Significantly speed up creating backups with isal via zlib-fast

isal is a drop in replacement for zlib with the
cavet that the compression level mappings are different.
zlib-fast is a tiny piece of middleware to convert
the standard zlib compression levels to isal compression
levels to allow for drop-in replacement

https://github.com/bdraco/zlib-fast/releases/tag/v0.1.0
https://github.com/pycompression/python-isal

Compression for backups is ~5x faster than the baseline

https://github.com/powturbo/TurboBench/issues/43

.devcontainer/devcontainer.json

@@ -7,34 +7,32 @@
   "appPort": ["9123:8123", "7357:4357"],
   "postCreateCommand": "bash devcontainer_bootstrap",
   "runArgs": ["-e", "GIT_EDITOR=code --wait", "--privileged"],
-  "extensions": [
-    "ms-python.python",
-    "ms-python.vscode-pylance",
-    "visualstudioexptteam.vscodeintellicode",
-    "esbenp.prettier-vscode"
-  ],
-  "mounts": ["type=volume,target=/var/lib/docker"],
-  "settings": {
-    "terminal.integrated.profiles.linux": {
-      "zsh": {
-        "path": "/usr/bin/zsh"
+  "customizations": {
+    "vscode": {
+      "extensions": [
+        "ms-python.python",
+        "ms-python.pylint",
+        "ms-python.vscode-pylance",
+        "visualstudioexptteam.vscodeintellicode",
+        "esbenp.prettier-vscode"
+      ],
+      "settings": {
+        "terminal.integrated.profiles.linux": {
+          "zsh": {
+            "path": "/usr/bin/zsh"
+          }
+        },
+        "terminal.integrated.defaultProfile.linux": "zsh",
+        "editor.formatOnPaste": false,
+        "editor.formatOnSave": true,
+        "editor.formatOnType": true,
+        "files.trimTrailingWhitespace": true,
+        "python.pythonPath": "/usr/local/bin/python3",
+        "python.formatting.provider": "black",
+        "python.formatting.blackArgs": ["--target-version", "py312"],
+        "python.formatting.blackPath": "/usr/local/bin/black"
       }
-    },
-    "terminal.integrated.defaultProfile.linux": "zsh",
-    "editor.formatOnPaste": false,
-    "editor.formatOnSave": true,
-    "editor.formatOnType": true,
-    "files.trimTrailingWhitespace": true,
-    "python.pythonPath": "/usr/local/bin/python3",
-    "python.linting.pylintEnabled": true,
-    "python.linting.enabled": true,
-    "python.formatting.provider": "black",
-    "python.formatting.blackArgs": ["--target-version", "py310"],
-    "python.formatting.blackPath": "/usr/local/bin/black",
-    "python.linting.banditPath": "/usr/local/bin/bandit",
-    "python.linting.flake8Path": "/usr/local/bin/flake8",
-    "python.linting.mypyPath": "/usr/local/bin/mypy",
-    "python.linting.pylintPath": "/usr/local/bin/pylint",
-    "python.linting.pydocstylePath": "/usr/local/bin/pydocstyle"
-  }
+    }
+  },
+  "mounts": ["type=volume,target=/var/lib/docker"]
 }

.github/workflows/builder.yml

@@ -33,7 +33,7 @@ on:
       - setup.py
 
 env:
-  DEFAULT_PYTHON: "3.11"
+  DEFAULT_PYTHON: "3.12"
   BUILD_NAME: supervisor
   BUILD_TYPE: supervisor
 
@@ -53,7 +53,7 @@ jobs:
       requirements: ${{ steps.requirements.outputs.changed }}
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
         with:
           fetch-depth: 0
 
@@ -70,13 +70,13 @@ jobs:
       - name: Get changed files
         id: changed_files
         if: steps.version.outputs.publish == 'false'
-        uses: jitterbit/get-changed-files@v1
+        uses: masesgroup/retrieve-changed-files@v3.0.0
 
       - name: Check if requirements files changed
         id: requirements
         run: |
-          if [[ "${{ steps.changed_files.outputs.all }}" =~ (requirements.txt|build.json) ]]; then
-            echo "::set-output name=changed::true"
+          if [[ "${{ steps.changed_files.outputs.all }}" =~ (requirements.txt|build.yaml) ]]; then
+            echo "changed=true" >> "$GITHUB_OUTPUT"
           fi
 
   build:
@@ -92,7 +92,7 @@ jobs:
         arch: ${{ fromJson(needs.init.outputs.architectures) }}
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
         with:
           fetch-depth: 0
 
@@ -106,13 +106,13 @@ jobs:
 
       - name: Build wheels
         if: needs.init.outputs.requirements == 'true'
-        uses: home-assistant/wheels@2023.04.0
+        uses: home-assistant/wheels@2024.01.0
         with:
-          abi: cp311
+          abi: cp312
           tag: musllinux_1_2
           arch: ${{ matrix.arch }}
           wheels-key: ${{ secrets.WHEELS_KEY }}
-          apk: "libffi-dev;openssl-dev"
+          apk: "libffi-dev;openssl-dev;yaml-dev"
           skip-binary: aiohttp
           env-file: true
           requirements: "requirements.txt"
@@ -125,20 +125,20 @@ jobs:
 
       - name: Set up Python ${{ env.DEFAULT_PYTHON }}
         if: needs.init.outputs.publish == 'true'
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         with:
           python-version: ${{ env.DEFAULT_PYTHON }}
 
       - name: Install Cosign
         if: needs.init.outputs.publish == 'true'
-        uses: sigstore/cosign-installer@v3.1.1
+        uses: sigstore/cosign-installer@v3.3.0
         with:
           cosign-release: "v2.0.2"
 
       - name: Install dirhash and calc hash
         if: needs.init.outputs.publish == 'true'
         run: |
-          pip3 install dirhash
+          pip3 install setuptools dirhash
           dir_hash="$(dirhash "${{ github.workspace }}/supervisor" -a sha256 --match "*.py")"
           echo "${dir_hash}" > rootfs/supervisor.sha256
 
@@ -149,7 +149,7 @@ jobs:
 
       - name: Login to GitHub Container Registry
         if: needs.init.outputs.publish == 'true'
-        uses: docker/login-action@v2.2.0
+        uses: docker/login-action@v3.0.0
         with:
           registry: ghcr.io
           username: ${{ github.repository_owner }}
@@ -160,7 +160,7 @@ jobs:
         run: echo "BUILD_ARGS=--test" >> $GITHUB_ENV
 
       - name: Build supervisor
-        uses: home-assistant/builder@2023.06.1
+        uses: home-assistant/builder@2024.01.0
         with:
           args: |
             $BUILD_ARGS \
@@ -178,7 +178,7 @@ jobs:
     steps:
       - name: Checkout the repository
         if: needs.init.outputs.publish == 'true'
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
 
       - name: Initialize git
         if: needs.init.outputs.publish == 'true'
@@ -203,11 +203,11 @@ jobs:
     timeout-minutes: 60
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
 
       - name: Build the Supervisor
         if: needs.init.outputs.publish != 'true'
-        uses: home-assistant/builder@2023.06.1
+        uses: home-assistant/builder@2024.01.0
         with:
           args: |
             --test \
@@ -324,7 +324,7 @@ jobs:
           if [ "$(echo $test | jq -r '.result')" != "ok" ]; then
             exit 1
           fi
-          echo "::set-output name=slug::$(echo $test | jq -r '.data.slug')"
+          echo "slug=$(echo $test | jq -r '.data.slug')" >> "$GITHUB_OUTPUT"
 
