.1
       - name: Clear www folder
         run: |
           rm -rf supervisor/api/panel/*

requirements.txt

@@ -1,4 +1,5 @@
 aiodns==3.5.0
+aiodocker==0.24.0
 aiohttp==3.13.2
 atomicwrites-homeassistant==1.4.1
 attrs==25.4.0
@@ -24,7 +25,7 @@ pyudev==0.24.4
 PyYAML==6.0.3
 requests==2.32.5
 securetar==2025.2.1
-sentry-sdk==2.43.0
+sentry-sdk==2.44.0
 setuptools==80.9.0
 voluptuous==0.15.2
 dbus-fast==2.45.1

requirements_tests.txt

@@ -2,13 +2,13 @@ astroid==4.0.2
 coverage==7.11.3
 mypy==1.18.2
 pre-commit==4.4.0
-pylint==4.0.2
+pylint==4.0.3
 pytest-aiohttp==1.1.0
-pytest-asyncio==1.2.0
+pytest-asyncio==1.3.0
 pytest-cov==7.0.0
 pytest-timeout==2.4.0
-pytest==8.4.2
-ruff==0.14.4
+pytest==9.0.1
+ruff==0.14.5
 time-machine==2.19.0
 types-docker==7.1.0.20251009
 types-pyyaml==6.0.12.20250915

supervisor/addons/addon.py

@@ -66,13 +66,21 @@ from ..docker.const import ContainerState
 from ..docker.monitor import DockerContainerStateEvent
 from ..docker.stats import DockerStats
 from ..exceptions import (
-    AddonConfigurationError,
+    AddonBackupMetadataInvalidError,
+    AddonConfigurationInvalidError,
+    AddonNotRunningError,
     AddonNotSupportedError,
+    AddonNotSupportedWriteStdinError,
+    AddonPrePostBackupCommandReturnedError,
     AddonsError,
     AddonsJobError,
+    AddonUnknownError,
+    BackupRestoreUnknownError,
     ConfigurationFileError,
+    DockerBuildError,
     DockerError,
     HostAppArmorError,
+    StoreAddonNotFoundError,
 )
 from ..hardware.data import Device
 from ..homeassistant.const import WSEvent
@@ -235,7 +243,7 @@ class Addon(AddonModel):
             await self.instance.check_image(self.version, default_image, self.arch)
         except DockerError:
             _LOGGER.info("No %s addon Docker image %s found", self.slug, self.image)
-            with suppress(DockerError):
+            with suppress(DockerError, AddonNotSupportedError):
                 await self.instance.install(self.version, default_image, arch=self.arch)
 
         self.persist[ATTR_IMAGE] = default_image
@@ -718,18 +726,16 @@ class Addon(AddonModel):
             options = self.schema.validate(self.options)
             await self.sys_run_in_executor(write_json_file, self.path_options, options)
         except vol.Invalid as ex:
-            _LOGGER.error(
-                "Add-on %s has invalid options: %s",
-                self.slug,
-                humanize_error(self.options, ex),
-            )
-        except ConfigurationFileError:
+            raise AddonConfigurationInvalidError(
+                _LOGGER.error,
+                addon=self.slug,
+                validation_error=humanize_error(self.options, ex),
+            ) from None
+        except ConfigurationFileError as err:
             _LOGGER.error("Add-on %s can't write options", self.slug)
-        else:
-            _LOGGER.debug("Add-on %s write options: %s", self.slug, options)
-            return
+            raise AddonUnknownError(addon=self.slug) from err
 
-        raise AddonConfigurationError()
+        _LOGGER.debug("Add-on %s write options: %s", self.slug, options)
 
     @Job(
         name="addon_unload",
@@ -772,7 +778,7 @@ class Addon(AddonModel):
     async def install(self) -> None:
         """Install and setup this addon."""
         if not self.addon_store:
-            raise AddonsError("Missing from store, cannot install!")
+            raise StoreAddonNotFoundError(addon=self.slug)
 
         await self.sys_addons.data.install(self.addon_store)
 
@@ -793,9 +799,17 @@ class Addon(AddonModel):
             await self.instance.install(
                 self.latest_version, self.addon_store.image, arch=self.arch
             )
-        except DockerError as err:
+        except AddonsError:
             await self.sys_addons.data.uninstall(self)
-            raise AddonsError() from err
+            raise
+        except DockerBuildError as err:
+            _LOGGER.error("Could not build image for addon %s: %s", self.slug, err)
+            await self.sys_addons.data.uninstall(self)
+            raise AddonUnknownError(addon=self.slug) from err
+        except DockerError as err:
+            _LOGGER.error("Could not pull image to update addon %s: %s", self.slug, err)
+            await self.sys_addons.data.uninstall(self)
+            raise AddonUnknownError(addon=self.slug) from err
 
         # Finish initialization and set up listeners
         await self.load()
@@ -819,7 +833,8 @@ class Addon(AddonModel):
         try:
             await self.instance.remove(remove_image=remove_image)
         except DockerError as err:
-            raise AddonsError() from err
+            _LOGGER.error("Could not remove image for addon %s: %s", self.slug, err)
+            raise AddonUnknownError(addon=self.slug) from err
 
         self.state = AddonState.UNKNOWN
 
@@ -884,7 +899,7 @@ class Addon(AddonModel):
         if it was running. Else nothing is returned.
         """
         if not self.addon_store:
-            raise AddonsError("Missing from store, cannot update!")
+            raise StoreAddonNotFoundError(addon=self.slug)
 
         old_image = self.image
         # Cache data to prevent races with other updates to global
@@ -892,8 +907,12 @@ class Addon(AddonModel):
 
         try:
             await self.instance.update(store.version, store.image, arch=self.arch)
+        except DockerBuildError as err:
+            _LOGGER.error("Could not build image for addon %s: %s", self.slug, err)
+            raise AddonUnknownError(addon=self.slug) from err
         except DockerError as err:
-            raise AddonsError() from err
+            _LOGGER.error("Could not pull image to update addon %s: %s", self.slug, err)
+            raise AddonUnknownError(addon=self.slug) from err
 
         # Stop the addon if running
         if (last_state := self.state) in {AddonState.STARTED, AddonState.STARTUP}:
@@ -935,12 +954,23 @@ class Addon(AddonModel):
         """
         last_state: AddonState = self.state
         try:
-            # remove docker container but not addon config
+            # remove docker container and image but not addon config
             try:
                 await self.instance.remove()
-                await self.instance.install(self.version)
             except DockerError as err:
-                raise AddonsError() from err
+                _LOGGER.error("Could not remove image for addon %s: %s", self.slug, err)
+                raise AddonUnknownError(addon=self.slug) from err
+
+            try:
+                await self.instance.install(self.version)
+            except DockerBuildError as err:
+                _LOGGER.error("Could not build image for addon %s: %s", self.slug, err)
+                raise AddonUnknownError(addon=self.slug) from err
+            except DockerError as err:
+                _LOGGER.error(
+                    "Could not pull image to update addon %s: %s", self.slug, err
+                )
+                raise AddonUnknownError(addon=self.slug) from err
 
             if self.addon_store:
                 await self.sys_addons.data.update(self.addon_store)
@@ -1111,8 +1141,9 @@ class Addon(AddonModel):
         try:
             await self.instance.run()
         except DockerError as err:
+            _LOGGER.error("Could not start container for addon %s: %s", self.slug, err)
             self.state = AddonState.ERROR
-            raise AddonsError() from err
+            raise AddonUnknownError(addon=self.slug) from err
 
         return self.sys_create_task(self._wait_for_startup())
 
@@ -1127,8 +1158,9 @@ class Addon(AddonModel):
         try:
             await self.instance.stop()
         except DockerError as err:
+            _LOGGER.error("Could not stop container for addon %s: %s", self.slug, err)
             self.state = AddonState.ERROR
-            raise AddonsError() from err
+            raise AddonUnknownError(addon=self.slug) from err
 
     @Job(
         name="addon_restart",
@@ -1161,9 +1193,15 @@ class Addon(AddonModel):
     async def stats(self) -> DockerStats:
         """Return stats of container."""
         try:
+            if not await self.is_running():
+                raise AddonNotRunningError(_LOGGER.warning, addon=self.slug)
+
             return await self.instance.stats()
         except DockerError as err:
-            raise AddonsError() from err
+            _LOGGER.error(
+                "Could not get stats of container for addon %s: %s", self.slug, err
+            )
+            raise AddonUnknownError(addon=self.slug) from err
 
     @Job(
         name="addon_write_stdin",
@@ -1173,14 +1211,18 @@ class Addon(AddonModel):
     async def write_stdin(self, data) -> None:
         """Write data to add-on stdin."""
         if not self.with_stdin:
-            raise AddonNotSupportedError(
-                f"Add-on {self.slug} does not support writing to stdin!", _LOGGER.error
-            )
+            raise AddonNotSupportedWriteStdinError(_LOGGER.error, addon=self.slug)
 
         try:
-            return await self.instance.write_stdin(data)
+            if not await self.is_running():
+                raise AddonNotRunningError(_LOGGER.warning, addon=self.slug)
+
+            await self.instance.write_stdin(data)
         except DockerError as err:
-            raise AddonsError() from err
+            _LOGGER.error(
+                "Could not write stdin to container for addon %s: %s", self.slug, err
+            )
+            raise AddonUnknownError(addon=self.slug) from err
 
     async def _backup_command(self, command: str) -> None:
         try:
@@ -1189,15 +1231,14 @@ class Addon(AddonModel):
                 _LOGGER.debug(
                     "Pre-/Post backup command failed with: %s", command_return.output
                 )
-                raise AddonsError(
-                    f"Pre-/Post backup command returned error code: {command_return.exit_code}",
-                    _LOGGER.error,
+                raise AddonPrePostBackupCommandReturnedError(
+                    _LOGGER.error, addon=self.slug, exit_code=command_return.exit_code
                 )
         except DockerError as err:
-            raise AddonsError(
-                f"Failed running pre-/post backup command {command}: {str(err)}",
-                _LOGGER.error,
-            ) from err
+            _LOGGER.error(
+                "Failed running pre-/post backup command %s: %s", command, err
+            )
+            raise AddonUnknownError(addon=self.slug) from err
 
     @Job(
         name="addon_begin_backup",
@@ -1286,15 +1327,14 @@ class Addon(AddonModel):
                     try:
                         self.instance.export_image(temp_path.joinpath("image.tar"))
                     except DockerError as err:
-                        raise AddonsError() from err
+                        raise BackupRestoreUnknownError() from err
 
                 # Store local configs/state
                 try:
                     write_json_file(temp_path.joinpath("addon.json"), metadata)
                 except ConfigurationFileError as err:
-                    raise AddonsError(
-                        f"Can't save meta for {self.slug}", _LOGGER.error
-                    ) from err
+                    _LOGGER.error("Can't save meta for %s: %s", self.slug, err)
+                    raise BackupRestoreUnknownError() from err
 
                 # Store AppArmor Profile
                 if apparmor_profile:
@@ -1304,9 +1344,7 @@ class Addon(AddonModel):
                             apparmor_profile, profile_backup_file
                         )
                     except HostAppArmorError as err:
-                        raise AddonsError(
-                            "Can't backup AppArmor profile", _LOGGER.error
-                        ) from err
+                        raise BackupRestoreUnknownError() from err
 
                 # Write tarfile
                 with tar_file as backup:
@@ -1360,7 +1398,8 @@ class Addon(AddonModel):
             )
             _LOGGER.info("Finish backup for addon %s", self.slug)
         except (tarfile.TarError, OSError, AddFileError) as err:
-            raise AddonsError(f"Can't write tarfile: {err}", _LOGGER.error) from err
+            _LOGGER.error("Can't write backup tarfile for addon %s: %s", self.slug, err)
+            raise BackupRestoreUnknownError() from err
         finally:
             if was_running:
                 wait_for_start = await self.end_backup()
@@ -1402,28 +1441,24 @@ class Addon(AddonModel):
         try:
             tmp, data = await self.sys_run_in_executor(_extract_tarfile)
         except tarfile.TarError as err:
-            raise AddonsError(
-                f"Can't read tarfile {tar_file}: {err}", _LOGGER.error
-            ) from err
+            _LOGGER.error("Can't extract backup tarfile for %s: %s", self.slug, err)
+            raise BackupRestoreUnknownError() from err
         except ConfigurationFileError as err:
-            raise AddonsError() from err
+            raise AddonUnknownError(addon=self.slug) from err
 
         try:
             # Validate
             try:
                 data = SCHEMA_ADDON_BACKUP(data)
             except vol.Invalid as err:
-                raise AddonsError(
-                    f"Can't validate {self.slug}, backup data: {humanize_error(data, err)}",
+                raise AddonBackupMetadataInvalidError(
                     _LOGGER.error,
+                    addon=self.slug,
+                    validation_error=humanize_error(data, err),
                 ) from err
 
-            # If available
-            if not self._available(data[ATTR_SYSTEM]):
-                raise AddonNotSupportedError(
-                    f"Add-on {self.slug} is not available for this platform",
-                    _LOGGER.error,
-                )
+            # Validate availability. Raises if not
+            self._validate_availability(data[ATTR_SYSTEM], logger=_LOGGER.error)
 
             # Restore local add-on information
             _LOGGER.info("Restore config for addon %s", self.slug)
@@ -1482,9 +1517,10 @@ class Addon(AddonModel):
                 try:
                     await self.sys_run_in_executor(_restore_data)
                 except shutil.Error as err:
-                    raise AddonsError(
-                        f"Can't restore origin data: {err}", _LOGGER.error
-                    ) from err
+                    _LOGGER.error(
+                        "Can't restore origin data for %s: %s", self.slug, err
+                    )
+                    raise BackupRestoreUnknownError() from err
 
                 # Restore AppArmor
                 profile_file = Path(tmp.name, "apparmor.txt")
@@ -1495,10 +1531,11 @@ class Addon(AddonModel):
                         )
                     except HostAppArmorError as err:
                         _LOGGER.error(
-                            "Can't restore AppArmor profile for add-on %s",
+                            "Can't restore AppArmor profile for add-on %s: %s",
                             self.slug,
+                            err,
                         )
-                        raise AddonsError() from err
+                        raise BackupRestoreUnknownError() from err
 
             finally:
                 # Is add-on loaded

supervisor/addons/build.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 from functools import cached_property
+import logging
 from pathlib import Path
 from typing import TYPE_CHECKING, Any
 
@@ -19,13 +20,20 @@ from ..const import (
 )
 from ..coresys import CoreSys, CoreSysAttributes
 from ..docker.interface import MAP_ARCH
-from ..exceptions import ConfigurationFileError, HassioArchNotFound
+from ..exceptions import (
+    AddonBuildArchitectureNotSupportedError,
+    AddonBuildDockerfileMissingError,
+    ConfigurationFileError,
+    HassioArchNotFound,
+)
 from ..utils.common import FileConfiguration, find_one_filetype
 from .validate import SCHEMA_BUILD_CONFIG
 
 if TYPE_CHECKING:
     from .manager import AnyAddon
 
+_LOGGER: logging.Logger = logging.getLogger(__name__)
+
 
 class AddonBuild(FileConfiguration, CoreSysAttributes):
     """Handle build options for add-ons."""
@@ -106,7 +114,7 @@ class AddonBuild(FileConfiguration, CoreSysAttributes):
             return self.addon.path_location.joinpath(f"Dockerfile.{self.arch}")
         return self.addon.path_location.joinpath("Dockerfile")
 
-    async def is_valid(self) -> bool:
+    async def is_valid(self) -> None:
         """Return true if the build env is valid."""
 
         def build_is_valid() -> bool:
@@ -118,9 +126,17 @@ class AddonBuild(FileConfiguration, CoreSysAttributes):
             )
 
         try:
-            return await self.sys_run_in_executor(build_is_valid)
+            if not await self.sys_run_in_executor(build_is_valid):
+                raise AddonBuildDockerfileMissingError(
+                    _LOGGER.error, addon=self.addon.slug
+                )
         except HassioArchNotFound:
-            return False
+            raise AddonBuildArchitectureNotSupportedError(
+                _LOGGER.error,
+                addon=self.addon.slug,
+                addon_arch_list=self.addon.supported_arch,
+                system_arch_list=self.sys_arch.supported,
+            ) from None
 
     def get_docker_args(
         self, version: AwesomeVersion, image_tag: str

supervisor/api/addons.py

@@ -100,6 +100,8 @@ from ..const import (
 from ..coresys import CoreSysAttributes
 from ..docker.stats import DockerStats
 from ..exceptions import (
+    AddonBootConfigCannotChangeError,
+    AddonConfigurationInvalidError,
     APIAddonNotInstalled,
     APIError,
     APIForbidden,
@@ -125,6 +127,7 @@ SCHEMA_OPTIONS = vol.Schema(
         vol.Optional(ATTR_AUDIO_INPUT): vol.Maybe(str),
         vol.Optional(ATTR_INGRESS_PANEL): vol.Boolean(),
         vol.Optional(ATTR_WATCHDOG): vol.Boolean(),
+        vol.Optional(ATTR_OPTIONS): vol.Maybe(dict),
     }
 )
 
@@ -300,19 +303,20 @@ class APIAddons(CoreSysAttributes):
         # Update secrets for validation
         await self.sys_homeassistant.secrets.reload()
 
-        # Extend schema with add-on specific validation
-        addon_schema = SCHEMA_OPTIONS.extend(
-            {vol.Optional(ATTR_OPTIONS): vol.Maybe(addon.schema)}
-        )
-
         # Validate/Process Body
-        body = await api_validate(addon_schema, request)
+        body = await api_validate(SCHEMA_OPTIONS, request)
         if ATTR_OPTIONS in body:
-            addon.options = body[ATTR_OPTIONS]
+            try:
+                addon.options = addon.schema(body[ATTR_OPTIONS])
+            except vol.Invalid as ex:
+                raise AddonConfigurationInvalidError(
+                    addon=addon.slug,
+                    validation_error=humanize_error(body[ATTR_OPTIONS], ex),
+                ) from None
         if ATTR_BOOT in body:
             if addon.boot_config == AddonBootConfig.MANUAL_ONLY:
-                raise APIError(
-                    f"Addon {addon.slug} boot option is set to {addon.boot_config} so it cannot be changed"
+                raise AddonBootConfigCannotChangeError(
+                    addon=addon.slug, boot_config=addon.boot_config.value
                 )
             addon.boot = body[ATTR_BOOT]
         if ATTR_AUTO_UPDATE in body:

supervisor/api/store.py

@@ -53,7 +53,7 @@ from ..const import (
     REQUEST_FROM,
 )
 from ..coresys import CoreSysAttributes
-from ..exceptions import APIError, APIForbidden, APINotFound
+from ..exceptions import APIError, APIForbidden, APINotFound, StoreAddonNotFoundError
 from ..store.addon import AddonStore
 from ..store.repository import Repository
 from ..store.validate import validate_repository
@@ -104,7 +104,7 @@ class APIStore(CoreSysAttributes):
         addon_slug: str = request.match_info["addon"]
 
         if not (addon := self.sys_addons.get(addon_slug)):
-            raise APINotFound(f"Addon {addon_slug} does not exist")
+            raise StoreAddonNotFoundError(addon=addon_slug)
 
         if installed and not addon.is_installed:
             raise APIError(f"Addon {addon_slug} is not installed")
@@ -112,7 +112,7 @@ class APIStore(CoreSysAttributes):
         if not installed and addon.is_installed:
             addon = cast(Addon, addon)
             if not addon.addon_store:
-                raise APINotFound(f"Addon {addon_slug} does not exist in the store")
+                raise StoreAddonNotFoundError(addon=addon_slug)
             return addon.addon_store
 
         return addon

supervisor/auth.py

@@ -9,8 +9,10 @@ from .addons.addon import Addon
 from .const import ATTR_PASSWORD, ATTR_TYPE, ATTR_USERNAME, FILE_HASSIO_AUTH
 from .coresys import CoreSys, CoreSysAttributes
 from .exceptions import (
-    AuthError,
+    AuthHomeAssistantAPIValidationError,
+    AuthInvalidNoneValueError,
     AuthListUsersError,
+    AuthListUsersNoneResponseError,
     AuthPasswordResetError,
     HomeAssistantAPIError,
     HomeAssistantWSError,
@@ -83,10 +85,8 @@ class Auth(FileConfiguration, CoreSysAttributes):
         self, addon: Addon, username: str | None, password: str | None
     ) -> bool:
         """Check username login."""
-        if password is None:
-            raise AuthError("None as password is not supported!", _LOGGER.error)
-        if username is None:
-            raise AuthError("None as username is not supported!", _LOGGER.error)
+        if username is None or password is None:
+            raise AuthInvalidNoneValueError(_LOGGER.error)
 