       - name: Uninstall SSH add-on
         run: |

.github/workflows/ci.yaml

@@ -8,8 +8,8 @@ on:
   pull_request: ~
 
 env:
-  DEFAULT_PYTHON: "3.11"
-  PRE_COMMIT_HOME: ~/.cache/pre-commit
+  DEFAULT_PYTHON: "3.12"
+  PRE_COMMIT_CACHE: ~/.cache/pre-commit
 
 concurrency:
   group: "${{ github.workflow }}-${{ github.ref }}"
@@ -25,15 +25,15 @@ jobs:
     name: Prepare Python dependencies
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python
         id: python
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         with:
           python-version: ${{ env.DEFAULT_PYTHON }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -47,9 +47,10 @@ jobs:
           pip install -r requirements.txt -r requirements_tests.txt
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
+          lookup-only: true
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
           restore-keys: |
@@ -66,15 +67,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -87,7 +88,7 @@ jobs:
       - name: Run black
         run: |
           . venv/bin/activate
-          black --target-version py38 --check supervisor tests setup.py
+          black --target-version py312 --check supervisor tests setup.py
 
   lint-dockerfile:
     name: Check Dockerfile
@@ -95,7 +96,7 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Register hadolint problem matcher
         run: |
           echo "::add-matcher::.github/workflows/matchers/hadolint.json"
@@ -110,15 +111,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -130,9 +131,9 @@ jobs:
           exit 1
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
       - name: Fail job if cache restore failed
@@ -154,15 +155,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -186,15 +187,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -206,9 +207,9 @@ jobs:
           exit 1
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
       - name: Fail job if cache restore failed
@@ -227,15 +228,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -247,9 +248,9 @@ jobs:
           exit 1
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
       - name: Fail job if cache restore failed
@@ -271,15 +272,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -303,15 +304,15 @@ jobs:
     needs: prepare
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -323,9 +324,9 @@ jobs:
           exit 1
       - name: Restore pre-commit environment from cache
         id: cache-precommit
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
-          path: ${{ env.PRE_COMMIT_HOME }}
+          path: ${{ env.PRE_COMMIT_CACHE }}
           key: |
             ${{ runner.os }}-pre-commit-${{ hashFiles('.pre-commit-config.yaml') }}
       - name: Fail job if cache restore failed
@@ -344,19 +345,19 @@ jobs:
     name: Run tests Python ${{ needs.prepare.outputs.python-version }}
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Install Cosign
-        uses: sigstore/cosign-installer@v3.1.1
+        uses: sigstore/cosign-installer@v3.3.0
         with:
           cosign-release: "v2.0.2"
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -391,7 +392,7 @@ jobs:
             -o console_output_style=count \
             tests
       - name: Upload coverage artifact
-        uses: actions/upload-artifact@v3.1.2
+        uses: actions/upload-artifact@v4.0.0
         with:
           name: coverage-${{ matrix.python-version }}
           path: .coverage
@@ -402,15 +403,15 @@ jobs:
     needs: ["pytest", "prepare"]
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Set up Python ${{ needs.prepare.outputs.python-version }}
-        uses: actions/setup-python@v4.6.1
+        uses: actions/setup-python@v5.0.0
         id: python
         with:
           python-version: ${{ needs.prepare.outputs.python-version }}
       - name: Restore Python virtual environment
         id: cache-venv
-        uses: actions/cache@v3.3.1
+        uses: actions/cache@v3.3.3
         with:
           path: venv
           key: |
@@ -421,7 +422,7 @@ jobs:
           echo "Failed to restore Python virtual environment from cache"
           exit 1
       - name: Download all coverage artifacts
-        uses: actions/download-artifact@v3
+        uses: actions/download-artifact@v4.1.1
       - name: Combine coverage results
         run: |
           . venv/bin/activate

.github/workflows/lock.yml

@@ -9,7 +9,7 @@ jobs:
   lock:
     runs-on: ubuntu-latest
     steps:
-      - uses: dessant/lock-threads@v4.0.1
+      - uses: dessant/lock-threads@v5.0.1
         with:
           github-token: ${{ github.token }}
           issue-inactive-days: "30"

.github/workflows/release-drafter.yml

@@ -11,7 +11,7 @@ jobs:
     name: Release Drafter
     steps:
       - name: Checkout the repository
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
         with:
           fetch-depth: 0
 
@@ -33,10 +33,10 @@ jobs:
 
           echo Current version:    $latest
           echo New target version: $datepre.$newpost
-          echo "::set-output name=version::$datepre.$newpost"
+          echo "version=$datepre.$newpost" >> "$GITHUB_OUTPUT"
 
       - name: Run Release Drafter
-        uses: release-drafter/release-drafter@v5.24.0
+        uses: release-drafter/release-drafter@v5.25.0
         with:
           tag: ${{ steps.version.outputs.version }}
           name: ${{ steps.version.outputs.version }}

.github/workflows/sentry.yaml

@@ -10,9 +10,9 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Check out code from GitHub
-        uses: actions/checkout@v3.5.3
+        uses: actions/checkout@v4.1.1
       - name: Sentry Release
-        uses: getsentry/action-release@v1.4.1
+        uses: getsentry/action-release@v1.6.0
         env:
           SENTRY_AUTH_TOKEN: ${{ secrets.SENTRY_AUTH_TOKEN }}
           SENTRY_ORG: ${{ secrets.SENTRY_ORG }}

.github/workflows/stale.yml

@@ -9,7 +9,7 @@ jobs:
   stale:
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/stale@v8.0.0
+      - uses: actions/stale@v9.0.0
         with:
           repo-token: ${{ secrets.GITHUB_TOKEN }}
           days-before-stale: 30

.hadolint.yaml

@@ -3,4 +3,5 @@ ignored:
   - DL3006
   - DL3013
   - DL3018
+  - DL3042
   - SC2155

.pre-commit-config.yaml

@@ -1,16 +1,16 @@
 repos:
   - repo: https://github.com/psf/black
-    rev: 23.1.0
+    rev: 23.12.1
     hooks:
       - id: black
         args:
           - --safe
           - --quiet
           - --target-version
-          - py310
+          - py312
         files: ^((supervisor|tests)/.+)?[^/]+\.py$
   - repo: https://github.com/PyCQA/flake8
-    rev: 6.0.0
+    rev: 7.0.0
     hooks:
       - id: flake8
         additional_dependencies:
@@ -18,17 +18,17 @@ repos:
           - pydocstyle==6.3.0
         files: ^(supervisor|script|tests)/.+\.py$
   - repo: https://github.com/pre-commit/pre-commit-hooks
-    rev: v4.3.0
+    rev: v4.5.0
     hooks:
       - id: check-executables-have-shebangs
         stages: [manual]
       - id: check-json
   - repo: https://github.com/PyCQA/isort
-    rev: 5.12.0
+    rev: 5.13.2
     hooks:
       - id: isort
   - repo: https://github.com/asottile/pyupgrade
-    rev: v3.4.0
+    rev: v3.15.0
     hooks:
       - id: pyupgrade
-        args: [--py310-plus]
+        args: [--py312-plus]

Dockerfile

@@ -15,6 +15,7 @@ WORKDIR /usr/src
 RUN \
     set -x \
     && apk add --no-cache \
+        findutils \
         eudev \
         eudev-libs \
         git \
@@ -22,6 +23,7 @@ RUN \
         libpulse \
         musl \
         openssl \
+        yaml \
     \
     && curl -Lso /usr/bin/cosign "https://github.com/home-assistant/cosign/releases/download/${COSIGN_VERSION}/cosign_${BUILD_ARCH}" \
     && chmod a+x /usr/bin/cosign
@@ -30,15 +32,14 @@ RUN \
 COPY requirements.txt .
 RUN \
     export MAKEFLAGS="-j$(nproc)" \
-    && pip3 install --no-cache-dir --no-index --only-binary=:all: --find-links \
-        "https://wheels.home-assistant.io/musllinux/" \
+    && pip3 install --only-binary=:all: \
         -r ./requirements.txt \
     && rm -f requirements.txt
 
 # Install Home Assistant Supervisor
 COPY . supervisor
 RUN \
-    pip3 install --no-cache-dir -e ./supervisor \
+    pip3 install -e ./supervisor \
     && python3 -m compileall ./supervisor/supervisor
 
 

build.yaml

@@ -1,10 +1,10 @@
 image: ghcr.io/home-assistant/{arch}-hassio-supervisor
 build_from:
-  aarch64: ghcr.io/home-assistant/aarch64-base-python:3.11-alpine3.16
-  armhf: ghcr.io/home-assistant/armhf-base-python:3.11-alpine3.16
-  armv7: ghcr.io/home-assistant/armv7-base-python:3.11-alpine3.16
-  amd64: ghcr.io/home-assistant/amd64-base-python:3.11-alpine3.16
-  i386: ghcr.io/home-assistant/i386-base-python:3.11-alpine3.16
+  aarch64: ghcr.io/home-assistant/aarch64-base-python:3.12-alpine3.18
+  armhf: ghcr.io/home-assistant/armhf-base-python:3.12-alpine3.18
+  armv7: ghcr.io/home-assistant/armv7-base-python:3.12-alpine3.18
+  amd64: ghcr.io/home-assistant/amd64-base-python:3.12-alpine3.18
+  i386: ghcr.io/home-assistant/i386-base-python:3.12-alpine3.18
 codenotary:
   signer: notary@home-assistant.io
   base_image: notary@home-assistant.io

pylintrc

@@ -1,45 +0,0 @@
-[MASTER]
-reports=no
-jobs=2
-
-good-names=id,i,j,k,ex,Run,_,fp,T,os
-
-extension-pkg-whitelist=
-  ciso8601
-
-# Reasons disabled:
-# format - handled by black
-# locally-disabled - it spams too much
-# duplicate-code - unavoidable
-# cyclic-import - doesn't test if both import on load
-# abstract-class-not-used - is flaky, should not show up but does
-# unused-argument - generic callbacks and setup methods create a lot of warnings
-# too-many-* - are not enforced for the sake of readability
-# too-few-* - same as too-many-*
-# abstract-method - with intro of async there are always methods missing
-disable=
-  format,
-  abstract-method,
-  cyclic-import,
-  duplicate-code,
-  locally-disabled,
-  no-else-return,
-  not-context-manager,
-  too-few-public-methods,
-  too-many-arguments,
-  too-many-branches,
-  too-many-instance-attributes,
-  too-many-lines,
-  too-many-locals,
-  too-many-public-methods,
-  too-many-return-statements,
-  too-many-statements,
-  unused-argument,
-  consider-using-with
-
-[EXCEPTIONS]
-overgeneral-exceptions=builtins.Exception
-
-
-[TYPECHECK]
-ignored-modules = distutils

pyproject.toml

@@ -0,0 +1,112 @@
+[build-system]
+requires = ["setuptools~=68.0.0", "wheel~=0.40.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "Supervisor"
+dynamic = ["version", "dependencies"]
+license = { text = "Apache-2.0" }
+description = "Open-source private cloud os for Home-Assistant based on HassOS"
+readme = "README.md"
+authors = [
+    { name = "The Home Assistant Authors", email = "hello@home-assistant.io" },
+]
+keywords = ["docker", "home-assistant", "api"]
+requires-python = ">=3.12.0"
+
+[project.urls]
+"Homepage" = "https://www.home-assistant.io/"
+"Source Code" = "https://github.com/home-assistant/supervisor"
+"Bug Reports" = "https://github.com/home-assistant/supervisor/issues"
+"Docs: Dev" = "https://developers.home-assistant.io/"
+"Discord" = "https://www.home-assistant.io/join-chat/"
+"Forum" = "https://community.home-assistant.io/"
+
+[tool.setuptools]
+platforms = ["any"]
+zip-safe = false
+include-package-data = true
+
+[tool.setuptools.packages.find]
+include = ["supervisor*"]
+
+[tool.pylint.MAIN]
+py-version = "3.11"
+# Use a conservative default here; 2 should speed up most setups and not hurt
+# any too bad. Override on command line as appropriate.
+jobs = 2
+persistent = false
+extension-pkg-allow-list = ["ciso8601"]
+
+[tool.pylint.BASIC]
+class-const-naming-style = "any"
+good-names = ["id", "i", "j", "k", "ex", "Run", "_", "fp", "T", "os"]
+
+[tool.pylint."MESSAGES CONTROL"]
+# Reasons disabled:
+# format - handled by black
+# abstract-method - with intro of async there are always methods missing
+# cyclic-import - doesn't test if both import on load
+# duplicate-code - unavoidable
+# locally-disabled - it spams too much
+# too-many-* - are not enforced for the sake of readability
+# too-few-* - same as too-many-*
+# unused-argument - generic callbacks and setup methods create a lot of warnings
+disable = [
+    "format",
+    "abstract-method",
+    "cyclic-import",
+    "duplicate-code",
+    "locally-disabled",
+    "no-else-return",
+    "not-context-manager",
+    "too-few-public-methods",
+    "too-many-arguments",
+    "too-many-branches",
+    "too-many-instance-attributes",
+    "too-many-lines",
+    "too-many-locals",
+    "too-many-public-methods",
+    "too-many-return-statements",
+    "too-many-statements",
+    "unused-argument",
+    "consider-using-with",
+]
+
+[tool.pylint.REPORTS]
+score = false
+
+[tool.pylint.TYPECHECK]
+ignored-modules = ["distutils"]
+
+[tool.pylint.FORMAT]
+expected-line-ending-format = "LF"
+
+[tool.pylint.EXCEPTIONS]
+overgeneral-exceptions = ["builtins.BaseException", "builtins.Exception"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+norecursedirs = [".git"]
+log_format = "%(asctime)s.%(msecs)03d %(levelname)-8s %(threadName)s %(name)s:%(filename)s:%(lineno)s %(message)s"
+log_date_format = "%Y-%m-%d %H:%M:%S"
+asyncio_mode = "auto"
+filterwarnings = [
+    "error",
+    "ignore:pkg_resources is deprecated as an API:DeprecationWarning:dirhash",
+    "ignore::pytest.PytestUnraisableExceptionWarning",
+]
+
+[tool.isort]
+multi_line_output = 3
+include_trailing_comma = true
+force_grid_wrap = 0
+line_length = 88
+indent = "    "
+force_sort_within_sections = true
+sections = ["FUTURE", "STDLIB", "THIRDPARTY", "FIRSTPARTY", "LOCALFOLDER"]
+default_section = "THIRDPARTY"
+forced_separate = "tests"
+combine_as_imports = true
+use_parentheses = true
+known_first_party = ["supervisor", "tests"]

pytest.ini

@@ -1,2 +0,0 @@
-[pytest]
-asyncio_mode = auto

requirements.txt

@@ -1,26 +1,30 @@
-aiodns==3.0.0
-aiohttp==3.8.4
-async_timeout==4.0.2
+aiodns==3.1.1
+aiohttp==3.9.1
+aiohttp-fast-url-dispatcher==0.3.0
+async_timeout==4.0.3
 atomicwrites-homeassistant==1.4.1
-attrs==23.1.0
-awesomeversion==23.5.0
-brotli==1.0.9
-ciso8601==2.3.0
-colorlog==6.7.0
+attrs==23.2.0
+awesomeversion==23.11.0
+brotli==1.1.0
+ciso8601==2.3.1
+colorlog==6.8.0
 cpe==1.2.1
-cryptography==41.0.1
-debugpy==1.6.7
-deepmerge==1.1.0
+cryptography==41.0.7
+debugpy==1.8.0
+deepmerge==1.1.1
 dirhash==0.2.1
-docker==6.1.3
-faust-cchardet==2.1.18
-gitpython==3.1.31
-jinja2==3.1.2
+docker==7.0.0
+faust-cchardet==2.1.19
+gitpython==3.1.41
+jinja2==3.1.3
+orjson==3.9.10
 pulsectl==23.5.2
 pyudev==0.24.1
-ruamel.yaml==0.17.21
-securetar==2023.3.0
-sentry-sdk==1.27.0
-voluptuous==0.13.1
-dbus-fast==1.86.0
-typing_extensions==4.7.1
+PyYAML==6.0.1
+securetar==2023.12.0
+sentry-sdk==1.39.2
+setuptools==69.0.3
+voluptuous==0.14.1
+dbus-fast==2.21.0
+typing_extensions==4.9.0
+zlib-fast==0.1.0

requirements_tests.txt

@@ -1,16 +1,16 @@
-black==23.3.0
-coverage==7.2.7
+black==23.12.1
+coverage==7.4.0
 flake8-docstrings==1.7.0
-flake8==6.0.0
-pre-commit==3.3.3
+flake8==7.0.0
+pre-commit==3.6.0
 pydocstyle==6.3.0
-pylint==2.17.4
-pytest-aiohttp==1.0.4
-pytest-asyncio==0.18.3
+pylint==3.0.3
+pytest-aiohttp==1.0.5
+pytest-asyncio==0.23.3
 pytest-cov==4.1.0
-pytest-timeout==2.1.0
-pytest==7.4.0
-pyupgrade==3.7.0
-time-machine==2.10.0
-typing_extensions==4.7.1
-urllib3==2.0.3
+pytest-timeout==2.2.0
+pytest==7.4.4
+pyupgrade==3.15.0
+time-machine==2.13.0
+typing_extensions==4.9.0
+urllib3==2.1.0

rootfs/etc/services.d/watchdog/run

@@ -15,7 +15,7 @@ do
     if [[ "${supervisor_state}" = "running"  ]]; then
 
         # Check API
-        if bashio::supervisor.ping; then
+        if bashio::supervisor.ping > /dev/null; then
             failed_count=0
         else
             bashio::log.warning "Maybe found an issue on API healthy"

setup.cfg

@@ -1,17 +1,3 @@
-[isort]
-multi_line_output = 3
-include_trailing_comma=True
-force_grid_wrap=0
-line_length=88
-indent = "    "
-force_sort_within_sections = true
-sections = FUTURE,STDLIB,THIRDPARTY,FIRSTPARTY,LOCALFOLDER
-default_section = THIRDPARTY
-forced_separate = tests
-combine_as_imports = true
-use_parentheses = true
-known_first_party = supervisor,tests
-
 [flake8]
 exclude = .venv,.git,.tox,docs,venv,bin,lib,deps,build
 doctests = True

setup.py

@@ -1,60 +1,27 @@
 """Home Assistant Supervisor setup."""
+from pathlib import Path
+import re
+
 from setuptools import setup
 