         _LOGGER.info("Auth request from '%s' for '%s'", addon.slug, username)
 
@@ -137,7 +137,7 @@ class Auth(FileConfiguration, CoreSysAttributes):
         finally:
             self._running.pop(username, None)
 
-        raise AuthError()
+        raise AuthHomeAssistantAPIValidationError()
 
     async def change_password(self, username: str, password: str) -> None:
         """Change user password login."""
@@ -155,7 +155,7 @@ class Auth(FileConfiguration, CoreSysAttributes):
         except HomeAssistantAPIError as err:
             _LOGGER.error("Can't request password reset on Home Assistant: %s", err)
 
-        raise AuthPasswordResetError()
+        raise AuthPasswordResetError(user=username)
 
     async def list_users(self) -> list[dict[str, Any]]:
         """List users on the Home Assistant instance."""
@@ -166,15 +166,12 @@ class Auth(FileConfiguration, CoreSysAttributes):
                 {ATTR_TYPE: "config/auth/list"}
             )
         except HomeAssistantWSError as err:
-            raise AuthListUsersError(
-                f"Can't request listing users on Home Assistant: {err}", _LOGGER.error
-            ) from err
+            _LOGGER.error("Can't request listing users on Home Assistant: %s", err)
+            raise AuthListUsersError() from err
 
         if users is not None:
             return users
-        raise AuthListUsersError(
-            "Can't request listing users on Home Assistant!", _LOGGER.error
-        )
+        raise AuthListUsersNoneResponseError(_LOGGER.error)
 
     @staticmethod
     def _rehash(value: str, salt2: str = "") -> str:

supervisor/backups/backup.py

@@ -628,9 +628,6 @@ class Backup(JobGroup):
                 if start_task := await self._addon_save(addon):
                     start_tasks.append(start_task)
             except BackupError as err:
-                err = BackupError(
-                    f"Can't backup add-on {addon.slug}: {str(err)}", _LOGGER.error
-                )
                 self.sys_jobs.current.capture_error(err)
 
         return start_tasks

supervisor/bus.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+from asyncio import Task
 from collections.abc import Callable, Coroutine
 import logging
 from typing import Any
@@ -38,11 +39,13 @@ class Bus(CoreSysAttributes):
         self._listeners.setdefault(event, []).append(listener)
         return listener
 
-    def fire_event(self, event: BusEvent, reference: Any) -> None:
+    def fire_event(self, event: BusEvent, reference: Any) -> list[Task]:
         """Fire an event to the bus."""
         _LOGGER.debug("Fire event '%s' with '%s'", event, reference)
+        tasks: list[Task] = []
         for listener in self._listeners.get(event, []):
-            self.sys_create_task(listener.callback(reference))
+            tasks.append(self.sys_create_task(listener.callback(reference)))
+        return tasks
 
     def remove_listener(self, listener: EventListener) -> None:
         """Unregister an listener."""

supervisor/docker/addon.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+from collections.abc import Awaitable
 from contextlib import suppress
 from ipaddress import IPv4Address
 import logging
@@ -9,6 +10,7 @@ import os
 from pathlib import Path
 from typing import TYPE_CHECKING, cast
 
+import aiodocker
 from attr import evolve
 from awesomeversion import AwesomeVersion
 import docker
@@ -32,6 +34,7 @@ from ..coresys import CoreSys
 from ..exceptions import (
     CoreDNSError,
     DBusError,
+    DockerBuildError,
     DockerError,
     DockerJobError,
     DockerNotFound,
@@ -679,9 +682,8 @@ class DockerAddon(DockerInterface):
     async def _build(self, version: AwesomeVersion, image: str | None = None) -> None:
         """Build a Docker container."""
         build_env = await AddonBuild(self.coresys, self.addon).load_config()
-        if not await build_env.is_valid():
-            _LOGGER.error("Invalid build environment, can't build this add-on!")
-            raise DockerError()
+        # Check if the build environment is valid, raises if not
+        await build_env.is_valid()
 
         _LOGGER.info("Starting build for %s:%s", self.image, version)
 
@@ -717,21 +719,24 @@ class DockerAddon(DockerInterface):
                 error_message = f"Docker build failed for {addon_image_tag} (exit code {result.exit_code}). Build output:\n{logs}"
                 raise docker.errors.DockerException(error_message)
 
-            addon_image = self.sys_docker.images.get(addon_image_tag)
-
-            return addon_image, logs
+            return addon_image_tag, logs
 
         try:
-            docker_image, log = await self.sys_run_in_executor(build_image)
+            addon_image_tag, log = await self.sys_run_in_executor(build_image)
 
             _LOGGER.debug("Build %s:%s done: %s", self.image, version, log)
 
             # Update meta data
-            self._meta = docker_image.attrs
+            self._meta = await self.sys_docker.images.inspect(addon_image_tag)
 
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            _LOGGER.error("Can't build %s:%s: %s", self.image, version, err)
-            raise DockerError() from err
+        except (
+            docker.errors.DockerException,
+            requests.RequestException,
+            aiodocker.DockerError,
+        ) as err:
+            raise DockerBuildError(
+                f"Can't build {self.image}:{version}: {err!s}", _LOGGER.error
+            ) from err
 
         _LOGGER.info("Build %s:%s done", self.image, version)
 
@@ -751,11 +756,8 @@ class DockerAddon(DockerInterface):
     )
     async def import_image(self, tar_file: Path) -> None:
         """Import a tar file as image."""
-        docker_image = await self.sys_run_in_executor(
-            self.sys_docker.import_image, tar_file
-        )
-        if docker_image:
-            self._meta = docker_image.attrs
+        if docker_image := await self.sys_docker.import_image(tar_file):
+            self._meta = docker_image
             _LOGGER.info("Importing image %s and version %s", tar_file, self.version)
 
             with suppress(DockerError):
@@ -769,17 +771,21 @@ class DockerAddon(DockerInterface):
         version: AwesomeVersion | None = None,
     ) -> None:
         """Check if old version exists and cleanup other versions of image not in use."""
-        await self.sys_run_in_executor(
-            self.sys_docker.cleanup_old_images,
-            (image := image or self.image),
-            version or self.version,
+        if not (use_image := image or self.image):
+            raise DockerError("Cannot determine image from metadata!", _LOGGER.error)
+        if not (use_version := version or self.version):
+            raise DockerError("Cannot determine version from metadata!", _LOGGER.error)
+
+        await self.sys_docker.cleanup_old_images(
+            use_image,
+            use_version,
             {old_image} if old_image else None,
             keep_images={
                 f"{addon.image}:{addon.version}"
                 for addon in self.sys_addons.installed
                 if addon.slug != self.addon.slug
                 and addon.image
-                and addon.image in {old_image, image}
+                and addon.image in {old_image, use_image}
             },
         )
 
@@ -788,12 +794,9 @@ class DockerAddon(DockerInterface):
         on_condition=DockerJobError,
         concurrency=JobConcurrency.GROUP_REJECT,
     )
-    async def write_stdin(self, data: bytes) -> None:
+    def write_stdin(self, data: bytes) -> Awaitable[None]:
         """Write to add-on stdin."""
-        if not await self.is_running():
-            raise DockerError()
-
-        await self.sys_run_in_executor(self._write_stdin, data)
+        return self.sys_run_in_executor(self._write_stdin, data)
 
     def _write_stdin(self, data: bytes) -> None:
         """Write to add-on stdin.

supervisor/docker/homeassistant.py

@@ -1,6 +1,5 @@
 """Init file for Supervisor Docker object."""
 
-from collections.abc import Awaitable
 from ipaddress import IPv4Address
 import logging
 import re
@@ -236,11 +235,10 @@ class DockerHomeAssistant(DockerInterface):
             environment={ENV_TIME: self.sys_timezone},
         )
 
-    def is_initialize(self) -> Awaitable[bool]:
+    async def is_initialize(self) -> bool:
         """Return True if Docker container exists."""
-        return self.sys_run_in_executor(
-            self.sys_docker.container_is_initialized,
-            self.name,
-            self.image,
-            self.sys_homeassistant.version,
+        if not self.sys_homeassistant.version:
+            return False
+        return await self.sys_docker.container_is_initialized(
+            self.name, self.image, self.sys_homeassistant.version
         )

supervisor/docker/interface.py

@@ -6,17 +6,18 @@ from abc import ABC, abstractmethod
 from collections import defaultdict
 from collections.abc import Awaitable
 from contextlib import suppress
+from http import HTTPStatus
 import logging
 import re
 from time import time
 from typing import Any, cast
 from uuid import uuid4
 
+import aiodocker
 from awesomeversion import AwesomeVersion
 from awesomeversion.strategy import AwesomeVersionStrategy
 import docker
 from docker.models.containers import Container
-from docker.models.images import Image
 import requests
 
 from ..bus import EventListener
@@ -33,6 +34,7 @@ from ..coresys import CoreSys
 from ..exceptions import (
     DockerAPIError,
     DockerError,
+    DockerHubRateLimitExceeded,
     DockerJobError,
     DockerLogOutOfOrder,
     DockerNotFound,
@@ -215,7 +217,7 @@ class DockerInterface(JobGroup, ABC):
         if not credentials:
             return
 
-        await self.sys_run_in_executor(self.sys_docker.docker.login, **credentials)
+        await self.sys_run_in_executor(self.sys_docker.dockerpy.login, **credentials)
 
     def _process_pull_image_log(  # noqa: C901
         self, install_job_id: str, reference: PullLogEntry
@@ -418,8 +420,7 @@ class DockerInterface(JobGroup, ABC):
             )
 
             # Pull new image
-            docker_image = await self.sys_run_in_executor(
-                self.sys_docker.pull_image,
+            docker_image = await self.sys_docker.pull_image(
                 self.sys_jobs.current.uuid,
                 image,
                 str(version),
@@ -431,22 +432,37 @@ class DockerInterface(JobGroup, ABC):
                 _LOGGER.info(
                     "Tagging image %s with version %s as latest", image, version
                 )
-                await self.sys_run_in_executor(docker_image.tag, image, tag="latest")
+                await self.sys_docker.images.tag(
+                    docker_image["Id"], image, tag="latest"
+                )
         except docker.errors.APIError as err:
-            if err.status_code == 429:
+            if err.status_code == HTTPStatus.TOO_MANY_REQUESTS:
                 self.sys_resolution.create_issue(
                     IssueType.DOCKER_RATELIMIT,
                     ContextType.SYSTEM,
                     suggestions=[SuggestionType.REGISTRY_LOGIN],
                 )
-                _LOGGER.info(
-                    "Your IP address has made too many requests to Docker Hub which activated a rate limit. "
-                    "For more details see https://www.home-assistant.io/more-info/dockerhub-rate-limit"
-                )
+                raise DockerHubRateLimitExceeded(_LOGGER.error) from err
+            await async_capture_exception(err)
             raise DockerError(
                 f"Can't install {image}:{version!s}: {err}", _LOGGER.error
             ) from err
-        except (docker.errors.DockerException, requests.RequestException) as err:
+        except aiodocker.DockerError as err:
+            if err.status == HTTPStatus.TOO_MANY_REQUESTS:
+                self.sys_resolution.create_issue(
+                    IssueType.DOCKER_RATELIMIT,
+                    ContextType.SYSTEM,
+                    suggestions=[SuggestionType.REGISTRY_LOGIN],
+                )
+                raise DockerHubRateLimitExceeded(_LOGGER.error) from err
+            await async_capture_exception(err)
+            raise DockerError(
+                f"Can't install {image}:{version!s}: {err}", _LOGGER.error
+            ) from err
+        except (
+            docker.errors.DockerException,
+            requests.RequestException,
+        ) as err:
             await async_capture_exception(err)
             raise DockerError(
                 f"Unknown error with {image}:{version!s} -> {err!s}", _LOGGER.error
@@ -455,46 +471,43 @@ class DockerInterface(JobGroup, ABC):
             if listener:
                 self.sys_bus.remove_listener(listener)
 
-        self._meta = docker_image.attrs
+        self._meta = docker_image
 
     async def exists(self) -> bool:
         """Return True if Docker image exists in local repository."""
-        with suppress(docker.errors.DockerException, requests.RequestException):
-            await self.sys_run_in_executor(
-                self.sys_docker.images.get, f"{self.image}:{self.version!s}"
-            )
+        with suppress(aiodocker.DockerError, requests.RequestException):
+            await self.sys_docker.images.inspect(f"{self.image}:{self.version!s}")
             return True
         return False
 
-    async def is_running(self) -> bool:
-        """Return True if Docker is running."""
+    async def _get_container(self) -> Container | None:
+        """Get docker container, returns None if not found."""
         try:
-            docker_container = await self.sys_run_in_executor(
+            return await self.sys_run_in_executor(
                 self.sys_docker.containers.get, self.name
             )
         except docker.errors.NotFound:
-            return False
+            return None
         except docker.errors.DockerException as err:
-            raise DockerAPIError() from err
+            raise DockerAPIError(
+                f"Docker API error occurred while getting container information: {err!s}"
+            ) from err
         except requests.RequestException as err:
-            raise DockerRequestError() from err
+            raise DockerRequestError(
+                f"Error communicating with Docker to get container information: {err!s}"
+            ) from err
 
-        return docker_container.status == "running"
+    async def is_running(self) -> bool:
+        """Return True if Docker is running."""
+        if docker_container := await self._get_container():
+            return docker_container.status == "running"
+        return False
 
     async def current_state(self) -> ContainerState:
         """Return current state of container."""
-        try:
-            docker_container = await self.sys_run_in_executor(
-                self.sys_docker.containers.get, self.name
-            )
-        except docker.errors.NotFound:
-            return ContainerState.UNKNOWN
-        except docker.errors.DockerException as err:
-            raise DockerAPIError() from err
-        except requests.RequestException as err:
-            raise DockerRequestError() from err
-
-        return _container_state_from_model(docker_container)
+        if docker_container := await self._get_container():
+            return _container_state_from_model(docker_container)
+        return ContainerState.UNKNOWN
 
     @Job(name="docker_interface_attach", concurrency=JobConcurrency.GROUP_QUEUE)
     async def attach(
@@ -521,15 +534,17 @@ class DockerInterface(JobGroup, ABC):
                     ),
                 )
 
-        with suppress(docker.errors.DockerException, requests.RequestException):
+        with suppress(aiodocker.DockerError, requests.RequestException):
             if not self._meta and self.image:
-                self._meta = self.sys_docker.images.get(
+                self._meta = await self.sys_docker.images.inspect(
                     f"{self.image}:{version!s}"
-                ).attrs
+                )
 
         # Successful?
         if not self._meta:
-            raise DockerError()
+            raise DockerError(
+                f"Could not get metadata on container or image for {self.name}"
+            )
         _LOGGER.info("Attaching to %s with version %s", self.image, self.version)
 
     @Job(
@@ -593,14 +608,17 @@ class DockerInterface(JobGroup, ABC):
     )
     async def remove(self, *, remove_image: bool = True) -> None:
         """Remove Docker images."""
+        if not self.image or not self.version:
+            raise DockerError(
+                "Cannot determine image and/or version from metadata!", _LOGGER.error
+            )
+
         # Cleanup container
         with suppress(DockerError):
             await self.stop()
 
         if remove_image:
-            await self.sys_run_in_executor(
-                self.sys_docker.remove_image, self.image, self.version
-            )
+            await self.sys_docker.remove_image(self.image, self.version)
 
         self._meta = None
 
@@ -622,18 +640,16 @@ class DockerInterface(JobGroup, ABC):
         image_name = f"{expected_image}:{version!s}"
         if self.image == expected_image:
             try:
-                image: Image = await self.sys_run_in_executor(
-                    self.sys_docker.images.get, image_name
-                )
-            except (docker.errors.DockerException, requests.RequestException) as err:
+                image = await self.sys_docker.images.inspect(image_name)
+            except (aiodocker.DockerError, requests.RequestException) as err:
                 raise DockerError(
                     f"Could not get {image_name} for check due to: {err!s}",
                     _LOGGER.error,
                 ) from err
 
-            image_arch = f"{image.attrs['Os']}/{image.attrs['Architecture']}"
-            if "Variant" in image.attrs:
-                image_arch = f"{image_arch}/{image.attrs['Variant']}"
+            image_arch = f"{image['Os']}/{image['Architecture']}"
+            if "Variant" in image:
+                image_arch = f"{image_arch}/{image['Variant']}"
 
             # If we have an image and its the right arch, all set
             # It seems that newer Docker version return a variant for arm64 images.
@@ -695,11 +711,13 @@ class DockerInterface(JobGroup, ABC):
         version: AwesomeVersion | None = None,
     ) -> None:
         """Check if old version exists and cleanup."""
-        await self.sys_run_in_executor(
-            self.sys_docker.cleanup_old_images,
-            image or self.image,
-            version or self.version,
-            {old_image} if old_image else None,
+        if not (use_image := image or self.image):
+            raise DockerError("Cannot determine image from metadata!", _LOGGER.error)
+        if not (use_version := version or self.version):
+            raise DockerError("Cannot determine version from metadata!", _LOGGER.error)
+
+        await self.sys_docker.cleanup_old_images(
+            use_image, use_version, {old_image} if old_image else None
         )
 
     @Job(
@@ -731,14 +749,8 @@ class DockerInterface(JobGroup, ABC):
 
     async def is_failed(self) -> bool:
         """Return True if Docker is failing state."""
-        try:
-            docker_container = await self.sys_run_in_executor(
-                self.sys_docker.containers.get, self.name
-            )
-        except docker.errors.NotFound:
+        if not (docker_container := await self._get_container()):
             return False
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError() from err
 
         # container is not running
         if docker_container.status != "exited":
@@ -751,10 +763,10 @@ class DockerInterface(JobGroup, ABC):
         """Return latest version of local image."""
         available_version: list[AwesomeVersion] = []
         try:
-            for image in await self.sys_run_in_executor(
-                self.sys_docker.images.list, self.image
+            for image in await self.sys_docker.images.list(
+                filters=f'{{"reference": ["{self.image}"]}}'
             ):
-                for tag in image.tags:
+                for tag in image["RepoTags"]:
                     version = AwesomeVersion(tag.partition(":")[2])
                     if version.strategy == AwesomeVersionStrategy.UNKNOWN:
                         continue
@@ -763,7 +775,7 @@ class DockerInterface(JobGroup, ABC):
             if not available_version:
                 raise ValueError()
 