-from supervisor.const import SUPERVISOR_VERSION
+RE_SUPERVISOR_VERSION = re.compile(r"^SUPERVISOR_VERSION =\s*(.+)$")
+
+SUPERVISOR_DIR = Path(__file__).parent
+REQUIREMENTS_FILE = SUPERVISOR_DIR / "requirements.txt"
+CONST_FILE = SUPERVISOR_DIR / "supervisor/const.py"
+
+REQUIREMENTS = REQUIREMENTS_FILE.read_text(encoding="utf-8")
+CONSTANTS = CONST_FILE.read_text(encoding="utf-8")
+
+
+def _get_supervisor_version():
+    for line in CONSTANTS.split("/n"):
+        if match := RE_SUPERVISOR_VERSION.match(line):
+            return match.group(1)
+    return "99.9.9dev"
+
 
 setup(
-    name="Supervisor",
-    version=SUPERVISOR_VERSION,
-    license="BSD License",
-    author="The Home Assistant Authors",
-    author_email="hello@home-assistant.io",
-    url="https://home-assistant.io/",
-    description=("Open-source private cloud os for Home-Assistant" " based on HassOS"),
-    long_description=(
-        "A maintainless private cloud operator system that"
-        "setup a Home-Assistant instance. Based on HassOS"
-    ),
-    classifiers=[
-        "Intended Audience :: End Users/Desktop",
-        "Intended Audience :: Developers",
-        "License :: OSI Approved :: Apache Software License",
-        "Operating System :: OS Independent",
-        "Topic :: Home Automation",
-        "Topic :: Software Development :: Libraries :: Python Modules",
-        "Topic :: Scientific/Engineering :: Atmospheric Science",
-        "Development Status :: 5 - Production/Stable",
-        "Intended Audience :: Developers",
-        "Programming Language :: Python :: 3.8",
-    ],
-    keywords=["docker", "home-assistant", "api"],
-    zip_safe=False,
-    platforms="any",
-    packages=[
-        "supervisor.addons",
-        "supervisor.api",
-        "supervisor.backups",
-        "supervisor.dbus.network",
-        "supervisor.dbus.network.setting",
-        "supervisor.dbus",
-        "supervisor.discovery.services",
-        "supervisor.discovery",
-        "supervisor.docker",
-        "supervisor.homeassistant",
-        "supervisor.host",
-        "supervisor.jobs",
-        "supervisor.misc",
-        "supervisor.plugins",
-        "supervisor.resolution.checks",
-        "supervisor.resolution.evaluations",
-        "supervisor.resolution.fixups",
-        "supervisor.resolution",
-        "supervisor.security",
-        "supervisor.services.modules",
-        "supervisor.services",
-        "supervisor.store",
-        "supervisor.utils",
-        "supervisor",
-    ],
-    include_package_data=True,
+    version=_get_supervisor_version(),
+    dependencies=REQUIREMENTS.split("/n"),
 )

supervisor/__main__.py

@@ -5,7 +5,13 @@ import logging
 from pathlib import Path
 import sys
 
-from supervisor import bootstrap
+import zlib_fast
+
+# Enable fast zlib before importing supervisor
+zlib_fast.enable()
+
+from supervisor import bootstrap  # noqa: E402
+from supervisor.utils.logging import activate_log_queue_handler  # noqa: E402
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
@@ -38,6 +44,8 @@ if __name__ == "__main__":
     executor = ThreadPoolExecutor(thread_name_prefix="SyncWorker")
     loop.set_default_executor(executor)
 
+    activate_log_queue_handler()
+
     _LOGGER.info("Initializing Supervisor setup")
     coresys = loop.run_until_complete(bootstrap.initialize_coresys())
     loop.set_debug(coresys.config.debug)