-        except (docker.errors.DockerException, ValueError) as err:
+        except (aiodocker.DockerError, ValueError) as err:
             raise DockerNotFound(
                 f"No version found for {self.image}", _LOGGER.info
             ) from err

supervisor/docker/manager.py

@@ -6,20 +6,24 @@ import asyncio
 from contextlib import suppress
 from dataclasses import dataclass
 from functools import partial
+from http import HTTPStatus
 from ipaddress import IPv4Address
+import json
 import logging
 import os
 from pathlib import Path
+import re
 from typing import Any, Final, Self, cast
 
+import aiodocker
+from aiodocker.images import DockerImages
+from aiohttp import ClientSession, ClientTimeout, UnixConnector
 import attr
 from awesomeversion import AwesomeVersion, AwesomeVersionCompareException
 from docker import errors as docker_errors
 from docker.api.client import APIClient
 from docker.client import DockerClient
-from docker.errors import DockerException, ImageNotFound, NotFound
 from docker.models.containers import Container, ContainerCollection
-from docker.models.images import Image, ImageCollection
 from docker.models.networks import Network
 from docker.types.daemon import CancellableStream
 import requests
@@ -53,6 +57,7 @@ _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 MIN_SUPPORTED_DOCKER: Final = AwesomeVersion("24.0.0")
 DOCKER_NETWORK_HOST: Final = "host"
+RE_IMPORT_IMAGE_STREAM = re.compile(r"(^Loaded image ID: |^Loaded image: )(.+)$")
 
 
 @attr.s(frozen=True)
@@ -204,7 +209,15 @@ class DockerAPI(CoreSysAttributes):
     def __init__(self, coresys: CoreSys):
         """Initialize Docker base wrapper."""
         self.coresys = coresys
-        self._docker: DockerClient | None = None
+        # We keep both until we can fully refactor to aiodocker
+        self._dockerpy: DockerClient | None = None
+        self.docker: aiodocker.Docker = aiodocker.Docker(
+            url="unix://localhost",  # dummy hostname for URL composition
+            connector=(connector := UnixConnector(SOCKET_DOCKER.as_posix())),
+            session=ClientSession(connector=connector, timeout=ClientTimeout(900)),
+            api_version="auto",
+        )
+
         self._network: DockerNetwork | None = None
         self._info: DockerInfo | None = None
         self.config: DockerConfig = DockerConfig()
@@ -212,28 +225,28 @@ class DockerAPI(CoreSysAttributes):
 
     async def post_init(self) -> Self:
         """Post init actions that must be done in event loop."""
-        self._docker = await asyncio.get_running_loop().run_in_executor(
+        self._dockerpy = await asyncio.get_running_loop().run_in_executor(
             None,
             partial(
                 DockerClient,
-                base_url=f"unix:/{str(SOCKET_DOCKER)}",
+                base_url=f"unix:/{SOCKET_DOCKER.as_posix()}",
                 version="auto",
                 timeout=900,
             ),
         )
-        self._info = DockerInfo.new(self.docker.info())
+        self._info = DockerInfo.new(self.dockerpy.info())
         await self.config.read_data()
-        self._network = await DockerNetwork(self.docker).post_init(
+        self._network = await DockerNetwork(self.dockerpy).post_init(
             self.config.enable_ipv6, self.config.mtu
         )
         return self
 
     @property
-    def docker(self) -> DockerClient:
+    def dockerpy(self) -> DockerClient:
         """Get docker API client."""
-        if not self._docker:
+        if not self._dockerpy:
             raise RuntimeError("Docker API Client not initialized!")
-        return self._docker
+        return self._dockerpy
 
     @property
     def network(self) -> DockerNetwork:
@@ -243,19 +256,19 @@ class DockerAPI(CoreSysAttributes):
         return self._network
 
     @property
-    def images(self) -> ImageCollection:
+    def images(self) -> DockerImages:
         """Return API images."""
         return self.docker.images
 
     @property
     def containers(self) -> ContainerCollection:
         """Return API containers."""
-        return self.docker.containers
+        return self.dockerpy.containers
 
     @property
     def api(self) -> APIClient:
         """Return API containers."""
-        return self.docker.api
+        return self.dockerpy.api
 
     @property
     def info(self) -> DockerInfo:
@@ -267,7 +280,7 @@ class DockerAPI(CoreSysAttributes):
     @property
     def events(self) -> CancellableStream:
         """Return docker event stream."""
-        return self.docker.events(decode=True)
+        return self.dockerpy.events(decode=True)
 
     @property
     def monitor(self) -> DockerMonitor:
@@ -383,7 +396,7 @@ class DockerAPI(CoreSysAttributes):
                 with suppress(DockerError):
                     self.network.detach_default_bridge(container)
         else:
-            host_network: Network = self.docker.networks.get(DOCKER_NETWORK_HOST)
+            host_network: Network = self.dockerpy.networks.get(DOCKER_NETWORK_HOST)
 
             # Check if container is register on host
             # https://github.com/moby/moby/issues/23302
@@ -410,35 +423,32 @@ class DockerAPI(CoreSysAttributes):
 
         return container
 
-    def pull_image(
+    async def pull_image(
         self,
         job_id: str,
         repository: str,
         tag: str = "latest",
         platform: str | None = None,
-    ) -> Image:
+    ) -> dict[str, Any]:
         """Pull the specified image and return it.
 
         This mimics the high level API of images.pull but provides better error handling by raising
         based on a docker error on pull. Whereas the high level API ignores all errors on pull and
         raises only if the get fails afterwards. Additionally it fires progress reports for the pull
         on the bus so listeners can use that to update status for users.
-
-        Must be run in executor.
         """
-        pull_log = self.docker.api.pull(
-            repository, tag=tag, platform=platform, stream=True, decode=True
-        )
-        for e in pull_log:
+        async for e in self.images.pull(
+            repository, tag=tag, platform=platform, stream=True
+        ):
             entry = PullLogEntry.from_pull_log_dict(job_id, e)
             if entry.error:
                 raise entry.exception
-            self.sys_loop.call_soon_threadsafe(
-                self.sys_bus.fire_event, BusEvent.DOCKER_IMAGE_PULL_UPDATE, entry
+            await asyncio.gather(
+                *self.sys_bus.fire_event(BusEvent.DOCKER_IMAGE_PULL_UPDATE, entry)
             )
 
         sep = "@" if tag.startswith("sha256:") else ":"
-        return self.images.get(f"{repository}{sep}{tag}")
+        return await self.images.inspect(f"{repository}{sep}{tag}")
 
     def run_command(
         self,
@@ -459,7 +469,7 @@ class DockerAPI(CoreSysAttributes):
         _LOGGER.info("Runing command '%s' on %s", command, image_with_tag)
         container = None
         try:
-            container = self.docker.containers.run(
+            container = self.dockerpy.containers.run(
                 image_with_tag,
                 command=command,
                 detach=True,
@@ -487,35 +497,35 @@ class DockerAPI(CoreSysAttributes):
         """Repair local docker overlayfs2 issues."""
         _LOGGER.info("Prune stale containers")
         try:
-            output = self.docker.api.prune_containers()
+            output = self.dockerpy.api.prune_containers()
             _LOGGER.debug("Containers prune: %s", output)
         except docker_errors.APIError as err:
             _LOGGER.warning("Error for containers prune: %s", err)
 
         _LOGGER.info("Prune stale images")
         try:
-            output = self.docker.api.prune_images(filters={"dangling": False})
+            output = self.dockerpy.api.prune_images(filters={"dangling": False})
             _LOGGER.debug("Images prune: %s", output)
         except docker_errors.APIError as err:
             _LOGGER.warning("Error for images prune: %s", err)
 
         _LOGGER.info("Prune stale builds")
         try:
-            output = self.docker.api.prune_builds()
+            output = self.dockerpy.api.prune_builds()
             _LOGGER.debug("Builds prune: %s", output)
         except docker_errors.APIError as err:
             _LOGGER.warning("Error for builds prune: %s", err)
 
         _LOGGER.info("Prune stale volumes")
         try:
-            output = self.docker.api.prune_builds()
+            output = self.dockerpy.api.prune_volumes()
             _LOGGER.debug("Volumes prune: %s", output)
         except docker_errors.APIError as err:
             _LOGGER.warning("Error for volumes prune: %s", err)
 
         _LOGGER.info("Prune stale networks")
         try:
-            output = self.docker.api.prune_networks()
+            output = self.dockerpy.api.prune_networks()
             _LOGGER.debug("Networks prune: %s", output)
         except docker_errors.APIError as err:
             _LOGGER.warning("Error for networks prune: %s", err)
@@ -537,11 +547,11 @@ class DockerAPI(CoreSysAttributes):
 
         Fix: https://github.com/moby/moby/issues/23302
         """
-        network: Network = self.docker.networks.get(network_name)
+        network: Network = self.dockerpy.networks.get(network_name)
 
         for cid, data in network.attrs.get("Containers", {}).items():
             try:
-                self.docker.containers.get(cid)
+                self.dockerpy.containers.get(cid)
                 continue
             except docker_errors.NotFound:
                 _LOGGER.debug(
@@ -556,22 +566,32 @@ class DockerAPI(CoreSysAttributes):
             with suppress(docker_errors.DockerException, requests.RequestException):
                 network.disconnect(data.get("Name", cid), force=True)
 
-    def container_is_initialized(
+    async def container_is_initialized(
         self, name: str, image: str, version: AwesomeVersion
     ) -> bool:
         """Return True if docker container exists in good state and is built from expected image."""
         try:
-            docker_container = self.containers.get(name)
-            docker_image = self.images.get(f"{image}:{version}")
-        except NotFound:
+            docker_container = await self.sys_run_in_executor(self.containers.get, name)
+            docker_image = await self.images.inspect(f"{image}:{version}")
+        except docker_errors.NotFound:
             return False
-        except (DockerException, requests.RequestException) as err:
-            raise DockerError() from err
+        except aiodocker.DockerError as err:
+            if err.status == HTTPStatus.NOT_FOUND:
+                return False
+            raise DockerError(
+                f"Could not get container {name} or image {image}:{version} to check state: {err!s}",
+                _LOGGER.error,
+            ) from err
+        except (docker_errors.DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Could not get container {name} or image {image}:{version} to check state: {err!s}",
+                _LOGGER.error,
+            ) from err
 
         # Check the image is correct and state is good
         return (
             docker_container.image is not None
-            and docker_container.image.id == docker_image.id
+            and docker_container.image.id == docker_image["Id"]
             and docker_container.status in ("exited", "running", "created")
         )
 
@@ -581,18 +601,22 @@ class DockerAPI(CoreSysAttributes):
         """Stop/remove Docker container."""
         try:
             docker_container: Container = self.containers.get(name)
-        except NotFound:
+        except docker_errors.NotFound:
+            # Generally suppressed so we don't log this
             raise DockerNotFound() from None
-        except (DockerException, requests.RequestException) as err:
-            raise DockerError() from err
+        except (docker_errors.DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Could not get container {name} for stopping: {err!s}",
+                _LOGGER.error,
+            ) from err
 
         if docker_container.status == "running":
             _LOGGER.info("Stopping %s application", name)
-            with suppress(DockerException, requests.RequestException):
+            with suppress(docker_errors.DockerException, requests.RequestException):
                 docker_container.stop(timeout=timeout)
 
         if remove_container:
-            with suppress(DockerException, requests.RequestException):
+            with suppress(docker_errors.DockerException, requests.RequestException):
                 _LOGGER.info("Cleaning %s application", name)
                 docker_container.remove(force=True, v=True)
 
@@ -604,11 +628,11 @@ class DockerAPI(CoreSysAttributes):
         """Start Docker container."""
         try:
             docker_container: Container = self.containers.get(name)
-        except NotFound:
+        except docker_errors.NotFound:
             raise DockerNotFound(
                 f"{name} not found for starting up", _LOGGER.error
             ) from None
-        except (DockerException, requests.RequestException) as err:
+        except (docker_errors.DockerException, requests.RequestException) as err:
             raise DockerError(
                 f"Could not get {name} for starting up", _LOGGER.error
             ) from err
@@ -616,36 +640,44 @@ class DockerAPI(CoreSysAttributes):
         _LOGGER.info("Starting %s", name)
         try:
             docker_container.start()
-        except (DockerException, requests.RequestException) as err:
+        except (docker_errors.DockerException, requests.RequestException) as err:
             raise DockerError(f"Can't start {name}: {err}", _LOGGER.error) from err
 
     def restart_container(self, name: str, timeout: int) -> None:
         """Restart docker container."""
         try:
             container: Container = self.containers.get(name)
-        except NotFound:
-            raise DockerNotFound() from None
-        except (DockerException, requests.RequestException) as err:
-            raise DockerError() from err
+        except docker_errors.NotFound:
+            raise DockerNotFound(
+                f"Container {name} not found for restarting", _LOGGER.warning
+            ) from None
+        except (docker_errors.DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Could not get container {name} for restarting: {err!s}", _LOGGER.error
+            ) from err
 
         _LOGGER.info("Restarting %s", name)
         try:
             container.restart(timeout=timeout)
-        except (DockerException, requests.RequestException) as err:
+        except (docker_errors.DockerException, requests.RequestException) as err:
             raise DockerError(f"Can't restart {name}: {err}", _LOGGER.warning) from err
 
     def container_logs(self, name: str, tail: int = 100) -> bytes:
         """Return Docker logs of container."""
         try:
             docker_container: Container = self.containers.get(name)
-        except NotFound:
-            raise DockerNotFound() from None
-        except (DockerException, requests.RequestException) as err:
-            raise DockerError() from err
+        except docker_errors.NotFound:
+            raise DockerNotFound(
+                f"Container {name} not found for logs", _LOGGER.warning
+            ) from None
+        except (docker_errors.DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Could not get container {name} for logs: {err!s}", _LOGGER.error
+            ) from err
 
         try:
             return docker_container.logs(tail=tail, stdout=True, stderr=True)
-        except (DockerException, requests.RequestException) as err:
+        except (docker_errors.DockerException, requests.RequestException) as err:
             raise DockerError(
                 f"Can't grep logs from {name}: {err}", _LOGGER.warning
             ) from err
@@ -654,10 +686,14 @@ class DockerAPI(CoreSysAttributes):
         """Read and return stats from container."""
         try:
             docker_container: Container = self.containers.get(name)
-        except NotFound:
-            raise DockerNotFound() from None
-        except (DockerException, requests.RequestException) as err:
-            raise DockerError() from err
+        except docker_errors.NotFound:
+            raise DockerNotFound(
+                f"Container {name} not found for stats", _LOGGER.warning
+            ) from None
+        except (docker_errors.DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Could not inspect container '{name}': {err!s}", _LOGGER.error
+            ) from err
 
         # container is not running
         if docker_container.status != "running":
@@ -665,7 +701,7 @@ class DockerAPI(CoreSysAttributes):
 
         try:
             return docker_container.stats(stream=False)
-        except (DockerException, requests.RequestException) as err:
+        except (docker_errors.DockerException, requests.RequestException) as err:
             raise DockerError(
                 f"Can't read stats from {name}: {err}", _LOGGER.error
             ) from err
@@ -674,61 +710,90 @@ class DockerAPI(CoreSysAttributes):
         """Execute a command inside Docker container."""
         try:
             docker_container: Container = self.containers.get(name)
-        except NotFound:
-            raise DockerNotFound() from None
-        except (DockerException, requests.RequestException) as err:
-            raise DockerError() from err
+        except docker_errors.NotFound:
+            raise DockerNotFound(
+                f"Container {name} not found for running command", _LOGGER.warning
+            ) from None
+        except (docker_errors.DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Can't get container {name} to run command: {err!s}"
+            ) from err
 
         # Execute
         try:
             code, output = docker_container.exec_run(command)
-        except (DockerException, requests.RequestException) as err:
-            raise DockerError() from err
+        except (docker_errors.DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Can't run command in container {name}: {err!s}"
+            ) from err
 
         return CommandReturn(code, output)
 
-    def remove_image(
+    async def remove_image(
         self, image: str, version: AwesomeVersion, latest: bool = True
     ) -> None:
         """Remove a Docker image by version and latest."""
         try:
             if latest:
                 _LOGGER.info("Removing image %s with latest", image)
-                with suppress(ImageNotFound):
-                    self.images.remove(image=f"{image}:latest", force=True)
+                try:
+                    await self.images.delete(f"{image}:latest", force=True)
+                except aiodocker.DockerError as err:
+                    if err.status != HTTPStatus.NOT_FOUND:
+                        raise
 
             _LOGGER.info("Removing image %s with %s", image, version)
-            with suppress(ImageNotFound):
-                self.images.remove(image=f"{image}:{version!s}", force=True)
+            try:
+                await self.images.delete(f"{image}:{version!s}", force=True)
+            except aiodocker.DockerError as err:
+                if err.status != HTTPStatus.NOT_FOUND:
+                    raise
 
-        except (DockerException, requests.RequestException) as err:
+        except (aiodocker.DockerError, requests.RequestException) as err:
             raise DockerError(
                 f"Can't remove image {image}: {err}", _LOGGER.warning
             ) from err
 
-    def import_image(self, tar_file: Path) -> Image | None:
+    async def import_image(self, tar_file: Path) -> dict[str, Any] | None:
         """Import a tar file as image."""
         try:
             with tar_file.open("rb") as read_tar:
-                docker_image_list: list[Image] = self.images.load(read_tar)  # type: ignore
-
-            if len(docker_image_list) != 1:
-                _LOGGER.warning(
-                    "Unexpected image count %d while importing image from tar",
-                    len(docker_image_list),
-                )
-                return None
-            return docker_image_list[0]
-        except (DockerException, OSError) as err:
+                resp: list[dict[str, Any]] = self.images.import_image(read_tar)
+        except (aiodocker.DockerError, OSError) as err:
             raise DockerError(
                 f"Can't import image from tar: {err}", _LOGGER.error
             ) from err
 
+        docker_image_list: list[str] = []
+        for chunk in resp:
+            if "errorDetail" in chunk:
+                raise DockerError(
+                    f"Can't import image from tar: {chunk['errorDetail']['message']}",
+                    _LOGGER.error,
+                )
+            if "stream" in chunk:
+                if match := RE_IMPORT_IMAGE_STREAM.search(chunk["stream"]):
+                    docker_image_list.append(match.group(2))
+
+        if len(docker_image_list) != 1:
+            _LOGGER.warning(
+                "Unexpected image count %d while importing image from tar",
+                len(docker_image_list),
+            )
+            return None
+
+        try:
+            return await self.images.inspect(docker_image_list[0])
+        except (aiodocker.DockerError, requests.RequestException) as err:
+            raise DockerError(
+                f"Could not inspect imported image due to: {err!s}", _LOGGER.error
+            ) from err
+
     def export_image(self, image: str, version: AwesomeVersion, tar_file: Path) -> None:
         """Export current images into a tar file."""
         try:
             docker_image = self.api.get_image(f"{image}:{version}")
-        except (DockerException, requests.RequestException) as err:
+        except (docker_errors.DockerException, requests.RequestException) as err:
             raise DockerError(
                 f"Can't fetch image {image}: {err}", _LOGGER.error
             ) from err
@@ -745,7 +810,7 @@ class DockerAPI(CoreSysAttributes):
 
         _LOGGER.info("Export image %s done", image)
 
-    def cleanup_old_images(
+    async def cleanup_old_images(
         self,
         current_image: str,
         current_version: AwesomeVersion,
@@ -756,46 +821,57 @@ class DockerAPI(CoreSysAttributes):
         """Clean up old versions of an image."""
         image = f"{current_image}:{current_version!s}"
         try:
-            keep = {cast(str, self.images.get(image).id)}
-        except ImageNotFound:
-            raise DockerNotFound(
-                f"{current_image} not found for cleanup", _LOGGER.warning
-            ) from None
-        except (DockerException, requests.RequestException) as err:
+            try:
+                image_attr = await self.images.inspect(image)
+            except aiodocker.DockerError as err:
+                if err.status == HTTPStatus.NOT_FOUND:
+                    raise DockerNotFound(
+                        f"{current_image} not found for cleanup", _LOGGER.warning
+                    ) from None
+                raise
+        except (aiodocker.DockerError, requests.RequestException) as err:
             raise DockerError(
                 f"Can't get {current_image} for cleanup", _LOGGER.warning
             ) from err
+        keep = {cast(str, image_attr["Id"])}
 
         if keep_images:
             keep_images -= {image}
-            try:
-                for image in keep_images:
-                    # If its not found, no need to preserve it from getting removed
-                    with suppress(ImageNotFound):
-                        keep.add(cast(str, self.images.get(image).id))
-            except (DockerException, requests.RequestException) as err:
-                raise DockerError(
-                    f"Failed to get one or more images from {keep} during cleanup",
-                    _LOGGER.warning,
-                ) from err
+            results = await asyncio.gather(
+                *[self.images.inspect(image) for image in keep_images],
+                return_exceptions=True,
+            )
+            for result in results:
+                # If its not found, no need to preserve it from getting removed
+                if (
+                    isinstance(result, aiodocker.DockerError)
+                    and result.status == HTTPStatus.NOT_FOUND
+                ):
+                    continue
+                if isinstance(result, BaseException):
+                    raise DockerError(
+                        f"Failed to get one or more images from {keep} during cleanup",
+                        _LOGGER.warning,
+                    ) from result
+                keep.add(cast(str, result["Id"]))
 
         # Cleanup old and current
         image_names = list(
             old_images | {current_image} if old_images else {current_image}
         )
         try:
-            # This API accepts a list of image names. Tested and confirmed working on docker==7.1.0
-            # Its typing does say only `str` though. Bit concerning, could an update break this?
-            images_list = self.images.list(name=image_names)  # type: ignore
-        except (DockerException, requests.RequestException) as err:
+            images_list = await self.images.list(
+                filters=json.dumps({"reference": image_names})
+            )
+        except (aiodocker.DockerError, requests.RequestException) as err:
             raise DockerError(
                 f"Corrupt docker overlayfs found: {err}", _LOGGER.warning
             ) from err
 
         for docker_image in images_list:
-            if docker_image.id in keep:
+            if docker_image["Id"] in keep:
                 continue
 
-            with suppress(DockerException, requests.RequestException):
-                _LOGGER.info("Cleanup images: %s", docker_image.tags)
-                self.images.remove(docker_image.id, force=True)
+            with suppress(aiodocker.DockerError, requests.RequestException):
+                _LOGGER.info("Cleanup images: %s", docker_image["RepoTags"])
+                await self.images.delete(docker_image["Id"], force=True)

supervisor/docker/monitor.py

@@ -89,7 +89,7 @@ class DockerMonitor(CoreSysAttributes, Thread):
                         DockerContainerStateEvent(
                             name=attributes["name"],
                             state=container_state,
-                            id=event["id"],
+                            id=event["Actor"]["ID"],
                             time=event["time"],
                         ),
                     )

supervisor/docker/supervisor.py

@@ -1,10 +1,12 @@
 """Init file for Supervisor Docker object."""
 
+import asyncio
 from collections.abc import Awaitable
 from ipaddress import IPv4Address
 import logging
 import os
 
+import aiodocker
 from awesomeversion.awesomeversion import AwesomeVersion
 import docker
 import requests
@@ -112,19 +114,18 @@ class DockerSupervisor(DockerInterface):
         name="docker_supervisor_update_start_tag",
         concurrency=JobConcurrency.GROUP_QUEUE,
     )
-    def update_start_tag(self, image: str, version: AwesomeVersion) -> Awaitable[None]:
+    async def update_start_tag(self, image: str, version: AwesomeVersion) -> None:
         """Update start tag to new version."""
-        return self.sys_run_in_executor(self._update_start_tag, image, version)
-
-    def _update_start_tag(self, image: str, version: AwesomeVersion) -> None:
-        """Update start tag to new version.
-
-        Need run inside executor.
-        """
         try:
-            docker_container = self.sys_docker.containers.get(self.name)
-            docker_image = self.sys_docker.images.get(f"{image}:{version!s}")
-        except (docker.errors.DockerException, requests.RequestException) as err:
+            docker_container = await self.sys_run_in_executor(
+                self.sys_docker.containers.get, self.name
+            )
+            docker_image = await self.sys_docker.images.inspect(f"{image}:{version!s}")
+        except (
+            aiodocker.DockerError,
+            docker.errors.DockerException,
+            requests.RequestException,
+        ) as err:
             raise DockerError(
                 f"Can't get image or container to fix start tag: {err}", _LOGGER.error
             ) from err
@@ -144,8 +145,14 @@ class DockerSupervisor(DockerInterface):
                 # If version tag
                 if start_tag != "latest":
                     continue
-                docker_image.tag(start_image, start_tag)
-                docker_image.tag(start_image, version.string)
+                await asyncio.gather(
+                    self.sys_docker.images.tag(
+                        docker_image["Id"], start_image, tag=start_tag
+                    ),
+                    self.sys_docker.images.tag(
+                        docker_image["Id"], start_image, tag=version.string
+                    ),
+                )
 
-        except (docker.errors.DockerException, requests.RequestException) as err:
+        except (aiodocker.DockerError, requests.RequestException) as err:
             raise DockerError(f"Can't fix start tag: {err}", _LOGGER.error) from err

supervisor/exceptions.py

@@ -3,23 +3,23 @@
 from collections.abc import Callable
 from typing import Any
 
+MESSAGE_CHECK_SUPERVISOR_LOGS = (
+    "Check supervisor logs for details (check with '{logs_command}')"
+)
+EXTRA_FIELDS_LOGS_COMMAND = {"logs_command": "ha supervisor logs"}
+
 
 class HassioError(Exception):
     """Root exception."""
 
     error_key: str | None = None
     message_template: str | None = None
+    extra_fields: dict[str, Any] | None = None
 
     def __init__(
-        self,
-        message: str | None = None,
-        logger: Callable[..., None] | None = None,
-        *,
-        extra_fields: dict[str, Any] | None = None,
+        self, message: str | None = None, logger: Callable[..., None] | None = None
     ) -> None:
         """Raise & log."""
-        self.extra_fields = extra_fields or {}
-
         if not message and self.message_template:
             message = (
                 self.message_template.format(**self.extra_fields)
@@ -41,6 +41,88 @@ class HassioNotSupportedError(HassioError):
     """Function is not supported."""
 
 
+# API
+
+
+class APIError(HassioError, RuntimeError):
+    """API errors."""
+
+    status = 400
+
+    def __init__(
+        self,
+        message: str | None = None,
+        logger: Callable[..., None] | None = None,
+        *,
+        job_id: str | None = None,
+    ) -> None:
+        """Raise & log, optionally with job."""
+        super().__init__(message, logger)
+        self.job_id = job_id
+
+
+class APIUnauthorized(APIError):
+    """API unauthorized error."""
+
+    status = 401
+
+
+class APIForbidden(APIError):
+    """API forbidden error."""
+
+    status = 403
+
+
+class APINotFound(APIError):
+    """API not found error."""
+
+    status = 404
+
+
+class APIGone(APIError):
+    """API is no longer available."""
+
+    status = 410
+
+
+class APITooManyRequests(APIError):
+    """API too many requests error."""
+
+    status = 429
+
+
+class APIInternalServerError(APIError):
+    """API internal server error."""
+
+    status = 500
+
+
+class APIAddonNotInstalled(APIError):
+    """Not installed addon requested at addons API."""
+
+
+class APIDBMigrationInProgress(APIError):
+    """Service is unavailable due to an offline DB migration is in progress."""
+
+    status = 503
+
+
+class APIUnknownSupervisorError(APIError):
+    """Unknown error occurred within supervisor. Adds supervisor check logs rider to mesage template."""
+
+    status = 500
+
+    def __init__(
+        self, logger: Callable[..., None] | None = None, *, job_id: str | None = None
+    ) -> None:
+        """Initialize exception."""
+        self.message_template = (
+            f"{self.message_template}. {MESSAGE_CHECK_SUPERVISOR_LOGS}"
+        )
+        self.extra_fields = (self.extra_fields or {}) | EXTRA_FIELDS_LOGS_COMMAND
+        super().__init__(None, logger, job_id=job_id)
+
+
 # JobManager
 
 
@@ -122,6 +204,13 @@ class SupervisorAppArmorError(SupervisorError):
     """Supervisor AppArmor error."""
 
 
+class SupervisorUnknownError(SupervisorError, APIUnknownSupervisorError):
+    """Raise when an unknown error occurs interacting with Supervisor or its container."""
+
+    error_key = "supervisor_unknown_error"
+    message_template = "An unknown error occurred with Supervisor"
+
+
 class SupervisorJobError(SupervisorError, JobException):
     """Raise on job errors."""
 
@@ -250,6 +339,54 @@ class AddonConfigurationError(AddonsError):
     """Error with add-on configuration."""
 
 
+class AddonConfigurationInvalidError(AddonConfigurationError, APIError):
+    """Raise if invalid configuration provided for addon."""
+
+    error_key = "addon_configuration_invalid_error"
+    message_template = "Add-on {addon} has invalid options: {validation_error}"
+
+    def __init__(
+        self,
+        logger: Callable[..., None] | None = None,
+        *,
+        addon: str,
+        validation_error: str,
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"addon": addon, "validation_error": validation_error}
+        super().__init__(None, logger)
+
+
+class AddonBootConfigCannotChangeError(AddonsError, APIError):
+    """Raise if user attempts to change addon boot config when it can't be changed."""
+
+    error_key = "addon_boot_config_cannot_change_error"
+    message_template = (
+        "Addon {addon} boot option is set to {boot_config} so it cannot be changed"
+    )
+
+    def __init__(
+        self, logger: Callable[..., None] | None = None, *, addon: str, boot_config: str
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"addon": addon, "boot_config": boot_config}
+        super().__init__(None, logger)
+
+
+class AddonNotRunningError(AddonsError, APIError):
+    """Raise when an addon is not running."""
+
+    error_key = "addon_not_running_error"
+    message_template = "Add-on {addon} is not running"
+
+    def __init__(
+        self, logger: Callable[..., None] | None = None, *, addon: str
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"addon": addon}
+        super().__init__(None, logger)
+
+
 class AddonNotSupportedError(HassioNotSupportedError):
     """Addon doesn't support a function."""
 
@@ -268,11 +405,8 @@ class AddonNotSupportedArchitectureError(AddonNotSupportedError):
         architectures: list[str],
     ) -> None:
         """Initialize exception."""
-        super().__init__(
-            None,
-            logger,
-            extra_fields={"slug": slug, "architectures": ", ".join(architectures)},
-        )
+        self.extra_fields = {"slug": slug, "architectures": ", ".join(architectures)}
+        super().__init__(None, logger)
 
 
 class AddonNotSupportedMachineTypeError(AddonNotSupportedError):
@@ -289,11 +423,8 @@ class AddonNotSupportedMachineTypeError(AddonNotSupportedError):
         machine_types: list[str],
     ) -> None:
         """Initialize exception."""
-        super().__init__(
-            None,
-            logger,
-            extra_fields={"slug": slug, "machine_types": ", ".join(machine_types)},
-        )
+        self.extra_fields = {"slug": slug, "machine_types": ", ".join(machine_types)}
+        super().__init__(None, logger)
 
 
 class AddonNotSupportedHomeAssistantVersionError(AddonNotSupportedError):
@@ -310,11 +441,80 @@ class AddonNotSupportedHomeAssistantVersionError(AddonNotSupportedError):
         version: str,
     ) -> None:
         """Initialize exception."""
-        super().__init__(
-            None,
-            logger,
-            extra_fields={"slug": slug, "version": version},
-        )
+        self.extra_fields = {"slug": slug, "version": version}
+        super().__init__(None, logger)
+
+
+class AddonNotSupportedWriteStdinError(AddonNotSupportedError, APIError):
+    """Addon does not support writing to stdin."""
+
+    error_key = "addon_not_supported_write_stdin_error"
+    message_template = "Add-on {addon} does not support writing to stdin"
+
+    def __init__(
+        self, logger: Callable[..., None] | None = None, *, addon: str
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"addon": addon}
+        super().__init__(None, logger)
+
+
+class AddonBuildDockerfileMissingError(AddonNotSupportedError, APIError):
+    """Raise when addon build invalid because dockerfile is missing."""
+
+    error_key = "addon_build_dockerfile_missing_error"
+    message_template = (
+        "Cannot build addon '{addon}' because dockerfile is missing. A repair "
+        "using '{repair_command}' will fix this if the cause is data "
+        "corruption. Otherwise please report this to the addon developer."
+    )
+
+    def __init__(
+        self, logger: Callable[..., None] | None = None, *, addon: str
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"addon": addon, "repair_command": "ha supervisor repair"}
+        super().__init__(None, logger)
+
+
+class AddonBuildArchitectureNotSupportedError(AddonNotSupportedError, APIError):
+    """Raise when addon cannot be built on system because it doesn't support its architecture."""
+
+    error_key = "addon_build_architecture_not_supported_error"
+    message_template = (
+        "Cannot build addon '{addon}' because its supported architectures "
+        "({addon_arches}) do not match the system supported architectures ({system_arches})"
+    )
+
+    def __init__(
+        self,
+        logger: Callable[..., None] | None = None,
+        *,
+        addon: str,
+        addon_arch_list: list[str],
+        system_arch_list: list[str],
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {
+            "addon": addon,
+            "addon_arches": ", ".join(addon_arch_list),
+            "system_arches": ", ".join(system_arch_list),
+        }
+        super().__init__(None, logger)
+
+
+class AddonUnknownError(AddonsError, APIUnknownSupervisorError):
+    """Raise when unknown error occurs taking an action for an addon."""
+
+    error_key = "addon_unknown_error"
+    message_template = "An unknown error occurred with addon {addon}"
+
+    def __init__(
+        self, logger: Callable[..., None] | None = None, *, addon: str
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"addon": addon}
+        super().__init__(logger)
 
 
 class AddonsJobError(AddonsError, JobException):
@@ -346,13 +546,59 @@ class AuthError(HassioError):
     """Auth errors."""
 
 
-class AuthPasswordResetError(HassioError):
+# This one uses the check logs rider even though its not a 500 error because it
+# is bad practice to return error specifics from a password reset API.
+class AuthPasswordResetError(AuthError, APIError):
     """Auth error if password reset failed."""
 
+    error_key = "auth_password_reset_error"
+    message_template = (
+        f"Unable to reset password for '{{user}}'. {MESSAGE_CHECK_SUPERVISOR_LOGS}"
+    )
 
-class AuthListUsersError(HassioError):
+    def __init__(self, logger: Callable[..., None] | None = None, *, user: str) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"user": user} | EXTRA_FIELDS_LOGS_COMMAND
+        super().__init__(None, logger)
+
+
+class AuthListUsersError(AuthError, APIUnknownSupervisorError):
     """Auth error if listing users failed."""
 
+    error_key = "auth_list_users_error"
+    message_template = "Can't request listing users on Home Assistant"
+
+
+class AuthListUsersNoneResponseError(AuthError, APIInternalServerError):
+    """Auth error if listing users returned invalid None response."""
+
+    error_key = "auth_list_users_none_response_error"
+    message_template = "Home Assistant returned invalid response of `{none}` instead of a list of users. Check Home Assistant logs for details (check with `{logs_command}`)"
+    extra_fields = {"none": "None", "logs_command": "ha core logs"}
+
+    def __init__(self, logger: Callable[..., None] | None = None) -> None:
+        """Initialize exception."""
+        super().__init__(None, logger)
+
+
+class AuthInvalidNoneValueError(AuthError, APIUnauthorized):
+    """Auth error if None provided as username or password."""
+
+    error_key = "auth_invalid_none_value_error"
+    message_template = "{none} as username or password is not supported"
+    extra_fields = {"none": "None"}
+
+    def __init__(self, logger: Callable[..., None] | None = None) -> None:
+        """Initialize exception."""
+        super().__init__(None, logger)
+
+
+class AuthHomeAssistantAPIValidationError(AuthError, APIUnknownSupervisorError):
+    """Error encountered trying to validate auth details via Home Assistant API."""
+
+    error_key = "auth_home_assistant_api_validation_error"
+    message_template = "Unable to validate authentication details with Home Assistant"
+
 
 # Host
 
@@ -385,60 +631,6 @@ class HostLogError(HostError):
     """Internal error with host log."""
 
 
-# API
-
-
-class APIError(HassioError, RuntimeError):
-    """API errors."""
-
-    status = 400
-
-    def __init__(
-        self,
-        message: str | None = None,
-        logger: Callable[..., None] | None = None,
-        *,
-        job_id: str | None = None,
-        error: HassioError | None = None,
-    ) -> None:
-        """Raise & log, optionally with job."""
-        # Allow these to be set from another error here since APIErrors essentially wrap others to add a status
-        self.error_key = error.error_key if error else None
-        self.message_template = error.message_template if error else None
-        super().__init__(
-            message, logger, extra_fields=error.extra_fields if error else None
-        )
-        self.job_id = job_id
-
-
-class APIForbidden(APIError):
-    """API forbidden error."""
-
-    status = 403
-
-
-class APINotFound(APIError):
-    """API not found error."""
-
-    status = 404
-
-
-class APIGone(APIError):
-    """API is no longer available."""
-
-    status = 410
-
-
-class APIAddonNotInstalled(APIError):
-    """Not installed addon requested at addons API."""
-
-
-class APIDBMigrationInProgress(APIError):
-    """Service is unavailable due to an offline DB migration is in progress."""
-
-    status = 503
-
-
 # Service / Discovery
 
 
@@ -616,6 +808,10 @@ class DockerError(HassioError):
     """Docker API/Transport errors."""
 
 
+class DockerBuildError(DockerError):
+    """Docker error during build."""
+
+
 class DockerAPIError(DockerError):
     """Docker API error."""
 
@@ -639,9 +835,29 @@ class DockerLogOutOfOrder(DockerError):
 class DockerNoSpaceOnDevice(DockerError):
     """Raise if a docker pull fails due to available space."""
 
+    error_key = "docker_no_space_on_device"
+    message_template = "No space left on disk"
+
     def __init__(self, logger: Callable[..., None] | None = None) -> None:
         """Raise & log."""
-        super().__init__("No space left on disk", logger=logger)
+        super().__init__(None, logger=logger)
+
+
+class DockerHubRateLimitExceeded(DockerError, APITooManyRequests):
+    """Raise for docker hub rate limit exceeded error."""
+
+    error_key = "dockerhub_rate_limit_exceeded"
+    message_template = (
+        "Your IP address has made too many requests to Docker Hub which activated a rate limit. "
+        "For more details see {dockerhub_rate_limit_url}"
+    )
+    extra_fields = {
+        "dockerhub_rate_limit_url": "https://www.home-assistant.io/more-info/dockerhub-rate-limit"
+    }
+
+    def __init__(self, logger: Callable[..., None] | None = None) -> None:
+        """Raise & log."""
+        super().__init__(None, logger=logger)
 
 
 class DockerJobError(DockerError, JobException):
@@ -712,6 +928,20 @@ class StoreNotFound(StoreError):
     """Raise if slug is not known."""
 