supervisor/addons/__init__.py

@@ -1,457 +1 @@
 """Init file for Supervisor add-ons."""
-import asyncio
-from collections.abc import Awaitable
-from contextlib import suppress
-import logging
-import tarfile
-from typing import Union
-
-from ..const import AddonBoot, AddonStartup, AddonState
-from ..coresys import CoreSys, CoreSysAttributes
-from ..exceptions import (
-    AddonConfigurationError,
-    AddonsError,
-    AddonsJobError,
-    AddonsNotSupportedError,
-    CoreDNSError,
-    DockerAPIError,
-    DockerError,
-    DockerNotFound,
-    HomeAssistantAPIError,
-    HostAppArmorError,
-)
-from ..jobs.decorator import Job, JobCondition
-from ..resolution.const import ContextType, IssueType, SuggestionType
-from ..store.addon import AddonStore
-from ..utils import check_exception_chain
-from ..utils.sentry import capture_exception
-from .addon import Addon
-from .const import ADDON_UPDATE_CONDITIONS
-from .data import AddonsData
-
-_LOGGER: logging.Logger = logging.getLogger(__name__)
-
-AnyAddon = Union[Addon, AddonStore]
-
-
-class AddonManager(CoreSysAttributes):
-    """Manage add-ons inside Supervisor."""
-
-    def __init__(self, coresys: CoreSys):
-        """Initialize Docker base wrapper."""
-        self.coresys: CoreSys = coresys
-        self.data: AddonsData = AddonsData(coresys)
-        self.local: dict[str, Addon] = {}
-        self.store: dict[str, AddonStore] = {}
-
-    @property
-    def all(self) -> list[AnyAddon]:
-        """Return a list of all add-ons."""
-        addons: dict[str, AnyAddon] = {**self.store, **self.local}
-        return list(addons.values())
-
-    @property
-    def installed(self) -> list[Addon]:
-        """Return a list of all installed add-ons."""
-        return list(self.local.values())
-
-    def get(self, addon_slug: str, local_only: bool = False) -> AnyAddon | None:
-        """Return an add-on from slug.
-
-        Prio:
-          1 - Local
-          2 - Store
-        """
-        if addon_slug in self.local:
-            return self.local[addon_slug]
-        if not local_only:
-            return self.store.get(addon_slug)
-        return None
-
-    def from_token(self, token: str) -> Addon | None:
-        """Return an add-on from Supervisor token."""
-        for addon in self.installed:
-            if token == addon.supervisor_token:
-                return addon
-        return None
-
-    async def load(self) -> None:
-        """Start up add-on management."""
-        tasks = []
-        for slug in self.data.system:
-            addon = self.local[slug] = Addon(self.coresys, slug)
-            tasks.append(self.sys_create_task(addon.load()))
-
-        # Run initial tasks
-        _LOGGER.info("Found %d installed add-ons", len(tasks))
-        if tasks:
-            await asyncio.wait(tasks)
-
-        # Sync DNS
-        await self.sync_dns()
-
-    async def boot(self, stage: AddonStartup) -> None:
-        """Boot add-ons with mode auto."""
-        tasks: list[Addon] = []
-        for addon in self.installed:
-            if addon.boot != AddonBoot.AUTO or addon.startup != stage:
-                continue
-            tasks.append(addon)
-
-        # Evaluate add-ons which need to be started
-        _LOGGER.info("Phase '%s' starting %d add-ons", stage, len(tasks))
-        if not tasks:
-            return
-
-        # Start Add-ons sequential
-        # avoid issue on slow IO
-        # Config.wait_boot is deprecated. Until addons update with healthchecks,
-        # add a sleep task for it to keep the same minimum amount of wait time
-        wait_boot: list[Awaitable[None]] = [asyncio.sleep(self.sys_config.wait_boot)]
-        for addon in tasks:
-            try:
-                if start_task := await addon.start():
-                    wait_boot.append(start_task)
-            except AddonsError as err:
-                # Check if there is an system/user issue
-                if check_exception_chain(
-                    err, (DockerAPIError, DockerNotFound, AddonConfigurationError)
-                ):
-                    addon.boot = AddonBoot.MANUAL
-                    addon.save_persist()
-            except Exception as err:  # pylint: disable=broad-except
-                capture_exception(err)
-            else:
-                continue
-
-            _LOGGER.warning("Can't start Add-on %s", addon.slug)
-
-        # Ignore exceptions from waiting for addon startup, addon errors handled elsewhere
-        await asyncio.gather(*wait_boot, return_exceptions=True)
-
-    async def shutdown(self, stage: AddonStartup) -> None:
-        """Shutdown addons."""
-        tasks: list[Addon] = []
-        for addon in self.installed:
-            if addon.state != AddonState.STARTED or addon.startup != stage:
-                continue
-            tasks.append(addon)
-
-        # Evaluate add-ons which need to be stopped
-        _LOGGER.info("Phase '%s' stopping %d add-ons", stage, len(tasks))
-        if not tasks:
-            return
-
-        # Stop Add-ons sequential
-        # avoid issue on slow IO
-        for addon in tasks:
-            try:
-                await addon.stop()
-            except Exception as err:  # pylint: disable=broad-except
-                _LOGGER.warning("Can't stop Add-on %s: %s", addon.slug, err)
-                capture_exception(err)
-
-    @Job(
-        conditions=ADDON_UPDATE_CONDITIONS,
-        on_condition=AddonsJobError,
-    )
-    async def install(self, slug: str) -> None:
-        """Install an add-on."""
-        if slug in self.local:
-            raise AddonsError(f"Add-on {slug} is already installed", _LOGGER.warning)
-        store = self.store.get(slug)
-
-        if not store:
-            raise AddonsError(f"Add-on {slug} does not exist", _LOGGER.error)
-
-        store.validate_availability()
-
-        self.data.install(store)
-        addon = Addon(self.coresys, slug)
-        await addon.load()
-
-        if not addon.path_data.is_dir():
-            _LOGGER.info(
-                "Creating Home Assistant add-on data folder %s", addon.path_data
-            )
-            addon.path_data.mkdir()
-
-        # Setup/Fix AppArmor profile
-        await addon.install_apparmor()
-
-        try:
-            await addon.instance.install(store.version, store.image, arch=addon.arch)
-        except DockerError as err:
-            self.data.uninstall(addon)
-            raise AddonsError() from err
-
-        self.local[slug] = addon
-
-        # Reload ingress tokens
-        if addon.with_ingress:
-            await self.sys_ingress.reload()
-
-        _LOGGER.info("Add-on '%s' successfully installed", slug)
-
-    async def uninstall(self, slug: str) -> None:
-        """Remove an add-on."""
-        if slug not in self.local:
-            _LOGGER.warning("Add-on %s is not installed", slug)
-            return
-        addon = self.local[slug]
-
-        try:
-            await addon.instance.remove()
-        except DockerError as err:
-            raise AddonsError() from err
-
-        addon.state = AddonState.UNKNOWN
-
-        await addon.unload()
-
-        # Cleanup audio settings
-        if addon.path_pulse.exists():
-            with suppress(OSError):
-                addon.path_pulse.unlink()
-
-        # Cleanup AppArmor profile
-        with suppress(HostAppArmorError):
-            await addon.uninstall_apparmor()
-
-        # Cleanup Ingress panel from sidebar
-        if addon.ingress_panel:
-            addon.ingress_panel = False
-            with suppress(HomeAssistantAPIError):
-                await self.sys_ingress.update_hass_panel(addon)
-
-        # Cleanup Ingress dynamic port assignment
-        if addon.with_ingress:
-            self.sys_create_task(self.sys_ingress.reload())
-            self.sys_ingress.del_dynamic_port(slug)
-
-        # Cleanup discovery data
-        for message in self.sys_discovery.list_messages:
-            if message.addon != addon.slug:
-                continue
-            self.sys_discovery.remove(message)
-
-        # Cleanup services data
-        for service in self.sys_services.list_services:
-            if addon.slug not in service.active:
-                continue
-            service.del_service_data(addon)
-
-        self.data.uninstall(addon)
-        self.local.pop(slug)
-
-        _LOGGER.info("Add-on '%s' successfully removed", slug)
-
-    @Job(
-        conditions=ADDON_UPDATE_CONDITIONS,
-        on_condition=AddonsJobError,
-    )
-    async def update(
-        self, slug: str, backup: bool | None = False
-    ) -> Awaitable[None] | None:
-        """Update add-on.
-
-        Returns a coroutine that completes when addon has state 'started' (see addon.start)
-        if addon is started after update. Else nothing is returned.
-        """
-        if slug not in self.local:
-            raise AddonsError(f"Add-on {slug} is not installed", _LOGGER.error)
-        addon = self.local[slug]
-
-        if addon.is_detached:
-            raise AddonsError(
-                f"Add-on {slug} is not available inside store", _LOGGER.error
-            )
-        store = self.store[slug]
-
-        if addon.version == store.version:
-            raise AddonsError(f"No update available for add-on {slug}", _LOGGER.warning)
-
-        # Check if available, Maybe something have changed
-        store.validate_availability()
-
-        if backup:
-            await self.sys_backups.do_backup_partial(
-                name=f"addon_{addon.slug}_{addon.version}",
-                homeassistant=False,
-                addons=[addon.slug],
-            )
-
-        # Update instance
-        last_state: AddonState = addon.state
-        old_image = addon.image
-        try:
-            await addon.instance.update(store.version, store.image)
-        except DockerError as err:
-            raise AddonsError() from err
-
-        _LOGGER.info("Add-on '%s' successfully updated", slug)
-        self.data.update(store)
-
-        # Cleanup
-        with suppress(DockerError):
-            await addon.instance.cleanup(old_image=old_image)
-
-        # Setup/Fix AppArmor profile
-        await addon.install_apparmor()
-
-        # restore state
-        return (
-            await addon.start()
-            if last_state in [AddonState.STARTED, AddonState.STARTUP]
-            else None
-        )
-
-    @Job(
-        conditions=[
-            JobCondition.FREE_SPACE,
-            JobCondition.INTERNET_HOST,
-            JobCondition.HEALTHY,
-        ],
-        on_condition=AddonsJobError,
-    )
-    async def rebuild(self, slug: str) -> Awaitable[None] | None:
-        """Perform a rebuild of local build add-on.
-
-        Returns a coroutine that completes when addon has state 'started' (see addon.start)
-        if addon is started after rebuild. Else nothing is returned.
-        """
-        if slug not in self.local:
-            raise AddonsError(f"Add-on {slug} is not installed", _LOGGER.error)
-        addon = self.local[slug]
-
-        if addon.is_detached:
-            raise AddonsError(
-                f"Add-on {slug} is not available inside store", _LOGGER.error
-            )
-        store = self.store[slug]
-
-        # Check if a rebuild is possible now
-        if addon.version != store.version:
-            raise AddonsError(
-                "Version changed, use Update instead Rebuild", _LOGGER.error
-            )
-        if not addon.need_build:
-            raise AddonsNotSupportedError(
-                "Can't rebuild a image based add-on", _LOGGER.error
-            )
-
-        # remove docker container but not addon config
-        last_state: AddonState = addon.state
-        try:
-            await addon.instance.remove()
-            await addon.instance.install(addon.version)
-        except DockerError as err:
-            raise AddonsError() from err
-
-        self.data.update(store)
-        _LOGGER.info("Add-on '%s' successfully rebuilt", slug)
-
-        # restore state
-        return (
-            await addon.start()
-            if last_state in [AddonState.STARTED, AddonState.STARTUP]
-            else None
-        )
-
-    @Job(
-        conditions=[
-            JobCondition.FREE_SPACE,
-            JobCondition.INTERNET_HOST,
-            JobCondition.HEALTHY,
-        ],
-        on_condition=AddonsJobError,
-    )
-    async def restore(
-        self, slug: str, tar_file: tarfile.TarFile
-    ) -> Awaitable[None] | None:
-        """Restore state of an add-on.
-
-        Returns a coroutine that completes when addon has state 'started' (see addon.start)
-        if addon is started after restore. Else nothing is returned.
-        """
-        if slug not in self.local:
-            _LOGGER.debug("Add-on %s is not local available for restore", slug)
-            addon = Addon(self.coresys, slug)
-        else:
-            _LOGGER.debug("Add-on %s is local available for restore", slug)
-            addon = self.local[slug]
-
-        wait_for_start = await addon.restore(tar_file)
-
-        # Check if new
-        if slug not in self.local:
-            _LOGGER.info("Detect new Add-on after restore %s", slug)
-            self.local[slug] = addon
-
-        # Update ingress
-        if addon.with_ingress:
-            await self.sys_ingress.reload()
-            with suppress(HomeAssistantAPIError):
-                await self.sys_ingress.update_hass_panel(addon)
-
-        return wait_for_start
-
-    @Job(conditions=[JobCondition.FREE_SPACE, JobCondition.INTERNET_HOST])
-    async def repair(self) -> None:
-        """Repair local add-ons."""
-        needs_repair: list[Addon] = []
-
-        # Evaluate Add-ons to repair
-        for addon in self.installed:
-            if await addon.instance.exists():
-                continue
-            needs_repair.append(addon)
-
-        _LOGGER.info("Found %d add-ons to repair", len(needs_repair))
-        if not needs_repair:
-            return
-
-        for addon in needs_repair:
-            _LOGGER.info("Repairing for add-on: %s", addon.slug)
-            with suppress(DockerError, KeyError):
-                # Need pull a image again
-                if not addon.need_build:
-                    await addon.instance.install(addon.version, addon.image)
-                    continue
-
-                # Need local lookup
-                if addon.need_build and not addon.is_detached:
-                    store = self.store[addon.slug]
-                    # If this add-on is available for rebuild
-                    if addon.version == store.version:
-                        await addon.instance.install(addon.version, addon.image)
-                        continue
-
-            _LOGGER.error("Can't repair %s", addon.slug)
-            with suppress(AddonsError):
-                await self.uninstall(addon.slug)
-
-    async def sync_dns(self) -> None:
-        """Sync add-ons DNS names."""
-        # Update hosts
-        for addon in self.installed:
-            try:
-                if not await addon.instance.is_running():
-                    continue
-            except DockerError as err:
-                _LOGGER.warning("Add-on %s is corrupt: %s", addon.slug, err)
-                self.sys_resolution.create_issue(
-                    IssueType.CORRUPT_DOCKER,
-                    ContextType.ADDON,
-                    reference=addon.slug,
-                    suggestions=[SuggestionType.EXECUTE_REPAIR],
-                )
-                capture_exception(err)
-            else:
-                self.sys_plugins.dns.add_host(
-                    ipv4=addon.ip_address, names=[addon.hostname], write=False
-                )
-
-        # Write hosts files
-        with suppress(CoreDNSError):
-            self.sys_plugins.dns.write_hosts()