 
+class StoreAddonNotFoundError(StoreError, APINotFound):
+    """Raise if a requested addon is not in the store."""
+
+    error_key = "store_addon_not_found_error"
+    message_template = "Addon {addon} does not exist in the store"
+
+    def __init__(
+        self, logger: Callable[..., None] | None = None, *, addon: str
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"addon": addon}
+        super().__init__(None, logger)
+
+
 class StoreJobError(StoreError, JobException):
     """Raise on job error with git."""
 
@@ -759,6 +989,55 @@ class BackupFileExistError(BackupError):
     """Raise if the backup file already exists."""
 
 
+class AddonBackupMetadataInvalidError(BackupError, APIError):
+    """Raise if invalid metadata file provided for addon in backup."""
+
+    error_key = "addon_backup_metadata_invalid_error"
+    message_template = (
+        "Metadata file for add-on {addon} in backup is invalid: {validation_error}"
+    )
+
+    def __init__(
+        self,
+        logger: Callable[..., None] | None = None,
+        *,
+        addon: str,
+        validation_error: str,
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {"addon": addon, "validation_error": validation_error}
+        super().__init__(None, logger)
+
+
+class AddonPrePostBackupCommandReturnedError(BackupError, APIError):
+    """Raise when addon's pre/post backup command returns an error."""
+
+    error_key = "addon_pre_post_backup_command_returned_error"
+    message_template = (
+        "Pre-/Post backup command for add-on {addon} returned error code: "
+        "{exit_code}. Please report this to the addon developer. Enable debug "
+        "logging to capture complete command output using {debug_logging_command}"
+    )
+
+    def __init__(
+        self, logger: Callable[..., None] | None = None, *, addon: str, exit_code: int
+    ) -> None:
+        """Initialize exception."""
+        self.extra_fields = {
+            "addon": addon,
+            "exit_code": exit_code,
+            "debug_logging_command": "ha supervisor options --logging debug",
+        }
+        super().__init__(None, logger)
+
+
+class BackupRestoreUnknownError(BackupError, APIUnknownSupervisorError):
+    """Raise when an unknown error occurs during backup or restore."""
+
+    error_key = "backup_restore_unknown_error"
+    message_template = "An unknown error occurred during backup/restore"
+
+
 # Security
 
 

supervisor/jobs/__init__.py

@@ -102,13 +102,17 @@ class SupervisorJobError:
         "Unknown error, see Supervisor logs (check with 'ha supervisor logs')"
     )
     stage: str | None = None
+    error_key: str | None = None
+    extra_fields: dict[str, Any] | None = None
 
-    def as_dict(self) -> dict[str, str | None]:
+    def as_dict(self) -> dict[str, Any]:
         """Return dictionary representation."""
         return {
             "type": self.type_.__name__,
             "message": self.message,
             "stage": self.stage,
+            "error_key": self.error_key,
+            "extra_fields": self.extra_fields,
         }
 
 
@@ -158,7 +162,9 @@ class SupervisorJob:
     def capture_error(self, err: HassioError | None = None) -> None:
         """Capture an error or record that an unknown error has occurred."""
         if err:
-            new_error = SupervisorJobError(type(err), str(err), self.stage)
+            new_error = SupervisorJobError(
+                type(err), str(err), self.stage, err.error_key, err.extra_fields
+            )
         else:
             new_error = SupervisorJobError(stage=self.stage)
         self.errors += [new_error]

supervisor/misc/filter.py

@@ -64,6 +64,19 @@ def filter_data(coresys: CoreSys, event: Event, hint: Hint) -> Event | None:
 
     # Not full startup - missing information
     if coresys.core.state in (CoreState.INITIALIZE, CoreState.SETUP):
+        # During SETUP, we have basic system info available for better debugging
+        if coresys.core.state == CoreState.SETUP:
+            event.setdefault("contexts", {}).update(
+                {
+                    "versions": {
+                        "docker": coresys.docker.info.version,
+                        "supervisor": coresys.supervisor.version,
+                    },
+                    "host": {
+                        "machine": coresys.machine,
+                    },
+                }
+            )
         return event
 
     # List installed addons

supervisor/supervisor.py

@@ -28,8 +28,8 @@ from .exceptions import (
     DockerError,
     HostAppArmorError,
     SupervisorAppArmorError,
-    SupervisorError,
     SupervisorJobError,
+    SupervisorUnknownError,
     SupervisorUpdateError,
 )
 from .jobs.const import JobCondition, JobThrottle
@@ -261,7 +261,7 @@ class Supervisor(CoreSysAttributes):
         try:
             return await self.instance.stats()
         except DockerError as err:
-            raise SupervisorError() from err
+            raise SupervisorUnknownError() from err
 
     async def repair(self):
         """Repair local Supervisor data."""

tests/addons/test_addon.py

@@ -3,24 +3,34 @@
 import asyncio
 from datetime import timedelta
 import errno
+from http import HTTPStatus
 from pathlib import Path
-from unittest.mock import MagicMock, PropertyMock, patch
+from typing import Any
+from unittest.mock import MagicMock, PropertyMock, call, patch
 
+import aiodocker
 from awesomeversion import AwesomeVersion
-from docker.errors import DockerException, ImageNotFound, NotFound
+from docker.errors import APIError, DockerException, NotFound
 import pytest
 from securetar import SecureTarFile
 
 from supervisor.addons.addon import Addon
 from supervisor.addons.const import AddonBackupMode
 from supervisor.addons.model import AddonModel
+from supervisor.config import CoreConfig
 from supervisor.const import AddonBoot, AddonState, BusEvent
 from supervisor.coresys import CoreSys
 from supervisor.docker.addon import DockerAddon
 from supervisor.docker.const import ContainerState
-from supervisor.docker.manager import CommandReturn
+from supervisor.docker.manager import CommandReturn, DockerAPI
 from supervisor.docker.monitor import DockerContainerStateEvent
-from supervisor.exceptions import AddonsError, AddonsJobError, AudioUpdateError
+from supervisor.exceptions import (
+    AddonPrePostBackupCommandReturnedError,
+    AddonsJobError,
+    AddonUnknownError,
+    AudioUpdateError,
+    HassioError,
+)
 from supervisor.hardware.helper import HwHelper
 from supervisor.ingress import Ingress
 from supervisor.store.repository import Repository
@@ -499,31 +509,26 @@ async def test_backup_with_pre_post_command(
 
 
 @pytest.mark.parametrize(
-    "get_error,exception_on_exec",
+    ("container_get_side_effect", "exec_run_side_effect", "exc_type_raised"),
     [
-        (NotFound("missing"), False),
-        (DockerException(), False),
-        (None, True),
-        (None, False),
+        (NotFound("missing"), [(1, None)], AddonUnknownError),
+        (DockerException(), [(1, None)], AddonUnknownError),
+        (None, DockerException(), AddonUnknownError),
+        (None, [(1, None)], AddonPrePostBackupCommandReturnedError),
     ],
 )
+@pytest.mark.usefixtures("tmp_supervisor_data", "path_extern")
 async def test_backup_with_pre_command_error(
     coresys: CoreSys,
     install_addon_ssh: Addon,
     container: MagicMock,
-    get_error: DockerException | None,
-    exception_on_exec: bool,
-    tmp_supervisor_data,
-    path_extern,
+    container_get_side_effect: DockerException | None,
+    exec_run_side_effect: DockerException | list[tuple[int, Any]],
+    exc_type_raised: type[HassioError],
 ) -> None:
     """Test backing up an addon with error running pre command."""
-    if get_error:
-        coresys.docker.containers.get.side_effect = get_error
-
-    if exception_on_exec:
-        container.exec_run.side_effect = DockerException()
-    else:
-        container.exec_run.return_value = (1, None)
+    coresys.docker.containers.get.side_effect = container_get_side_effect
+    container.exec_run.side_effect = exec_run_side_effect
 
     install_addon_ssh.path_data.mkdir()
     await install_addon_ssh.load()
@@ -532,7 +537,7 @@ async def test_backup_with_pre_command_error(
     with (
         patch.object(DockerAddon, "is_running", return_value=True),
         patch.object(Addon, "backup_pre", new=PropertyMock(return_value="backup_pre")),
-        pytest.raises(AddonsError),
+        pytest.raises(exc_type_raised),
     ):
         assert await install_addon_ssh.backup(tarfile) is None
 
@@ -861,16 +866,14 @@ async def test_addon_loads_wrong_image(
 
     container.remove.assert_called_with(force=True, v=True)
     # one for removing the addon, one for removing the addon builder
-    assert coresys.docker.images.remove.call_count == 2
+    assert coresys.docker.images.delete.call_count == 2
 
-    assert coresys.docker.images.remove.call_args_list[0].kwargs == {
-        "image": "local/aarch64-addon-ssh:latest",
-        "force": True,
-    }
-    assert coresys.docker.images.remove.call_args_list[1].kwargs == {
-        "image": "local/aarch64-addon-ssh:9.2.1",
-        "force": True,
-    }
+    assert coresys.docker.images.delete.call_args_list[0] == call(
+        "local/aarch64-addon-ssh:latest", force=True
+    )
+    assert coresys.docker.images.delete.call_args_list[1] == call(
+        "local/aarch64-addon-ssh:9.2.1", force=True
+    )
     mock_run_command.assert_called_once()
     assert mock_run_command.call_args.args[0] == "docker.io/library/docker"
     assert mock_run_command.call_args.kwargs["version"] == "1.0.0-cli"
@@ -894,7 +897,9 @@ async def test_addon_loads_missing_image(
     mock_amd64_arch_supported,
 ):
     """Test addon corrects a missing image on load."""
-    coresys.docker.images.get.side_effect = ImageNotFound("missing")
+    coresys.docker.images.inspect.side_effect = aiodocker.DockerError(
+        HTTPStatus.NOT_FOUND, {"message": "missing"}
+    )
 
     with (
         patch("pathlib.Path.is_file", return_value=True),
@@ -926,41 +931,51 @@ async def test_addon_loads_missing_image(
     assert install_addon_ssh.image == "local/amd64-addon-ssh"
 
 
+@pytest.mark.parametrize(
+    "pull_image_exc",
+    [APIError("error"), aiodocker.DockerError(400, {"message": "error"})],
+)
+@pytest.mark.usefixtures("container", "mock_amd64_arch_supported")
 async def test_addon_load_succeeds_with_docker_errors(
     coresys: CoreSys,
     install_addon_ssh: Addon,
-    container: MagicMock,
     caplog: pytest.LogCaptureFixture,
-    mock_amd64_arch_supported,
+    pull_image_exc: Exception,
 ):
     """Docker errors while building/pulling an image during load should not raise and fail setup."""
     # Build env invalid failure
-    coresys.docker.images.get.side_effect = ImageNotFound("missing")
+    coresys.docker.images.inspect.side_effect = aiodocker.DockerError(
+        HTTPStatus.NOT_FOUND, {"message": "missing"}
+    )
     caplog.clear()
     await install_addon_ssh.load()
-    assert "Invalid build environment" in caplog.text
+    assert "Cannot build addon 'local_ssh' because dockerfile is missing" in caplog.text
 
     # Image build failure
-    coresys.docker.images.build.side_effect = DockerException()
     caplog.clear()
     with (
         patch("pathlib.Path.is_file", return_value=True),
         patch.object(
-            type(coresys.config),
-            "local_to_extern_path",
-            return_value="/addon/path/on/host",
+            CoreConfig, "local_to_extern_path", return_value="/addon/path/on/host"
+        ),
+        patch.object(
+            DockerAPI,
+            "run_command",
+            return_value=MagicMock(exit_code=1, output=b"error"),
         ),
     ):
         await install_addon_ssh.load()
-    assert "Can't build local/amd64-addon-ssh:9.2.1" in caplog.text
+    assert (
+        "Can't build local/amd64-addon-ssh:9.2.1: Docker build failed for local/amd64-addon-ssh:9.2.1 (exit code 1). Build output:\nerror"
+        in caplog.text
+    )
 
     # Image pull failure
     install_addon_ssh.data["image"] = "test/amd64-addon-ssh"
-    coresys.docker.images.build.reset_mock(side_effect=True)
-    coresys.docker.pull_image.side_effect = DockerException()
     caplog.clear()
-    await install_addon_ssh.load()
-    assert "Unknown error with test/amd64-addon-ssh:9.2.1" in caplog.text
+    with patch.object(DockerAPI, "pull_image", side_effect=pull_image_exc):
+        await install_addon_ssh.load()
+    assert "Can't install test/amd64-addon-ssh:9.2.1:" in caplog.text
 
 
 async def test_addon_manual_only_boot(coresys: CoreSys, install_addon_example: Addon):

tests/addons/test_build.py

@@ -3,10 +3,12 @@
 from unittest.mock import PropertyMock, patch
 
 from awesomeversion import AwesomeVersion
+import pytest
 
 from supervisor.addons.addon import Addon
 from supervisor.addons.build import AddonBuild
 from supervisor.coresys import CoreSys
+from supervisor.exceptions import AddonBuildDockerfileMissingError
 
 from tests.common import is_in_list
 
@@ -102,11 +104,11 @@ async def test_build_valid(coresys: CoreSys, install_addon_ssh: Addon):
             type(coresys.arch), "default", new=PropertyMock(return_value="aarch64")
         ),
     ):
-        assert await build.is_valid()
+        assert (await build.is_valid()) is None
 
 
 async def test_build_invalid(coresys: CoreSys, install_addon_ssh: Addon):
-    """Test platform set in docker args."""
+    """Test build not supported because Dockerfile missing for specified architecture."""
     build = await AddonBuild(coresys, install_addon_ssh).load_config()
     with (
         patch.object(
@@ -115,5 +117,6 @@ async def test_build_invalid(coresys: CoreSys, install_addon_ssh: Addon):
         patch.object(
             type(coresys.arch), "default", new=PropertyMock(return_value="amd64")
         ),
+        pytest.raises(AddonBuildDockerfileMissingError),
     ):
-        assert not await build.is_valid()
+        await build.is_valid()

tests/addons/test_manager.py

@@ -4,7 +4,7 @@ import asyncio
 from collections.abc import AsyncGenerator, Generator
 from copy import deepcopy
 from pathlib import Path
-from unittest.mock import AsyncMock, MagicMock, Mock, PropertyMock, patch
+from unittest.mock import AsyncMock, MagicMock, Mock, PropertyMock, call, patch
 
 from awesomeversion import AwesomeVersion
 import pytest
@@ -514,19 +514,13 @@ async def test_shared_image_kept_on_uninstall(
     latest = f"{install_addon_example.image}:latest"
 
     await coresys.addons.uninstall("local_example2")
-    coresys.docker.images.remove.assert_not_called()
+    coresys.docker.images.delete.assert_not_called()
     assert not coresys.addons.get("local_example2", local_only=True)
 
     await coresys.addons.uninstall("local_example")
-    assert coresys.docker.images.remove.call_count == 2
-    assert coresys.docker.images.remove.call_args_list[0].kwargs == {
-        "image": latest,
-        "force": True,
-    }
-    assert coresys.docker.images.remove.call_args_list[1].kwargs == {
-        "image": image,
-        "force": True,
-    }
+    assert coresys.docker.images.delete.call_count == 2
+    assert coresys.docker.images.delete.call_args_list[0] == call(latest, force=True)
+    assert coresys.docker.images.delete.call_args_list[1] == call(image, force=True)
     assert not coresys.addons.get("local_example", local_only=True)
 
 
@@ -554,19 +548,17 @@ async def test_shared_image_kept_on_update(
     assert example_2.version == "1.2.0"
     assert install_addon_example_image.version == "1.2.0"
 
-    image_new = MagicMock()
-    image_new.id = "image_new"
-    image_old = MagicMock()
-    image_old.id = "image_old"
-    docker.images.get.side_effect = [image_new, image_old]
+    image_new = {"Id": "image_new", "RepoTags": ["image_new:latest"]}
+    image_old = {"Id": "image_old", "RepoTags": ["image_old:latest"]}
+    docker.images.inspect.side_effect = [image_new, image_old]
     docker.images.list.return_value = [image_new, image_old]
 
     with patch.object(DockerAPI, "pull_image", return_value=image_new):
         await coresys.addons.update("local_example2")
-        docker.images.remove.assert_not_called()
+        docker.images.delete.assert_not_called()
         assert example_2.version == "1.3.0"
 
-        docker.images.get.side_effect = [image_new]
+        docker.images.inspect.side_effect = [image_new]
         await coresys.addons.update("local_example_image")
-        docker.images.remove.assert_called_once_with("image_old", force=True)
+        docker.images.delete.assert_called_once_with("image_old", force=True)
         assert install_addon_example_image.version == "1.3.0"

tests/api/test_addons.py

@@ -482,6 +482,11 @@ async def test_addon_options_boot_mode_manual_only_invalid(
         body["message"]
         == "Addon local_example boot option is set to manual_only so it cannot be changed"
     )
+    assert body["error_key"] == "addon_boot_config_cannot_change_error"
+    assert body["extra_fields"] == {
+        "addon": "local_example",
+        "boot_config": "manual_only",
+    }
 
 
 async def get_message(resp: ClientResponse, json_expected: bool) -> str:
@@ -550,3 +555,94 @@ async def test_addon_not_installed(
     resp = await api_client.request(method, url)
     assert resp.status == 400
     assert await get_message(resp, json_expected) == "Addon is not installed"
+
+
+async def test_addon_set_options(api_client: TestClient, install_addon_example: Addon):
+    """Test setting options for an addon."""
+    resp = await api_client.post(
+        "/addons/local_example/options", json={"options": {"message": "test"}}
+    )
+    assert resp.status == 200
+    assert install_addon_example.options == {"message": "test"}
+
+
+async def test_addon_set_options_error(
+    api_client: TestClient, install_addon_example: Addon
+):
+    """Test setting options for an addon."""
+    resp = await api_client.post(
+        "/addons/local_example/options", json={"options": {"message": True}}
+    )
+    assert resp.status == 400
+    body = await resp.json()
+    assert (
+        body["message"]
+        == "Add-on local_example has invalid options: not a valid value. Got {'message': True}"
+    )
+    assert body["error_key"] == "addon_configuration_invalid_error"
+    assert body["extra_fields"] == {
+        "addon": "local_example",
+        "validation_error": "not a valid value. Got {'message': True}",
+    }
+
+
+async def test_addon_start_options_error(
+    api_client: TestClient,
+    install_addon_example: Addon,
+    caplog: pytest.LogCaptureFixture,
+):
+    """Test error writing options when trying to start addon."""
+    install_addon_example.options = {"message": "hello"}
+
+    # Simulate OS error trying to write the file
+    with patch("supervisor.utils.json.atomic_write", side_effect=OSError("fail")):
+        resp = await api_client.post("/addons/local_example/start")
+        assert resp.status == 500
+        body = await resp.json()
+        assert (
+            body["message"]
+            == "An unknown error occurred with addon local_example. Check supervisor logs for details (check with 'ha supervisor logs')"
+        )
+        assert body["error_key"] == "addon_unknown_error"
+        assert body["extra_fields"] == {
+            "addon": "local_example",
+            "logs_command": "ha supervisor logs",
+        }
+        assert "Add-on local_example can't write options" in caplog.text
+
+    # Simulate an update with a breaking change for options schema creating failure on start
+    caplog.clear()
+    install_addon_example.data["schema"] = {"message": "bool"}
+    resp = await api_client.post("/addons/local_example/start")
+    assert resp.status == 400
+    body = await resp.json()
+    assert (
+        body["message"]
+        == "Add-on local_example has invalid options: expected boolean. Got {'message': 'hello'}"
+    )
+    assert body["error_key"] == "addon_configuration_invalid_error"
+    assert body["extra_fields"] == {
+        "addon": "local_example",
+        "validation_error": "expected boolean. Got {'message': 'hello'}",
+    }
+    assert (
+        "Add-on local_example has invalid options: expected boolean. Got {'message': 'hello'}"
+        in caplog.text
+    )
+
+
+@pytest.mark.parametrize(("method", "action"), [("get", "stats"), ("post", "stdin")])
+async def test_addon_not_running_error(
+    api_client: TestClient, install_addon_example: Addon, method: str, action: str
+):
+    """Test addon not running error for endpoints that require that."""
+    with patch.object(
+        Addon, "with_stdin", return_value=PropertyMock(return_value=True)
+    ):
+        resp = await api_client.request(method, f"/addons/local_example/{action}")
+
+    assert resp.status == 400
+    body = await resp.json()
+    assert body["message"] == "Add-on local_example is not running"
+    assert body["error_key"] == "addon_not_running_error"
+    assert body["extra_fields"] == {"addon": "local_example"}

tests/api/test_auth.py

@@ -6,9 +6,11 @@ from unittest.mock import AsyncMock, MagicMock, patch
 from aiohttp.hdrs import WWW_AUTHENTICATE
 from aiohttp.test_utils import TestClient
 import pytest
+from securetar import Any
 
 from supervisor.addons.addon import Addon
 from supervisor.coresys import CoreSys
+from supervisor.exceptions import HomeAssistantAPIError, HomeAssistantWSError
 
 from tests.common import MockResponse
 from tests.const import TEST_ADDON_SLUG
@@ -100,6 +102,52 @@ async def test_password_reset(
     assert "Successful password reset for 'john'" in caplog.text
 