supervisor/addons/addon.py

@@ -3,6 +3,7 @@ import asyncio
 from collections.abc import Awaitable
 from contextlib import suppress
 from copy import deepcopy
+import errno
 from ipaddress import IPv4Address
 import logging
 from pathlib import Path, PurePath
@@ -64,12 +65,15 @@ from ..exceptions import (
     AddonsNotSupportedError,
     ConfigurationFileError,
     DockerError,
+    HomeAssistantAPIError,
     HostAppArmorError,
 )
 from ..hardware.data import Device
 from ..homeassistant.const import WSEvent, WSType
 from ..jobs.const import JobExecutionLimit
 from ..jobs.decorator import Job
+from ..resolution.const import UnhealthyReason
+from ..store.addon import AddonStore
 from ..utils import check_port
 from ..utils.apparmor import adjust_profile
 from ..utils.json import read_json_file, write_json_file
@@ -80,6 +84,7 @@ from .const import (
     WATCHDOG_THROTTLE_MAX_CALLS,
     WATCHDOG_THROTTLE_PERIOD,
     AddonBackupMode,
+    MappingType,
 )
 from .model import AddonModel, Data
 from .options import AddonOptions
@@ -129,54 +134,7 @@ class Addon(AddonModel):
         )
         self._listeners: list[EventListener] = []
         self._startup_event = asyncio.Event()
-
-        @Job(
-            name=f"addon_{slug}_restart_after_problem",
-            limit=JobExecutionLimit.THROTTLE_RATE_LIMIT,
-            throttle_period=WATCHDOG_THROTTLE_PERIOD,
-            throttle_max_calls=WATCHDOG_THROTTLE_MAX_CALLS,
-            on_condition=AddonsJobError,
-        )
-        async def restart_after_problem(addon: Addon, state: ContainerState):
-            """Restart unhealthy or failed addon."""
-            attempts = 0
-            while await addon.instance.current_state() == state:
-                if not addon.in_progress:
-                    _LOGGER.warning(
-                        "Watchdog found addon %s is %s, restarting...",
-                        addon.name,
-                        state.value,
-                    )
-                    try:
-                        if state == ContainerState.FAILED:
-                            # Ensure failed container is removed before attempting reanimation
-                            if attempts == 0:
-                                with suppress(DockerError):
-                                    await addon.instance.stop(remove_container=True)
-
-                            await (await addon.start())
-                        else:
-                            await (await addon.restart())
-                    except AddonsError as err:
-                        attempts = attempts + 1
-                        _LOGGER.error(
-                            "Watchdog restart of addon %s failed!", addon.name
-                        )
-                        capture_exception(err)
-                    else:
-                        break
-
-                if attempts >= WATCHDOG_MAX_ATTEMPTS:
-                    _LOGGER.critical(
-                        "Watchdog cannot restart addon %s, failed all %s attempts",
-                        addon.name,
-                        attempts,
-                    )
-                    break
-
-                await asyncio.sleep(WATCHDOG_RETRY_SECONDS)
-
-        self._restart_after_problem = restart_after_problem
+        self._startup_task: asyncio.Task | None = None
 
     def __repr__(self) -> str:
         """Return internal representation."""
@@ -228,6 +186,7 @@ class Addon(AddonModel):
             )
         )
 
+        await self._check_ingress_port()
         with suppress(DockerError):
             await self.instance.attach(version=self.version)
 
@@ -246,6 +205,11 @@ class Addon(AddonModel):
         """Return add-on data from store."""
         return self.sys_store.data.addons.get(self.slug, self.data)
 
+    @property
+    def addon_store(self) -> AddonStore | None:
+        """Return store representation of addon."""
+        return self.sys_addons.store.get(self.slug)
+
     @property
     def persist(self) -> Data:
         """Return add-on data/config."""
@@ -434,7 +398,7 @@ class Addon(AddonModel):
 
         port = self.data[ATTR_INGRESS_PORT]
         if port == 0:
-            return self.sys_ingress.get_dynamic_port(self.slug)
+            raise RuntimeError(f"No port set for add-on {self.slug}")
         return port
 
     @property
@@ -500,6 +464,21 @@ class Addon(AddonModel):
         """Return add-on data path external for Docker."""
         return PurePath(self.sys_config.path_extern_addons_data, self.slug)
 
+    @property
+    def addon_config_used(self) -> bool:
+        """Add-on is using its public config folder."""
+        return MappingType.ADDON_CONFIG in self.map_volumes
+
+    @property
+    def path_config(self) -> Path:
+        """Return add-on config path inside Supervisor."""
+        return Path(self.sys_config.path_addon_configs, self.slug)
+
+    @property
+    def path_extern_config(self) -> PurePath:
+        """Return add-on config path external for Docker."""
+        return PurePath(self.sys_config.path_extern_addon_configs, self.slug)
+
     @property
     def path_options(self) -> Path:
         """Return path to add-on options."""
@@ -563,7 +542,7 @@ class Addon(AddonModel):
 
         # TCP monitoring
         if s_prefix == "tcp":
-            return await self.sys_run_in_executor(check_port, self.ip_address, port)
+            return await check_port(self.ip_address, port)
 
         # lookup the correct protocol from config
         if t_proto:
@@ -579,7 +558,7 @@ class Addon(AddonModel):
             ) as req:
                 if req.status < 300:
                     return True
-        except (asyncio.TimeoutError, aiohttp.ClientError):
+        except (TimeoutError, aiohttp.ClientError):
             pass
 
         return False
@@ -606,16 +585,201 @@ class Addon(AddonModel):
 
         raise AddonConfigurationError()
 
+    @Job(
+        name="addon_unload",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
     async def unload(self) -> None:
         """Unload add-on and remove data."""
+        if self._startup_task:
+            # If we were waiting on startup, cancel that and let the task finish before proceeding
+            self._startup_task.cancel(f"Removing add-on {self.name} from system")
+            with suppress(asyncio.CancelledError):
+                await self._startup_task
+
         for listener in self._listeners:
             self.sys_bus.remove_listener(listener)
 
-        if not self.path_data.is_dir():
+        if self.path_data.is_dir():
+            _LOGGER.info("Removing add-on data folder %s", self.path_data)
+            await remove_data(self.path_data)
+
+    async def _check_ingress_port(self):
+        """Assign a ingress port if dynamic port selection is used."""
+        if not self.with_ingress:
             return
 
-        _LOGGER.info("Removing add-on data folder %s", self.path_data)
-        await remove_data(self.path_data)
+        if self.data[ATTR_INGRESS_PORT] == 0:
+            self.data[ATTR_INGRESS_PORT] = await self.sys_ingress.get_dynamic_port(
+                self.slug
+            )
+
+    @Job(
+        name="addon_install",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def install(self) -> None:
+        """Install and setup this addon."""
+        self.sys_addons.data.install(self.addon_store)
+        await self.load()
+
+        if not self.path_data.is_dir():
+            _LOGGER.info(
+                "Creating Home Assistant add-on data folder %s", self.path_data
+            )
+            self.path_data.mkdir()
+
+        # Setup/Fix AppArmor profile
+        await self.install_apparmor()
+
+        # Install image
+        try:
+            await self.instance.install(
+                self.latest_version, self.addon_store.image, arch=self.arch
+            )
+        except DockerError as err:
+            self.sys_addons.data.uninstall(self)
+            raise AddonsError() from err
+
+        # Add to addon manager
+        self.sys_addons.local[self.slug] = self
+
+        # Reload ingress tokens
+        if self.with_ingress:
+            await self.sys_ingress.reload()
+
+    @Job(
+        name="addon_uninstall",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def uninstall(self) -> None:
+        """Uninstall and cleanup this addon."""
+        try:
+            await self.instance.remove()
+        except DockerError as err:
+            raise AddonsError() from err
+
+        self.state = AddonState.UNKNOWN
+
+        await self.unload()
+
+        # Cleanup audio settings
+        if self.path_pulse.exists():
+            with suppress(OSError):
+                self.path_pulse.unlink()
+
+        # Cleanup AppArmor profile
+        with suppress(HostAppArmorError):
+            await self.uninstall_apparmor()
+
+        # Cleanup Ingress panel from sidebar
+        if self.ingress_panel:
+            self.ingress_panel = False
+            with suppress(HomeAssistantAPIError):
+                await self.sys_ingress.update_hass_panel(self)
+
+        # Cleanup Ingress dynamic port assignment
+        if self.with_ingress:
+            self.sys_create_task(self.sys_ingress.reload())
+            self.sys_ingress.del_dynamic_port(self.slug)
+
+        # Cleanup discovery data
+        for message in self.sys_discovery.list_messages:
+            if message.addon != self.slug:
+                continue
+            self.sys_discovery.remove(message)
+
+        # Cleanup services data
+        for service in self.sys_services.list_services:
+            if self.slug not in service.active:
+                continue
+            service.del_service_data(self)
+
+        # Remove from addon manager
+        self.sys_addons.data.uninstall(self)
+        self.sys_addons.local.pop(self.slug)
+
+    @Job(
+        name="addon_update",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def update(self) -> asyncio.Task | None:
+        """Update this addon to latest version.
+
+        Returns a Task that completes when addon has state 'started' (see start)
+        if it was running. Else nothing is returned.
+        """
+        old_image = self.image
+        # Cache data to prevent races with other updates to global
+        store = self.addon_store.clone()
+
+        try:
+            await self.instance.update(store.version, store.image, arch=self.arch)
+        except DockerError as err:
+            raise AddonsError() from err
+
+        # Stop the addon if running
+        if (last_state := self.state) in {AddonState.STARTED, AddonState.STARTUP}:
+            await self.stop()
+
+        try:
+            _LOGGER.info("Add-on '%s' successfully updated", self.slug)
+            self.sys_addons.data.update(store)
+            await self._check_ingress_port()
+
+            # Cleanup
+            with suppress(DockerError):
+                await self.instance.cleanup(
+                    old_image=old_image, image=store.image, version=store.version
+                )
+
+            # Setup/Fix AppArmor profile
+            await self.install_apparmor()
+
+        finally:
+            # restore state. Return Task for caller if no exception
+            out = (
+                await self.start()
+                if last_state in {AddonState.STARTED, AddonState.STARTUP}
+                else None
+            )
+        return out
+
+    @Job(
+        name="addon_rebuild",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def rebuild(self) -> asyncio.Task | None:
+        """Rebuild this addons container and image.
+
+        Returns a Task that completes when addon has state 'started' (see start)
+        if it was running. Else nothing is returned.
+        """
+        last_state: AddonState = self.state
+        try:
+            # remove docker container but not addon config
+            try:
+                await self.instance.remove()
+                await self.instance.install(self.version)
+            except DockerError as err:
+                raise AddonsError() from err
+
+            self.sys_addons.data.update(self.addon_store)
+            _LOGGER.info("Add-on '%s' successfully rebuilt", self.slug)
+
+        finally:
+            # restore state
+            out = (
+                await self.start()
+                if last_state in [AddonState.STARTED, AddonState.STARTUP]
+                else None
+            )
+        return out
 
     def write_pulse(self) -> None:
         """Write asound config to file and return True on success."""
@@ -631,6 +795,8 @@ class Addon(AddonModel):
         try:
             self.path_pulse.write_text(pulse_config, encoding="utf-8")
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             _LOGGER.error(
                 "Add-on %s can't write pulse/client.config: %s", self.slug, err
             )
@@ -699,24 +865,34 @@ class Addon(AddonModel):
     async def _wait_for_startup(self) -> None:
         """Wait for startup event to be set with timeout."""
         try:
-            await asyncio.wait_for(self._startup_event.wait(), STARTUP_TIMEOUT)
-        except asyncio.TimeoutError:
+            self._startup_task = self.sys_create_task(self._startup_event.wait())
+            await asyncio.wait_for(self._startup_task, STARTUP_TIMEOUT)
+        except TimeoutError:
             _LOGGER.warning(
-                "Timeout while waiting for addon %s to start, took more then %s seconds",
+                "Timeout while waiting for addon %s to start, took more than %s seconds",
                 self.name,
                 STARTUP_TIMEOUT,
             )
+        except asyncio.CancelledError as err:
+            _LOGGER.info("Wait for addon startup task cancelled due to: %s", err)
+        finally:
+            self._startup_task = None
 
-    async def start(self) -> Awaitable[None]:
+    @Job(
+        name="addon_start",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def start(self) -> asyncio.Task:
         """Set options and start add-on.
 
-        Returns a coroutine that completes when addon has state 'started'.
+        Returns a Task that completes when addon has state 'started'.
         For addons with a healthcheck, that is when they become healthy or unhealthy.
         Addons without a healthcheck have state 'started' immediately.
         """
         if await self.instance.is_running():
             _LOGGER.warning("%s is already running!", self.slug)
-            return self._wait_for_startup()
+            return self.sys_create_task(self._wait_for_startup())
 
         # Access Token
         self.persist[ATTR_ACCESS_TOKEN] = secrets.token_hex(56)
@@ -729,6 +905,18 @@ class Addon(AddonModel):
         if self.with_audio:
             self.write_pulse()
 
+        def _check_addon_config_dir():
+            if self.path_config.is_dir():
+                return
+
+            _LOGGER.info(
+                "Creating Home Assistant add-on config folder %s", self.path_config
+            )
+            self.path_config.mkdir()
+
+        if self.addon_config_used:
+            await self.sys_run_in_executor(_check_addon_config_dir)
+
         # Start Add-on
         self._startup_event.clear()
         try:
@@ -737,8 +925,13 @@ class Addon(AddonModel):
             self.state = AddonState.ERROR
             raise AddonsError() from err
 
-        return self._wait_for_startup()
+        return self.sys_create_task(self._wait_for_startup())
 
+    @Job(
+        name="addon_stop",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
     async def stop(self) -> None:
         """Stop add-on."""
         self._manual_stop = True
@@ -748,10 +941,15 @@ class Addon(AddonModel):
             self.state = AddonState.ERROR
             raise AddonsError() from err
 
-    async def restart(self) -> Awaitable[None]:
+    @Job(
+        name="addon_restart",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def restart(self) -> asyncio.Task:
         """Restart add-on.
 
-        Returns a coroutine that completes when addon has state 'started' (see start).
+        Returns a Task that completes when addon has state 'started' (see start).
         """
         with suppress(AddonsError):
             await self.stop()
@@ -778,11 +976,13 @@ class Addon(AddonModel):
         except DockerError as err:
             raise AddonsError() from err
 
+    @Job(
+        name="addon_write_stdin",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
     async def write_stdin(self, data) -> None:
-        """Write data to add-on stdin.
-
-        Return a coroutine.
-        """
+        """Write data to add-on stdin."""
         if not self.with_stdin:
             raise AddonsNotSupportedError(
                 f"Add-on {self.slug} does not support writing to stdin!", _LOGGER.error
@@ -810,14 +1010,59 @@ class Addon(AddonModel):
                 _LOGGER.error,
             ) from err
 
-    async def backup(self, tar_file: tarfile.TarFile) -> Awaitable[None] | None:
+    @Job(
+        name="addon_begin_backup",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def begin_backup(self) -> bool:
+        """Execute pre commands or stop addon if necessary.
+
+        Returns value of `is_running`. Caller should not call `end_backup` if return is false.
+        """
+        if not await self.is_running():
+            return False
+
+        if self.backup_mode == AddonBackupMode.COLD:
+            _LOGGER.info("Shutdown add-on %s for cold backup", self.slug)
+            await self.stop()
+
+        elif self.backup_pre is not None:
+            await self._backup_command(self.backup_pre)
+
+        return True
+
+    @Job(
+        name="addon_end_backup",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def end_backup(self) -> asyncio.Task | None:
+        """Execute post commands or restart addon if necessary.
+
+        Returns a Task that completes when addon has state 'started' (see start)
+        for cold backup. Else nothing is returned.
+        """
+        if self.backup_mode is AddonBackupMode.COLD:
+            _LOGGER.info("Starting add-on %s again", self.slug)
+            return await self.start()
+
+        if self.backup_post is not None:
+            await self._backup_command(self.backup_post)
+        return None
+
+    @Job(
+        name="addon_backup",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def backup(self, tar_file: tarfile.TarFile) -> asyncio.Task | None:
         """Backup state of an add-on.
 