 
+@pytest.mark.parametrize(
+    ("post_mock", "expected_log"),
+    [
+        (
+            MagicMock(return_value=MockResponse(status=400)),
+            "The user 'john' is not registered",
+        ),
+        (
+            MagicMock(side_effect=HomeAssistantAPIError("fail")),
+            "Can't request password reset on Home Assistant: fail",
+        ),
+    ],
+)
+async def test_failed_password_reset(
+    api_client: TestClient,
+    coresys: CoreSys,
+    caplog: pytest.LogCaptureFixture,
+    websession: MagicMock,
+    post_mock: MagicMock,
+    expected_log: str,
+):
+    """Test failed password reset."""
+    coresys.homeassistant.api.access_token = "abc123"
+    # pylint: disable-next=protected-access
+    coresys.homeassistant.api._access_token_expires = datetime.now(tz=UTC) + timedelta(
+        days=1
+    )
+
+    websession.post = post_mock
+    resp = await api_client.post(
+        "/auth/reset", json={"username": "john", "password": "doe"}
+    )
+    assert resp.status == 400
+    body = await resp.json()
+    assert (
+        body["message"]
+        == "Unable to reset password for 'john'. Check supervisor logs for details (check with 'ha supervisor logs')"
+    )
+    assert body["error_key"] == "auth_password_reset_error"
+    assert body["extra_fields"] == {
+        "user": "john",
+        "logs_command": "ha supervisor logs",
+    }
+    assert expected_log in caplog.text
+
+
 async def test_list_users(
     api_client: TestClient, coresys: CoreSys, ha_ws_client: AsyncMock
 ):
@@ -120,6 +168,48 @@ async def test_list_users(
     ]
 
 
+@pytest.mark.parametrize(
+    ("send_command_mock", "error_response", "expected_log"),
+    [
+        (
+            AsyncMock(return_value=None),
+            {
+                "result": "error",
+                "message": "Home Assistant returned invalid response of `None` instead of a list of users. Check Home Assistant logs for details (check with `ha core logs`)",
+                "error_key": "auth_list_users_none_response_error",
+                "extra_fields": {"none": "None", "logs_command": "ha core logs"},
+            },
+            "Home Assistant returned invalid response of `None` instead of a list of users. Check Home Assistant logs for details (check with `ha core logs`)",
+        ),
+        (
+            AsyncMock(side_effect=HomeAssistantWSError("fail")),
+            {
+                "result": "error",
+                "message": "Can't request listing users on Home Assistant. Check supervisor logs for details (check with 'ha supervisor logs')",
+                "error_key": "auth_list_users_error",
+                "extra_fields": {"logs_command": "ha supervisor logs"},
+            },
+            "Can't request listing users on Home Assistant: fail",
+        ),
+    ],
+)
+async def test_list_users_failure(
+    api_client: TestClient,
+    ha_ws_client: AsyncMock,
+    caplog: pytest.LogCaptureFixture,
+    send_command_mock: AsyncMock,
+    error_response: dict[str, Any],
+    expected_log: str,
+):
+    """Test failure listing users via API."""
+    ha_ws_client.async_send_command = send_command_mock
+    resp = await api_client.get("/auth/list")
+    assert resp.status == 500
+    result = await resp.json()
+    assert result == error_response
+    assert expected_log in caplog.text
+
+
 @pytest.mark.parametrize(
     ("field", "api_client"),
     [("username", TEST_ADDON_SLUG), ("user", TEST_ADDON_SLUG)],

tests/api/test_backups.py

@@ -17,6 +17,7 @@ from supervisor.const import CoreState
 from supervisor.coresys import CoreSys
 from supervisor.docker.manager import DockerAPI
 from supervisor.exceptions import (
+    AddonPrePostBackupCommandReturnedError,
     AddonsError,
     BackupInvalidError,
     HomeAssistantBackupError,
@@ -24,6 +25,7 @@ from supervisor.exceptions import (
 from supervisor.homeassistant.core import HomeAssistantCore
 from supervisor.homeassistant.module import HomeAssistant
 from supervisor.homeassistant.websocket import HomeAssistantWebSocket
+from supervisor.jobs import SupervisorJob
 from supervisor.mounts.mount import Mount
 from supervisor.supervisor import Supervisor
 
@@ -401,6 +403,8 @@ async def test_api_backup_errors(
             "type": "BackupError",
             "message": str(err),
             "stage": None,
+            "error_key": None,
+            "extra_fields": None,
         }
     ]
     assert job["child_jobs"][2]["name"] == "backup_store_folders"
@@ -437,6 +441,8 @@ async def test_api_backup_errors(
             "type": "HomeAssistantBackupError",
             "message": "Backup error",
             "stage": "home_assistant",
+            "error_key": None,
+            "extra_fields": None,
         }
     ]
     assert job["child_jobs"][0]["name"] == "backup_store_homeassistant"
@@ -445,6 +451,8 @@ async def test_api_backup_errors(
             "type": "HomeAssistantBackupError",
             "message": "Backup error",
             "stage": None,
+            "error_key": None,
+            "extra_fields": None,
         }
     ]
     assert len(job["child_jobs"]) == 1
@@ -749,6 +757,8 @@ async def test_backup_to_multiple_locations_error_on_copy(
             "type": "BackupError",
             "message": "Could not copy backup to .cloud_backup due to: ",
             "stage": None,
+            "error_key": None,
+            "extra_fields": None,
         }
     ]
 
@@ -1483,3 +1493,44 @@ async def test_immediate_list_after_missing_file_restore(
     result = await resp.json()
     assert len(result["data"]["backups"]) == 1
     assert result["data"]["backups"][0]["slug"] == "93b462f8"
+
+
+@pytest.mark.parametrize("command", ["backup_pre", "backup_post"])
+@pytest.mark.usefixtures("install_addon_example", "tmp_supervisor_data")
+async def test_pre_post_backup_command_error(
+    api_client: TestClient, coresys: CoreSys, container: MagicMock, command: str
+):
+    """Test pre/post backup command error."""
+    await coresys.core.set_state(CoreState.RUNNING)
+    coresys.hardware.disk.get_disk_free_space = lambda x: 5000
+
+    container.status = "running"
+    container.exec_run.return_value = (1, b"")
+    with patch.object(Addon, command, return_value=PropertyMock(return_value="test")):
+        resp = await api_client.post(
+            "/backups/new/partial", json={"addons": ["local_example"]}
+        )
+
+    assert resp.status == 200
+    body = await resp.json()
+    job_id = body["data"]["job_id"]
+    job: SupervisorJob | None = None
+    for j in coresys.jobs.jobs:
+        if j.name == "backup_store_addons" and j.parent_id == job_id:
+            job = j
+            break
+
+    assert job
+    assert job.done is True
+    assert job.errors[0].type_ == AddonPrePostBackupCommandReturnedError
+    assert job.errors[0].message == (
+        "Pre-/Post backup command for add-on local_example returned error code: "
+        "1. Please report this to the addon developer. Enable debug "
+        "logging to capture complete command output using ha supervisor options --logging debug"
+    )
+    assert job.errors[0].error_key == "addon_pre_post_backup_command_returned_error"
+    assert job.errors[0].extra_fields == {
+        "addon": "local_example",
+        "exit_code": 1,
+        "debug_logging_command": "ha supervisor options --logging debug",
+    }

tests/api/test_homeassistant.py

@@ -19,7 +19,7 @@ from supervisor.homeassistant.core import HomeAssistantCore
 from supervisor.homeassistant.module import HomeAssistant
 
 from tests.api import common_test_api_advanced_logs
-from tests.common import load_json_fixture
+from tests.common import AsyncIterator, load_json_fixture
 
 
 @pytest.mark.parametrize("legacy_route", [True, False])
@@ -283,9 +283,9 @@ async def test_api_progress_updates_home_assistant_update(
     """Test progress updates sent to Home Assistant for updates."""
     coresys.hardware.disk.get_disk_free_space = lambda x: 5000
     coresys.core.set_state(CoreState.RUNNING)
-    coresys.docker.docker.api.pull.return_value = load_json_fixture(
-        "docker_pull_image_log.json"
-    )
+
+    logs = load_json_fixture("docker_pull_image_log.json")
+    coresys.docker.images.pull.return_value = AsyncIterator(logs)
     coresys.homeassistant.version = AwesomeVersion("2025.8.0")
 
     with (

tests/api/test_jobs.py

@@ -374,6 +374,8 @@ async def test_job_with_error(
                     "type": "SupervisorError",
                     "message": "bad",
                     "stage": "test",
+                    "error_key": None,
+                    "extra_fields": None,
                 }
             ],
             "child_jobs": [
@@ -391,6 +393,8 @@ async def test_job_with_error(
                             "type": "SupervisorError",
                             "message": "bad",
                             "stage": None,
+                            "error_key": None,
+                            "extra_fields": None,
                         }
                     ],
                     "child_jobs": [],

tests/api/test_store.py

@@ -24,7 +24,7 @@ from supervisor.homeassistant.module import HomeAssistant
 from supervisor.store.addon import AddonStore
 from supervisor.store.repository import Repository
 
-from tests.common import load_json_fixture
+from tests.common import AsyncIterator, load_json_fixture
 from tests.const import TEST_ADDON_SLUG
 
 REPO_URL = "https://github.com/awesome-developer/awesome-repo"
@@ -323,7 +323,10 @@ async def test_store_addon_not_found(
     """Test store addon not found error."""
     resp = await api_client.request(method, url)
     assert resp.status == 404
-    assert await get_message(resp, json_expected) == "Addon bad does not exist"
+    assert (
+        await get_message(resp, json_expected)
+        == "Addon bad does not exist in the store"
+    )
 
 
 @pytest.mark.parametrize(
@@ -732,9 +735,10 @@ async def test_api_progress_updates_addon_install_update(
     """Test progress updates sent to Home Assistant for installs/updates."""
     coresys.hardware.disk.get_disk_free_space = lambda x: 5000
     coresys.core.set_state(CoreState.RUNNING)
-    coresys.docker.docker.api.pull.return_value = load_json_fixture(
-        "docker_pull_image_log.json"
-    )
+
+    logs = load_json_fixture("docker_pull_image_log.json")
+    coresys.docker.images.pull.return_value = AsyncIterator(logs)
+
     coresys.arch._supported_arch = ["amd64"]  # pylint: disable=protected-access
     install_addon_example.data_store["version"] = AwesomeVersion("2.0.0")
 

tests/api/test_supervisor.py

@@ -7,6 +7,7 @@ from unittest.mock import AsyncMock, MagicMock, PropertyMock, patch
 from aiohttp.test_utils import TestClient
 from awesomeversion import AwesomeVersion
 from blockbuster import BlockingError
+from docker.errors import DockerException
 import pytest
 
 from supervisor.const import CoreState
@@ -19,7 +20,7 @@ from supervisor.supervisor import Supervisor
 from supervisor.updater import Updater
 
 from tests.api import common_test_api_advanced_logs
-from tests.common import load_json_fixture
+from tests.common import AsyncIterator, load_json_fixture
 from tests.dbus_service_mocks.base import DBusServiceMock
 from tests.dbus_service_mocks.os_agent import OSAgent as OSAgentService
 
@@ -332,9 +333,9 @@ async def test_api_progress_updates_supervisor_update(
     """Test progress updates sent to Home Assistant for updates."""
     coresys.hardware.disk.get_disk_free_space = lambda x: 5000
     coresys.core.set_state(CoreState.RUNNING)
-    coresys.docker.docker.api.pull.return_value = load_json_fixture(
-        "docker_pull_image_log.json"
-    )
+
+    logs = load_json_fixture("docker_pull_image_log.json")
+    coresys.docker.images.pull.return_value = AsyncIterator(logs)
 
     with (
         patch.object(
@@ -417,3 +418,37 @@ async def test_api_progress_updates_supervisor_update(
             "done": True,
         },
     ]
+
+
+async def test_api_supervisor_stats(api_client: TestClient, coresys: CoreSys):
+    """Test supervisor stats."""
+    coresys.docker.containers.get.return_value.status = "running"
+    coresys.docker.containers.get.return_value.stats.return_value = load_json_fixture(
+        "container_stats.json"
+    )
+
+    resp = await api_client.get("/supervisor/stats")
+    assert resp.status == 200
+    result = await resp.json()
+    assert result["data"]["cpu_percent"] == 90.0
+    assert result["data"]["memory_usage"] == 59700000
+    assert result["data"]["memory_limit"] == 4000000000
+    assert result["data"]["memory_percent"] == 1.49
+
+
+async def test_supervisor_api_stats_failure(
+    api_client: TestClient, coresys: CoreSys, caplog: pytest.LogCaptureFixture
+):
+    """Test supervisor stats failure."""
+    coresys.docker.containers.get.side_effect = DockerException("fail")
+
+    resp = await api_client.get("/supervisor/stats")
+    assert resp.status == 500
+    body = await resp.json()
+    assert (
+        body["message"]
+        == "Can't get stats for Supervisor container. Check supervisor logs for details (check with 'ha supervisor logs')"
+    )
+    assert body["error_key"] == "supervisor_stats_error"
+    assert body["extra_fields"] == {"logs_command": "ha supervisor logs"}
+    assert "Could not inspect container 'hassio_supervisor': fail" in caplog.text

tests/common.py

@@ -1,13 +1,14 @@
 """Common test functions."""
 
 import asyncio
+from collections.abc import Sequence
 from datetime import datetime
 from functools import partial
 from importlib import import_module
 from inspect import getclosurevars
 import json
 from pathlib import Path
-from typing import Any
+from typing import Any, Self
 
 from dbus_fast.aio.message_bus import MessageBus
 
@@ -145,3 +146,22 @@ class MockResponse:
 
     async def __aexit__(self, exc_type, exc, tb):
         """Exit the context manager."""
+
+
+class AsyncIterator:
+    """Make list/fixture into async iterator for test mocks."""
+
+    def __init__(self, seq: Sequence[Any]) -> None:
+        """Initialize with sequence."""
+        self.iter = iter(seq)
+
+    def __aiter__(self) -> Self:
+        """Implement aiter."""
+        return self
+
+    async def __anext__(self) -> Any:
+        """Return next in sequence."""
+        try:
+            return next(self.iter)
+        except StopIteration:
+            raise StopAsyncIteration() from None

tests/conftest.py

@@ -9,6 +9,7 @@ import subprocess
 from unittest.mock import AsyncMock, MagicMock, Mock, PropertyMock, patch
 from uuid import uuid4
 
+from aiodocker.docker import DockerImages
 from aiohttp import ClientSession, web
 from aiohttp.test_utils import TestClient
 from awesomeversion import AwesomeVersion
@@ -55,6 +56,7 @@ from supervisor.store.repository import Repository
 from supervisor.utils.dt import utcnow
 
 from .common import (
+    AsyncIterator,
     MockResponse,
     load_binary_fixture,
     load_fixture,
@@ -112,40 +114,46 @@ async def supervisor_name() -> None:
 @pytest.fixture
 async def docker() -> DockerAPI:
     """Mock DockerAPI."""
-    images = [MagicMock(tags=["ghcr.io/home-assistant/amd64-hassio-supervisor:latest"])]
-    image = MagicMock()
-    image.attrs = {"Os": "linux", "Architecture": "amd64"}
+    image_inspect = {
+        "Os": "linux",
+        "Architecture": "amd64",
+        "Id": "test123",
+        "RepoTags": ["ghcr.io/home-assistant/amd64-hassio-supervisor:latest"],
+    }
 
     with (
         patch("supervisor.docker.manager.DockerClient", return_value=MagicMock()),
-        patch("supervisor.docker.manager.DockerAPI.images", return_value=MagicMock()),
         patch(
             "supervisor.docker.manager.DockerAPI.containers", return_value=MagicMock()
         ),
-        patch(
-            "supervisor.docker.manager.DockerAPI.api",
-            return_value=(api_mock := MagicMock()),
-        ),
-        patch("supervisor.docker.manager.DockerAPI.images.get", return_value=image),
-        patch("supervisor.docker.manager.DockerAPI.images.list", return_value=images),
-        patch(
-            "supervisor.docker.manager.DockerAPI.info",
-            return_value=MagicMock(),
-        ),
+        patch("supervisor.docker.manager.DockerAPI.api", return_value=MagicMock()),
+        patch("supervisor.docker.manager.DockerAPI.info", return_value=MagicMock()),
         patch("supervisor.docker.manager.DockerAPI.unload"),
+        patch("supervisor.docker.manager.aiodocker.Docker", return_value=MagicMock()),
+        patch(
+            "supervisor.docker.manager.DockerAPI.images",
+            new=PropertyMock(
+                return_value=(docker_images := MagicMock(spec=DockerImages))
+            ),
+        ),
     ):
         docker_obj = await DockerAPI(MagicMock()).post_init()
         docker_obj.config._data = {"registries": {}}
         with patch("supervisor.docker.monitor.DockerMonitor.load"):
             await docker_obj.load()
 
+        docker_images.inspect.return_value = image_inspect
+        docker_images.list.return_value = [image_inspect]
+        docker_images.import_image.return_value = [
+            {"stream": "Loaded image: test:latest\n"}
+        ]
+
+        docker_images.pull.return_value = AsyncIterator([{}])
+
         docker_obj.info.logging = "journald"
         docker_obj.info.storage = "overlay2"
         docker_obj.info.version = AwesomeVersion("1.0.0")
 
-        # Need an iterable for logs
-        api_mock.pull.return_value = []
-
         yield docker_obj
 
 
@@ -838,11 +846,9 @@ async def container(docker: DockerAPI) -> MagicMock:
     """Mock attrs and status for container on attach."""
     docker.containers.get.return_value = addon = MagicMock()
     docker.containers.create.return_value = addon
-    docker.images.build.return_value = (addon, "")
     addon.status = "stopped"
     addon.attrs = {"State": {"ExitCode": 0}}
-    with patch.object(DockerAPI, "pull_image", return_value=addon):
-        yield addon
+    yield addon
 
 
 @pytest.fixture

tests/docker/test_interface.py

@@ -5,10 +5,10 @@ from pathlib import Path
 from typing import Any
 from unittest.mock import ANY, AsyncMock, MagicMock, Mock, PropertyMock, call, patch
 
+import aiodocker
 from awesomeversion import AwesomeVersion
 from docker.errors import DockerException, NotFound
 from docker.models.containers import Container
-from docker.models.images import Image
 import pytest
 from requests import RequestException
 
@@ -28,7 +28,7 @@ from supervisor.exceptions import (
 )
 from supervisor.jobs import JobSchedulerOptions, SupervisorJob
 
-from tests.common import load_json_fixture
+from tests.common import AsyncIterator, load_json_fixture
 
 
 @pytest.mark.parametrize(
@@ -48,35 +48,30 @@ async def test_docker_image_platform(
     platform: str,
 ):
     """Test platform set correctly from arch."""
-    with patch.object(
-        coresys.docker.images, "get", return_value=Mock(id="test:1.2.3")
-    ) as get:
-        await test_docker_interface.install(
-            AwesomeVersion("1.2.3"), "test", arch=cpu_arch
-        )
-        coresys.docker.docker.api.pull.assert_called_once_with(
-            "test", tag="1.2.3", platform=platform, stream=True, decode=True
-        )
-        get.assert_called_once_with("test:1.2.3")
+    coresys.docker.images.inspect.return_value = {"Id": "test:1.2.3"}
+    await test_docker_interface.install(AwesomeVersion("1.2.3"), "test", arch=cpu_arch)
+    coresys.docker.images.pull.assert_called_once_with(
+        "test", tag="1.2.3", platform=platform, stream=True
+    )
+    coresys.docker.images.inspect.assert_called_once_with("test:1.2.3")
 
 
 async def test_docker_image_default_platform(
     coresys: CoreSys, test_docker_interface: DockerInterface
 ):
     """Test platform set using supervisor arch when omitted."""
+    coresys.docker.images.inspect.return_value = {"Id": "test:1.2.3"}
     with (
         patch.object(
             type(coresys.supervisor), "arch", PropertyMock(return_value="i386")
         ),
-        patch.object(
-            coresys.docker.images, "get", return_value=Mock(id="test:1.2.3")
-        ) as get,
     ):
         await test_docker_interface.install(AwesomeVersion("1.2.3"), "test")
-        coresys.docker.docker.api.pull.assert_called_once_with(
-            "test", tag="1.2.3", platform="linux/386", stream=True, decode=True
+        coresys.docker.images.pull.assert_called_once_with(
+            "test", tag="1.2.3", platform="linux/386", stream=True
         )
-        get.assert_called_once_with("test:1.2.3")
+
+    coresys.docker.images.inspect.assert_called_once_with("test:1.2.3")
 
 
 @pytest.mark.parametrize(
@@ -207,57 +202,40 @@ async def test_attach_existing_container(
 
 async def test_attach_container_failure(coresys: CoreSys):
     """Test attach fails to find container but finds image."""
-    container_collection = MagicMock()
-    container_collection.get.side_effect = DockerException()
-    image_collection = MagicMock()
-    image_config = {"Image": "sha256:abc123"}
-    image_collection.get.return_value = Image({"Config": image_config})
-    with (
-        patch(
-            "supervisor.docker.manager.DockerAPI.containers",
-            new=PropertyMock(return_value=container_collection),
-        ),
-        patch(
-            "supervisor.docker.manager.DockerAPI.images",
-            new=PropertyMock(return_value=image_collection),
-        ),
-        patch.object(type(coresys.bus), "fire_event") as fire_event,
-    ):
+    coresys.docker.containers.get.side_effect = DockerException()
+    coresys.docker.images.inspect.return_value.setdefault("Config", {})["Image"] = (
+        "sha256:abc123"
+    )
+    with patch.object(type(coresys.bus), "fire_event") as fire_event:
         await coresys.homeassistant.core.instance.attach(AwesomeVersion("2022.7.3"))
         assert not [
             event
             for event in fire_event.call_args_list
             if event.args[0] == BusEvent.DOCKER_CONTAINER_STATE_CHANGE
         ]
-        assert coresys.homeassistant.core.instance.meta_config == image_config
+        assert (
+            coresys.homeassistant.core.instance.meta_config["Image"] == "sha256:abc123"
+        )
 
 
 async def test_attach_total_failure(coresys: CoreSys):
     """Test attach fails to find container or image."""
-    container_collection = MagicMock()
-    container_collection.get.side_effect = DockerException()
-    image_collection = MagicMock()
-    image_collection.get.side_effect = DockerException()
-    with (
-        patch(
-            "supervisor.docker.manager.DockerAPI.containers",
-            new=PropertyMock(return_value=container_collection),
-        ),
-        patch(
-            "supervisor.docker.manager.DockerAPI.images",
-            new=PropertyMock(return_value=image_collection),
-        ),
-        pytest.raises(DockerError),
-    ):
+    coresys.docker.containers.get.side_effect = DockerException
+    coresys.docker.images.inspect.side_effect = aiodocker.DockerError(
+        400, {"message": ""}
+    )
+    with pytest.raises(DockerError):
         await coresys.homeassistant.core.instance.attach(AwesomeVersion("2022.7.3"))
 
 
-@pytest.mark.parametrize("err", [DockerException(), RequestException()])
+@pytest.mark.parametrize(
+    "err", [aiodocker.DockerError(400, {"message": ""}), RequestException()]
+)
 async def test_image_pull_fail(
     coresys: CoreSys, capture_exception: Mock, err: Exception
 ):
     """Test failure to pull image."""
-    coresys.docker.images.get.side_effect = err
+    coresys.docker.images.inspect.side_effect = err
     with pytest.raises(DockerError):
         await coresys.homeassistant.core.instance.install(
             AwesomeVersion("2022.7.3"), arch=CpuArch.AMD64
@@ -289,8 +267,9 @@ async def test_install_fires_progress_events(
     coresys: CoreSys, test_docker_interface: DockerInterface
 ):
     """Test progress events are fired during an install for listeners."""
+
     # This is from a sample pull. Filtered log to just one per unique status for test
-    coresys.docker.docker.api.pull.return_value = [
+    logs = [
         {
             "status": "Pulling from home-assistant/odroid-n2-homeassistant",
             "id": "2025.7.2",
@@ -312,7 +291,11 @@ async def test_install_fires_progress_events(
             "id": "1578b14a573c",
         },
         {"status": "Pull complete", "progressDetail": {}, "id": "1578b14a573c"},
-        {"status": "Verifying Checksum", "progressDetail": {}, "id": "6a1e931d8f88"},
+        {
+            "status": "Verifying Checksum",
+            "progressDetail": {},
+            "id": "6a1e931d8f88",
+        },
         {
             "status": "Digest: sha256:490080d7da0f385928022927990e04f604615f7b8c622ef3e58253d0f089881d"
         },
@@ -320,6 +303,7 @@ async def test_install_fires_progress_events(
             "status": "Status: Downloaded newer image for ghcr.io/home-assistant/odroid-n2-homeassistant:2025.7.2"
         },
     ]
+    coresys.docker.images.pull.return_value = AsyncIterator(logs)
 
     events: list[PullLogEntry] = []
 
@@ -334,10 +318,10 @@ async def test_install_fires_progress_events(
         ),
     ):
         await test_docker_interface.install(AwesomeVersion("1.2.3"), "test")
-        coresys.docker.docker.api.pull.assert_called_once_with(
-            "test", tag="1.2.3", platform="linux/386", stream=True, decode=True
+        coresys.docker.images.pull.assert_called_once_with(
+            "test", tag="1.2.3", platform="linux/386", stream=True
         )
-        coresys.docker.images.get.assert_called_once_with("test:1.2.3")
+        coresys.docker.images.inspect.assert_called_once_with("test:1.2.3")
 
     await asyncio.sleep(1)
     assert events == [
@@ -415,10 +399,11 @@ async def test_install_progress_rounding_does_not_cause_misses(
 ):
     """Test extremely close progress events do not create rounding issues."""
     coresys.core.set_state(CoreState.RUNNING)
+
     # Current numbers chosen to create a rounding issue with original code
     # Where a progress update came in with a value between the actual previous
     # value and what it was rounded to. It should not raise an out of order exception
-    coresys.docker.docker.api.pull.return_value = [
+    logs = [
         {
             "status": "Pulling from home-assistant/odroid-n2-homeassistant",
             "id": "2025.7.1",
@@ -458,6 +443,7 @@ async def test_install_progress_rounding_does_not_cause_misses(
             "status": "Status: Downloaded newer image for ghcr.io/home-assistant/odroid-n2-homeassistant:2025.7.1"
         },
     ]
+    coresys.docker.images.pull.return_value = AsyncIterator(logs)
 
     with (
         patch.object(
@@ -513,7 +499,8 @@ async def test_install_raises_on_pull_error(
     exc_msg: str,
 ):
     """Test exceptions raised from errors in pull log."""
-    coresys.docker.docker.api.pull.return_value = [
+
+    logs = [
         {
             "status": "Pulling from home-assistant/odroid-n2-homeassistant",
             "id": "2025.7.2",
@@ -526,6 +513,7 @@ async def test_install_raises_on_pull_error(
         },
         error_log,
     ]
+    coresys.docker.images.pull.return_value = AsyncIterator(logs)
 
     with pytest.raises(exc_type, match=exc_msg):
         await test_docker_interface.install(AwesomeVersion("1.2.3"), "test")
@@ -539,11 +527,11 @@ async def test_install_progress_handles_download_restart(
 ):
     """Test install handles docker progress events that include a download restart."""
     coresys.core.set_state(CoreState.RUNNING)
+
     # Fixture emulates a download restart as it docker logs it
     # A log out of order exception should not be raised
-    coresys.docker.docker.api.pull.return_value = load_json_fixture(
-        "docker_pull_image_log_restart.json"
-    )
+    logs = load_json_fixture("docker_pull_image_log_restart.json")
+    coresys.docker.images.pull.return_value = AsyncIterator(logs)
 
     with (
         patch.object(
@@ -586,7 +574,7 @@ async def test_install_progress_handles_layers_skipping_download(
 
     # Reproduce EXACT sequence from SupervisorNoUpdateProgressLogs.txt:
     # Small layer (02a6e69d8d00) completes BEFORE normal layer (3f4a84073184) starts downloading
-    coresys.docker.docker.api.pull.return_value = [
+    logs = [
         {"status": "Pulling from test/image", "id": "latest"},
         # Small layer that skips downloading (02a6e69d8d00 in logs, 96 bytes)
         {"status": "Pulling fs layer", "progressDetail": {}, "id": "02a6e69d8d00"},
@@ -634,6 +622,7 @@ async def test_install_progress_handles_layers_skipping_download(
         {"status": "Digest: sha256:test"},
         {"status": "Status: Downloaded newer image for test/image:latest"},
     ]
+    coresys.docker.images.pull.return_value = AsyncIterator(logs)
 
     # Capture immutable snapshots of install job progress using job.as_dict()
     # This solves the mutable object problem - we snapshot state at call time

tests/docker/test_manager.py

@@ -1,9 +1,10 @@
 """Test Docker manager."""
 
 import asyncio
+from pathlib import Path
 from unittest.mock import MagicMock, patch
 
-from docker.errors import DockerException
+from docker.errors import APIError, DockerException, NotFound
 import pytest
 from requests import RequestException
 
@@ -20,7 +21,7 @@ async def test_run_command_success(docker: DockerAPI):
     mock_container.logs.return_value = b"command output"
 
     # Mock docker containers.run to return our mock container
-    docker.docker.containers.run.return_value = mock_container
+    docker.dockerpy.containers.run.return_value = mock_container
 
     # Execute the command
     result = docker.run_command(
@@ -33,7 +34,7 @@ async def test_run_command_success(docker: DockerAPI):
     assert result.output == b"command output"
 
     # Verify docker.containers.run was called correctly
-    docker.docker.containers.run.assert_called_once_with(
+    docker.dockerpy.containers.run.assert_called_once_with(
         "alpine:3.18",
         command="echo hello",
         detach=True,
@@ -55,7 +56,7 @@ async def test_run_command_with_defaults(docker: DockerAPI):
     mock_container.logs.return_value = b"error output"
 
     # Mock docker containers.run to return our mock container
-    docker.docker.containers.run.return_value = mock_container
+    docker.dockerpy.containers.run.return_value = mock_container
 
     # Execute the command with minimal parameters
     result = docker.run_command(image="ubuntu")
@@ -66,7 +67,7 @@ async def test_run_command_with_defaults(docker: DockerAPI):
     assert result.output == b"error output"
 
     # Verify docker.containers.run was called with defaults
-    docker.docker.containers.run.assert_called_once_with(
+    docker.dockerpy.containers.run.assert_called_once_with(
         "ubuntu:latest",  # default tag
         command=None,  # default command
         detach=True,
@@ -81,7 +82,7 @@ async def test_run_command_with_defaults(docker: DockerAPI):
 async def test_run_command_docker_exception(docker: DockerAPI):
     """Test command execution when Docker raises an exception."""
     # Mock docker containers.run to raise DockerException
-    docker.docker.containers.run.side_effect = DockerException("Docker error")
+    docker.dockerpy.containers.run.side_effect = DockerException("Docker error")
 
     # Execute the command and expect DockerError
     with pytest.raises(DockerError, match="Can't execute command: Docker error"):
@@ -91,7 +92,7 @@ async def test_run_command_docker_exception(docker: DockerAPI):
 async def test_run_command_request_exception(docker: DockerAPI):
     """Test command execution when requests raises an exception."""
     # Mock docker containers.run to raise RequestException
-    docker.docker.containers.run.side_effect = RequestException("Connection error")
+    docker.dockerpy.containers.run.side_effect = RequestException("Connection error")
 
     # Execute the command and expect DockerError
     with pytest.raises(DockerError, match="Can't execute command: Connection error"):
@@ -104,7 +105,7 @@ async def test_run_command_cleanup_on_exception(docker: DockerAPI):
     mock_container = MagicMock()
 
     # Mock docker.containers.run to return container, but container.wait to raise exception
-    docker.docker.containers.run.return_value = mock_container
+    docker.dockerpy.containers.run.return_value = mock_container
     mock_container.wait.side_effect = DockerException("Wait failed")
 
     # Execute the command and expect DockerError
@@ -123,7 +124,7 @@ async def test_run_command_custom_stdout_stderr(docker: DockerAPI):
     mock_container.logs.return_value = b"output"
 
     # Mock docker containers.run to return our mock container
-    docker.docker.containers.run.return_value = mock_container
+    docker.dockerpy.containers.run.return_value = mock_container
 
     # Execute the command with custom stdout/stderr
     result = docker.run_command(
@@ -150,7 +151,7 @@ async def test_run_container_with_cidfile(
     cidfile_path = coresys.config.path_cid_files / f"{container_name}.cid"
     extern_cidfile_path = coresys.config.path_extern_cid_files / f"{container_name}.cid"
 
-    docker.docker.containers.run.return_value = mock_container
+    docker.dockerpy.containers.run.return_value = mock_container
 
     # Mock container creation
     with patch.object(
@@ -351,3 +352,101 @@ async def test_run_container_with_leftover_cidfile_directory(
         assert cidfile_path.read_text() == mock_container.id
 
         assert result == mock_container
+
+
+async def test_repair(coresys: CoreSys, caplog: pytest.LogCaptureFixture):
+    """Test repair API."""
+    coresys.docker.dockerpy.networks.get.side_effect = [
+        hassio := MagicMock(
+            attrs={
+                "Containers": {
+                    "good": {"Name": "good"},
+                    "corrupt": {"Name": "corrupt"},
+                    "fail": {"Name": "fail"},
+                }
+            }
+        ),
+        host := MagicMock(attrs={"Containers": {}}),
+    ]
+    coresys.docker.dockerpy.containers.get.side_effect = [
+        MagicMock(),
+        NotFound("corrupt"),
+        DockerException("fail"),
+    ]
+
+    await coresys.run_in_executor(coresys.docker.repair)
+
+    coresys.docker.dockerpy.api.prune_containers.assert_called_once()
+    coresys.docker.dockerpy.api.prune_images.assert_called_once_with(
+        filters={"dangling": False}
+    )
+    coresys.docker.dockerpy.api.prune_builds.assert_called_once()
+    coresys.docker.dockerpy.api.prune_volumes.assert_called_once()
+    coresys.docker.dockerpy.api.prune_networks.assert_called_once()
+    hassio.disconnect.assert_called_once_with("corrupt", force=True)
+    host.disconnect.assert_not_called()
+    assert "Docker fatal error on container fail on hassio" in caplog.text
+
+
+async def test_repair_failures(coresys: CoreSys, caplog: pytest.LogCaptureFixture):
+    """Test repair proceeds best it can through failures."""
+    coresys.docker.dockerpy.api.prune_containers.side_effect = APIError("fail")
+    coresys.docker.dockerpy.api.prune_images.side_effect = APIError("fail")
+    coresys.docker.dockerpy.api.prune_builds.side_effect = APIError("fail")
+    coresys.docker.dockerpy.api.prune_volumes.side_effect = APIError("fail")
+    coresys.docker.dockerpy.api.prune_networks.side_effect = APIError("fail")
+    coresys.docker.dockerpy.networks.get.side_effect = NotFound("missing")
+
+    await coresys.run_in_executor(coresys.docker.repair)
+
+    assert "Error for containers prune: fail" in caplog.text
+    assert "Error for images prune: fail" in caplog.text
+    assert "Error for builds prune: fail" in caplog.text
+    assert "Error for volumes prune: fail" in caplog.text
+    assert "Error for networks prune: fail" in caplog.text
+    assert "Error for networks hassio prune: missing" in caplog.text
+    assert "Error for networks host prune: missing" in caplog.text
+
+
+@pytest.mark.parametrize("log_starter", [("Loaded image ID"), ("Loaded image")])
+async def test_import_image(coresys: CoreSys, tmp_path: Path, log_starter: str):
+    """Test importing an image into docker."""
+    (test_tar := tmp_path / "test.tar").touch()
+    coresys.docker.images.import_image.return_value = [
+        {"stream": f"{log_starter}: imported"}
+    ]
+    coresys.docker.images.inspect.return_value = {"Id": "imported"}
+
+    image = await coresys.docker.import_image(test_tar)
+
+    assert image["Id"] == "imported"
+    coresys.docker.images.inspect.assert_called_once_with("imported")
+
+
+async def test_import_image_error(coresys: CoreSys, tmp_path: Path):
+    """Test failure importing an image into docker."""
+    (test_tar := tmp_path / "test.tar").touch()
+    coresys.docker.images.import_image.return_value = [
+        {"errorDetail": {"message": "fail"}}
+    ]
+
+    with pytest.raises(DockerError, match="Can't import image from tar: fail"):
+        await coresys.docker.import_image(test_tar)
+
+    coresys.docker.images.inspect.assert_not_called()
+
+
+async def test_import_multiple_images_in_tar(
+    coresys: CoreSys, tmp_path: Path, caplog: pytest.LogCaptureFixture
+):
+    """Test importing an image into docker."""
+    (test_tar := tmp_path / "test.tar").touch()
+    coresys.docker.images.import_image.return_value = [
+        {"stream": "Loaded image: imported-1"},
+        {"stream": "Loaded image: imported-2"},
+    ]
+
+    assert await coresys.docker.import_image(test_tar) is None
+
+    assert "Unexpected image count 2 while importing image from tar" in caplog.text
+    coresys.docker.images.inspect.assert_not_called()

tests/docker/test_monitor.py

@@ -88,7 +88,7 @@ async def test_events(
 ):
     """Test events created from docker events."""
     event["Actor"]["Attributes"]["name"] = "some_container"
-    event["id"] = "abc123"
+    event["Actor"]["ID"] = "abc123"
     event["time"] = 123
     with (
         patch(
@@ -131,12 +131,12 @@ async def test_unlabeled_container(coresys: CoreSys):
             new=PropertyMock(
                 return_value=[
                     {
-                        "id": "abc123",
                         "time": 123,
                         "Type": "container",
                         "Action": "die",
                         "Actor": {
-                            "Attributes": {"name": "homeassistant", "exitCode": "137"}
+                            "ID": "abc123",
+                            "Attributes": {"name": "homeassistant", "exitCode": "137"},
                         },
                     }
                 ]

tests/homeassistant/test_core.py

@@ -1,11 +1,14 @@
 """Test Home Assistant core."""
 
 from datetime import datetime, timedelta
-from unittest.mock import ANY, MagicMock, Mock, PropertyMock, patch
+from http import HTTPStatus
+from unittest.mock import ANY, MagicMock, Mock, PropertyMock, call, patch
 
+import aiodocker
 from awesomeversion import AwesomeVersion
-from docker.errors import APIError, DockerException, ImageNotFound, NotFound
+from docker.errors import APIError, DockerException, NotFound
 import pytest
+from requests import RequestException
 from time_machine import travel
 
 from supervisor.const import CpuArch
@@ -23,8 +26,12 @@ from supervisor.exceptions import (
 from supervisor.homeassistant.api import APIState
 from supervisor.homeassistant.core import HomeAssistantCore
 from supervisor.homeassistant.module import HomeAssistant
+from supervisor.resolution.const import ContextType, IssueType
+from supervisor.resolution.data import Issue
 from supervisor.updater import Updater
 
+from tests.common import AsyncIterator
+
 
 async def test_update_fails_if_out_of_date(coresys: CoreSys):
     """Test update of Home Assistant fails when supervisor or plugin is out of date."""
@@ -52,11 +59,23 @@ async def test_update_fails_if_out_of_date(coresys: CoreSys):
         await coresys.homeassistant.core.update()
 
 
-async def test_install_landingpage_docker_error(
-    coresys: CoreSys, capture_exception: Mock, caplog: pytest.LogCaptureFixture
+@pytest.mark.parametrize(
+    "err",
+    [
+        aiodocker.DockerError(HTTPStatus.TOO_MANY_REQUESTS, {"message": "ratelimit"}),
+        APIError("ratelimit", MagicMock(status_code=HTTPStatus.TOO_MANY_REQUESTS)),
+    ],
+)
+async def test_install_landingpage_docker_ratelimit_error(
+    coresys: CoreSys,
+    capture_exception: Mock,
+    caplog: pytest.LogCaptureFixture,
+    err: Exception,
 ):
-    """Test install landing page fails due to docker error."""
+    """Test install landing page fails due to docker ratelimit error."""
     coresys.security.force = True
+    coresys.docker.images.pull.side_effect = [err, AsyncIterator([{}])]
+
     with (
         patch.object(DockerHomeAssistant, "attach", side_effect=DockerError),
         patch.object(
@@ -69,19 +88,35 @@ async def test_install_landingpage_docker_error(
         ),
         patch("supervisor.homeassistant.core.asyncio.sleep") as sleep,
     ):
-        coresys.docker.images.get.side_effect = [APIError("fail"), MagicMock()]
         await coresys.homeassistant.core.install_landingpage()
         sleep.assert_awaited_once_with(30)
 
     assert "Failed to install landingpage, retrying after 30sec" in caplog.text
     capture_exception.assert_not_called()
+    assert (
+        Issue(IssueType.DOCKER_RATELIMIT, ContextType.SYSTEM)
+        in coresys.resolution.issues
+    )
 
 
+@pytest.mark.parametrize(
+    "err",
+    [
+        aiodocker.DockerError(HTTPStatus.INTERNAL_SERVER_ERROR, {"message": "fail"}),
+        APIError("fail"),
+        DockerException(),
+        RequestException(),
+        OSError(),
+    ],
+)
 async def test_install_landingpage_other_error(
-    coresys: CoreSys, capture_exception: Mock, caplog: pytest.LogCaptureFixture
+    coresys: CoreSys,
+    capture_exception: Mock,
+    caplog: pytest.LogCaptureFixture,
+    err: Exception,
 ):
     """Test install landing page fails due to other error."""
-    coresys.docker.images.get.side_effect = [(err := OSError()), MagicMock()]
+    coresys.docker.images.inspect.side_effect = [err, MagicMock()]
 
     with (
         patch.object(DockerHomeAssistant, "attach", side_effect=DockerError),
@@ -102,11 +137,23 @@ async def test_install_landingpage_other_error(
     capture_exception.assert_called_once_with(err)
 
 
-async def test_install_docker_error(
-    coresys: CoreSys, capture_exception: Mock, caplog: pytest.LogCaptureFixture
+@pytest.mark.parametrize(
+    "err",
+    [
+        aiodocker.DockerError(HTTPStatus.TOO_MANY_REQUESTS, {"message": "ratelimit"}),
+        APIError("ratelimit", MagicMock(status_code=HTTPStatus.TOO_MANY_REQUESTS)),
+    ],
+)
+async def test_install_docker_ratelimit_error(
+    coresys: CoreSys,
+    capture_exception: Mock,
+    caplog: pytest.LogCaptureFixture,
+    err: Exception,
 ):
-    """Test install fails due to docker error."""
+    """Test install fails due to docker ratelimit error."""
     coresys.security.force = True
+    coresys.docker.images.pull.side_effect = [err, AsyncIterator([{}])]
+
     with (
         patch.object(HomeAssistantCore, "start"),
         patch.object(DockerHomeAssistant, "cleanup"),
@@ -123,19 +170,35 @@ async def test_install_docker_error(
         ),
         patch("supervisor.homeassistant.core.asyncio.sleep") as sleep,
     ):
-        coresys.docker.images.get.side_effect = [APIError("fail"), MagicMock()]
         await coresys.homeassistant.core.install()
         sleep.assert_awaited_once_with(30)
 
     assert "Error on Home Assistant installation. Retrying in 30sec" in caplog.text
     capture_exception.assert_not_called()
+    assert (
+        Issue(IssueType.DOCKER_RATELIMIT, ContextType.SYSTEM)
+        in coresys.resolution.issues
+    )
 
 
+@pytest.mark.parametrize(
+    "err",
+    [
+        aiodocker.DockerError(HTTPStatus.INTERNAL_SERVER_ERROR, {"message": "fail"}),
+        APIError("fail"),
+        DockerException(),
+        RequestException(),
+        OSError(),
+    ],
+)
 async def test_install_other_error(
-    coresys: CoreSys, capture_exception: Mock, caplog: pytest.LogCaptureFixture
+    coresys: CoreSys,
+    capture_exception: Mock,
+    caplog: pytest.LogCaptureFixture,
+    err: Exception,
 ):
     """Test install fails due to other error."""
-    coresys.docker.images.get.side_effect = [(err := OSError()), MagicMock()]
+    coresys.docker.images.inspect.side_effect = [err, MagicMock()]
 
     with (
         patch.object(HomeAssistantCore, "start"),
@@ -161,21 +224,29 @@ async def test_install_other_error(
 
 
 @pytest.mark.parametrize(
-    "container_exists,image_exists", [(False, True), (True, False), (True, True)]
+    ("container_exc", "image_exc", "remove_calls"),
+    [
+        (NotFound("missing"), None, []),
+        (
+            None,
+            aiodocker.DockerError(404, {"message": "missing"}),
+            [call(force=True, v=True)],
+        ),
+        (None, None, [call(force=True, v=True)]),
+    ],
 )
+@pytest.mark.usefixtures("path_extern")
 async def test_start(
-    coresys: CoreSys, container_exists: bool, image_exists: bool, path_extern
+    coresys: CoreSys,
+    container_exc: DockerException | None,
+    image_exc: aiodocker.DockerError | None,
+    remove_calls: list[call],
 ):
     """Test starting Home Assistant."""
-    if image_exists:
-        coresys.docker.images.get.return_value.id = "123"
-    else:
-        coresys.docker.images.get.side_effect = ImageNotFound("missing")
-
-    if container_exists:
-        coresys.docker.containers.get.return_value.image.id = "123"
-    else:
-        coresys.docker.containers.get.side_effect = NotFound("missing")
+    coresys.docker.images.inspect.return_value = {"Id": "123"}
+    coresys.docker.images.inspect.side_effect = image_exc
+    coresys.docker.containers.get.return_value.id = "123"
+    coresys.docker.containers.get.side_effect = container_exc
 
     with (
         patch.object(
@@ -198,18 +269,14 @@ async def test_start(
         assert run.call_args.kwargs["hostname"] == "homeassistant"
 
     coresys.docker.containers.get.return_value.stop.assert_not_called()
-    if container_exists:
-        coresys.docker.containers.get.return_value.remove.assert_called_once_with(
-            force=True,
-            v=True,
-        )
-    else:
-        coresys.docker.containers.get.return_value.remove.assert_not_called()
+    assert (
+        coresys.docker.containers.get.return_value.remove.call_args_list == remove_calls
+    )
 
 
 async def test_start_existing_container(coresys: CoreSys, path_extern):
     """Test starting Home Assistant when container exists and is viable."""
-    coresys.docker.images.get.return_value.id = "123"
+    coresys.docker.images.inspect.return_value = {"Id": "123"}
     coresys.docker.containers.get.return_value.image.id = "123"
     coresys.docker.containers.get.return_value.status = "exited"
 
@@ -394,24 +461,32 @@ async def test_core_loads_wrong_image_for_machine(
     """Test core is loaded with wrong image for machine."""
     coresys.homeassistant.set_image("ghcr.io/home-assistant/odroid-n2-homeassistant")
     coresys.homeassistant.version = AwesomeVersion("2024.4.0")
-    container.attrs["Config"] = {"Labels": {"io.hass.version": "2024.4.0"}}
 
-    await coresys.homeassistant.core.load()
+    with patch.object(
+        DockerAPI,
+        "pull_image",
+        return_value={
+            "Id": "abc123",
+            "Config": {"Labels": {"io.hass.version": "2024.4.0"}},
+        },
+    ) as pull_image:
+        container.attrs |= pull_image.return_value
+        await coresys.homeassistant.core.load()
+        pull_image.assert_called_once_with(
+            ANY,
+            "ghcr.io/home-assistant/qemux86-64-homeassistant",
+            "2024.4.0",
+            platform="linux/amd64",
+        )
 
     container.remove.assert_called_once_with(force=True, v=True)
-    assert coresys.docker.images.remove.call_args_list[0].kwargs == {
-        "image": "ghcr.io/home-assistant/odroid-n2-homeassistant:latest",
-        "force": True,
-    }
-    assert coresys.docker.images.remove.call_args_list[1].kwargs == {
-        "image": "ghcr.io/home-assistant/odroid-n2-homeassistant:2024.4.0",
-        "force": True,
-    }
-    coresys.docker.pull_image.assert_called_once_with(
-        ANY,
-        "ghcr.io/home-assistant/qemux86-64-homeassistant",
-        "2024.4.0",
-        platform="linux/amd64",
+    assert coresys.docker.images.delete.call_args_list[0] == call(
+        "ghcr.io/home-assistant/odroid-n2-homeassistant:latest",
+        force=True,
+    )
+    assert coresys.docker.images.delete.call_args_list[1] == call(
+        "ghcr.io/home-assistant/odroid-n2-homeassistant:2024.4.0",
+        force=True,
     )
     assert (
         coresys.homeassistant.image == "ghcr.io/home-assistant/qemux86-64-homeassistant"
@@ -428,8 +503,8 @@ async def test_core_load_allows_image_override(coresys: CoreSys, container: Magi
     await coresys.homeassistant.core.load()
 
     container.remove.assert_not_called()
-    coresys.docker.images.remove.assert_not_called()
-    coresys.docker.images.get.assert_not_called()
+    coresys.docker.images.delete.assert_not_called()
+    coresys.docker.images.inspect.assert_not_called()
     assert (
         coresys.homeassistant.image == "ghcr.io/home-assistant/odroid-n2-homeassistant"
     )
@@ -440,27 +515,36 @@ async def test_core_loads_wrong_image_for_architecture(
 ):
     """Test core is loaded with wrong image for architecture."""
     coresys.homeassistant.version = AwesomeVersion("2024.4.0")
-    container.attrs["Config"] = {"Labels": {"io.hass.version": "2024.4.0"}}
-    coresys.docker.images.get("ghcr.io/home-assistant/qemux86-64-homeassistant").attrs[
-        "Architecture"
-    ] = "arm64"
+    coresys.docker.images.inspect.return_value = img_data = (
+        coresys.docker.images.inspect.return_value
+        | {
+            "Architecture": "arm64",
+            "Config": {"Labels": {"io.hass.version": "2024.4.0"}},
+        }
+    )
+    container.attrs |= img_data
 
-    await coresys.homeassistant.core.load()
+    with patch.object(
+        DockerAPI,
+        "pull_image",
+        return_value=img_data | {"Architecture": "amd64"},
+    ) as pull_image:
+        await coresys.homeassistant.core.load()
+        pull_image.assert_called_once_with(
+            ANY,
+            "ghcr.io/home-assistant/qemux86-64-homeassistant",
+            "2024.4.0",
+            platform="linux/amd64",
+        )
 
     container.remove.assert_called_once_with(force=True, v=True)
-    assert coresys.docker.images.remove.call_args_list[0].kwargs == {
-        "image": "ghcr.io/home-assistant/qemux86-64-homeassistant:latest",
-        "force": True,
-    }
-    assert coresys.docker.images.remove.call_args_list[1].kwargs == {
-        "image": "ghcr.io/home-assistant/qemux86-64-homeassistant:2024.4.0",
-        "force": True,
-    }
-    coresys.docker.pull_image.assert_called_once_with(
-        ANY,
-        "ghcr.io/home-assistant/qemux86-64-homeassistant",
-        "2024.4.0",
-        platform="linux/amd64",
+    assert coresys.docker.images.delete.call_args_list[0] == call(
+        "ghcr.io/home-assistant/qemux86-64-homeassistant:latest",
+        force=True,
+    )
+    assert coresys.docker.images.delete.call_args_list[1] == call(
+        "ghcr.io/home-assistant/qemux86-64-homeassistant:2024.4.0",
+        force=True,
     )
     assert (
         coresys.homeassistant.image == "ghcr.io/home-assistant/qemux86-64-homeassistant"

tests/jobs/test_job_manager.py

@@ -200,6 +200,8 @@ async def test_notify_on_change(coresys: CoreSys, ha_ws_client: AsyncMock):
                                 "type": "HassioError",
                                 "message": "Unknown error, see Supervisor logs (check with 'ha supervisor logs')",
                                 "stage": "test",
+                                "error_key": None,
+                                "extra_fields": None,
                             }
                         ],
                         "created": ANY,
@@ -228,6 +230,8 @@ async def test_notify_on_change(coresys: CoreSys, ha_ws_client: AsyncMock):
                             "type": "HassioError",
                             "message": "Unknown error, see Supervisor logs (check with 'ha supervisor logs')",
                             "stage": "test",
+                            "error_key": None,
+                            "extra_fields": None,
                         }
                     ],
                     "created": ANY,

tests/misc/test_filter_data.py

@@ -115,7 +115,17 @@ async def test_not_started(coresys):
     assert filter_data(coresys, SAMPLE_EVENT, {}) == SAMPLE_EVENT
 
     await coresys.core.set_state(CoreState.SETUP)
-    assert filter_data(coresys, SAMPLE_EVENT, {}) == SAMPLE_EVENT
+    filtered = filter_data(coresys, SAMPLE_EVENT, {})
+    # During SETUP, we should have basic system info available
+    assert "contexts" in filtered
+    assert "versions" in filtered["contexts"]
+    assert "docker" in filtered["contexts"]["versions"]
+    assert "supervisor" in filtered["contexts"]["versions"]
+    assert "host" in filtered["contexts"]
+    assert "machine" in filtered["contexts"]["host"]
+    assert filtered["contexts"]["versions"]["docker"] == coresys.docker.info.version
+    assert filtered["contexts"]["versions"]["supervisor"] == coresys.supervisor.version
+    assert filtered["contexts"]["host"]["machine"] == coresys.machine
 
 
 async def test_defaults(coresys):

tests/plugins/test_plugin_base.py

@@ -2,7 +2,7 @@
 
 import asyncio
 from pathlib import Path
-from unittest.mock import ANY, MagicMock, Mock, PropertyMock, patch
+from unittest.mock import ANY, MagicMock, Mock, PropertyMock, call, patch
 
 from awesomeversion import AwesomeVersion
 import pytest
@@ -11,6 +11,7 @@ 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.manager import DockerAPI
 from supervisor.docker.monitor import DockerContainerStateEvent
 from supervisor.exceptions import (
     AudioError,
@@ -359,21 +360,26 @@ async def test_load_with_incorrect_image(
     plugin.version = AwesomeVersion("2024.4.0")
 
     container.status = "running"
-    container.attrs["Config"] = {"Labels": {"io.hass.version": "2024.4.0"}}
+    coresys.docker.images.inspect.return_value = img_data = (
+        coresys.docker.images.inspect.return_value
+        | {"Config": {"Labels": {"io.hass.version": "2024.4.0"}}}
+    )
+    container.attrs |= img_data
 
-    await plugin.load()
+    with patch.object(DockerAPI, "pull_image", return_value=img_data) as pull_image:
+        await plugin.load()
+        pull_image.assert_called_once_with(
+            ANY, correct_image, "2024.4.0", platform="linux/amd64"
+        )
 
     container.remove.assert_called_once_with(force=True, v=True)
-    assert coresys.docker.images.remove.call_args_list[0].kwargs == {
-        "image": f"{old_image}:latest",
-        "force": True,
-    }
-    assert coresys.docker.images.remove.call_args_list[1].kwargs == {
-        "image": f"{old_image}:2024.4.0",
-        "force": True,
-    }
-    coresys.docker.pull_image.assert_called_once_with(
-        ANY, correct_image, "2024.4.0", platform="linux/amd64"
+    assert coresys.docker.images.delete.call_args_list[0] == call(
+        f"{old_image}:latest",
+        force=True,
+    )
+    assert coresys.docker.images.delete.call_args_list[1] == call(
+        f"{old_image}:2024.4.0",
+        force=True,
     )
     assert plugin.image == correct_image
 

tests/resolution/fixup/test_addon_execute_repair.py

@@ -1,8 +1,9 @@
 """Test fixup addon execute repair."""
 
-from unittest.mock import MagicMock, patch
+from http import HTTPStatus
+from unittest.mock import patch
 
-from docker.errors import NotFound
+import aiodocker
 import pytest
 
 from supervisor.addons.addon import Addon
@@ -17,7 +18,9 @@ from supervisor.resolution.fixups.addon_execute_repair import FixupAddonExecuteR
 
 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")
+    docker.images.inspect.side_effect = aiodocker.DockerError(
+        HTTPStatus.NOT_FOUND, {"message": "missing"}
+    )
     install_addon_ssh.data["image"] = "test_image"
 
     addon_execute_repair = FixupAddonExecuteRepair(coresys)
@@ -41,7 +44,9 @@ async def test_fixup_max_auto_attempts(
     docker: DockerAPI, coresys: CoreSys, install_addon_ssh: Addon
 ):
     """Test fixup stops being auto-applied after 5 failures."""
-    docker.images.get.side_effect = NotFound("missing")
+    docker.images.inspect.side_effect = aiodocker.DockerError(
+        HTTPStatus.NOT_FOUND, {"message": "missing"}
+    )
     install_addon_ssh.data["image"] = "test_image"
 
     addon_execute_repair = FixupAddonExecuteRepair(coresys)
@@ -82,8 +87,6 @@ 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