-        Returns a coroutine that completes when addon has state 'started' (see start)
+        Returns a Task that completes when addon has state 'started' (see start)
         for cold backup. Else nothing is returned.
         """
         wait_for_start: Awaitable[None] | None = None
-        is_running = await self.is_running()
 
         with TemporaryDirectory(dir=self.sys_config.path_tmp) as temp:
             temp_path = Path(temp)
@@ -869,16 +1114,16 @@ class Addon(AddonModel):
                         arcname="data",
                     )
 
-            if (
-                is_running
-                and self.backup_mode == AddonBackupMode.HOT
-                and self.backup_pre is not None
-            ):
-                await self._backup_command(self.backup_pre)
-            elif is_running and self.backup_mode == AddonBackupMode.COLD:
-                _LOGGER.info("Shutdown add-on %s for cold backup", self.slug)
-                await self.instance.stop()
+                    # Backup config
+                    if self.addon_config_used:
+                        atomic_contents_add(
+                            backup,
+                            self.path_config,
+                            excludes=self.backup_exclude,
+                            arcname="config",
+                        )
 
+            is_running = await self.begin_backup()
             try:
                 _LOGGER.info("Building backup for add-on %s", self.slug)
                 await self.sys_run_in_executor(_write_tarfile)
@@ -887,23 +1132,21 @@ class Addon(AddonModel):
                     f"Can't write tarfile {tar_file}: {err}", _LOGGER.error
                 ) from err
             finally:
-                if (
-                    is_running
-                    and self.backup_mode == AddonBackupMode.HOT
-                    and self.backup_post is not None
-                ):
-                    await self._backup_command(self.backup_post)
-                elif is_running and self.backup_mode is AddonBackupMode.COLD:
-                    _LOGGER.info("Starting add-on %s again", self.slug)
-                    wait_for_start = await self.start()
+                if is_running:
+                    wait_for_start = await self.end_backup()
 
         _LOGGER.info("Finish backup for addon %s", self.slug)
         return wait_for_start
 
-    async def restore(self, tar_file: tarfile.TarFile) -> Awaitable[None] | None:
+    @Job(
+        name="addon_restore",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=AddonsJobError,
+    )
+    async def restore(self, tar_file: tarfile.TarFile) -> asyncio.Task | None:
         """Restore state of an add-on.
 
-        Returns a coroutine that completes when addon has state 'started' (see start)
+        Returns a Task that completes when addon has state 'started' (see start)
         if addon is started after restore. Else nothing is returned.
         """
         wait_for_start: Awaitable[None] | None = None
@@ -912,7 +1155,11 @@ class Addon(AddonModel):
             def _extract_tarfile():
                 """Extract tar backup."""
                 with tar_file as backup:
-                    backup.extractall(path=Path(temp), members=secure_path(backup))
+                    backup.extractall(
+                        path=Path(temp),
+                        members=secure_path(backup),
+                        filter="fully_trusted",
+                    )
 
             try:
                 await self.sys_run_in_executor(_extract_tarfile)
@@ -950,64 +1197,81 @@ class Addon(AddonModel):
                 self.slug, data[ATTR_USER], data[ATTR_SYSTEM], restore_image
             )
 
-            # Check version / restore image
-            version = data[ATTR_VERSION]
-            if not await self.instance.exists():
-                _LOGGER.info("Restore/Install of image for addon %s", self.slug)
+            # Stop it first if its running
+            if await self.instance.is_running():
+                await self.stop()
 
-                image_file = Path(temp, "image.tar")
-                if image_file.is_file():
-                    with suppress(DockerError):
-                        await self.instance.import_image(image_file)
-                else:
-                    with suppress(DockerError):
-                        await self.instance.install(version, restore_image)
-                        await self.instance.cleanup()
-            elif self.instance.version != version or self.legacy:
-                _LOGGER.info("Restore/Update of image for addon %s", self.slug)
-                with suppress(DockerError):
-                    await self.instance.update(version, restore_image)
-            else:
-                with suppress(DockerError):
-                    await self.instance.stop()
-
-            # Restore data
-            def _restore_data():
-                """Restore data."""
-                temp_data = Path(temp, "data")
-                if temp_data.is_dir():
-                    shutil.copytree(temp_data, self.path_data, symlinks=True)
-                else:
-                    self.path_data.mkdir()
-
-            _LOGGER.info("Restoring data for addon %s", self.slug)
-            if self.path_data.is_dir():
-                await remove_data(self.path_data)
             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
+                # Check version / restore image
+                version = data[ATTR_VERSION]
+                if not await self.instance.exists():
+                    _LOGGER.info("Restore/Install of image for addon %s", self.slug)
+
+                    image_file = Path(temp, "image.tar")
+                    if image_file.is_file():
+                        with suppress(DockerError):
+                            await self.instance.import_image(image_file)
+                    else:
+                        with suppress(DockerError):
+                            await self.instance.install(
+                                version, restore_image, self.arch
+                            )
+                            await self.instance.cleanup()
+                elif self.instance.version != version or self.legacy:
+                    _LOGGER.info("Restore/Update of image for addon %s", self.slug)
+                    with suppress(DockerError):
+                        await self.instance.update(version, restore_image, self.arch)
+                self._check_ingress_port()
+
+                # Restore data and config
+                def _restore_data():
+                    """Restore data and config."""
+                    temp_data = Path(temp, "data")
+                    if temp_data.is_dir():
+                        shutil.copytree(temp_data, self.path_data, symlinks=True)
+                    else:
+                        self.path_data.mkdir()
+
+                    temp_config = Path(temp, "config")
+                    if temp_config.is_dir():
+                        shutil.copytree(temp_config, self.path_config, symlinks=True)
+                    elif self.addon_config_used:
+                        self.path_config.mkdir()
+
+                _LOGGER.info("Restoring data and config for addon %s", self.slug)
+                if self.path_data.is_dir():
+                    await remove_data(self.path_data)
+                if self.path_config.is_dir():
+                    await remove_data(self.path_config)
 
-            # Restore AppArmor
-            profile_file = Path(temp, "apparmor.txt")
-            if profile_file.exists():
                 try:
-                    await self.sys_host.apparmor.load_profile(self.slug, profile_file)
-                except HostAppArmorError as err:
-                    _LOGGER.error(
-                        "Can't restore AppArmor profile for add-on %s", self.slug
-                    )
-                    raise AddonsError() from err
+                    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
 
-            # Is add-on loaded
-            if not self.loaded:
-                await self.load()
+                # Restore AppArmor
+                profile_file = Path(temp, "apparmor.txt")
+                if profile_file.exists():
+                    try:
+                        await self.sys_host.apparmor.load_profile(
+                            self.slug, profile_file
+                        )
+                    except HostAppArmorError as err:
+                        _LOGGER.error(
+                            "Can't restore AppArmor profile for add-on %s", self.slug
+                        )
+                        raise AddonsError() from err
 
-            # Run add-on
-            if data[ATTR_STATE] == AddonState.STARTED:
-                wait_for_start = await self.start()
+                # Is add-on loaded
+                if not self.loaded:
+                    await self.load()
+
+            finally:
+                # Run add-on
+                if data[ATTR_STATE] == AddonState.STARTED:
+                    wait_for_start = await self.start()
 
         _LOGGER.info("Finished restore for add-on %s", self.slug)
         return wait_for_start
@@ -1019,6 +1283,50 @@ class Addon(AddonModel):
         """
         return self.instance.check_trust()
 
+    @Job(
+        name="addon_restart_after_problem",
+        limit=JobExecutionLimit.GROUP_THROTTLE_RATE_LIMIT,
+        throttle_period=WATCHDOG_THROTTLE_PERIOD,
+        throttle_max_calls=WATCHDOG_THROTTLE_MAX_CALLS,
+        on_condition=AddonsJobError,
+    )
+    async def _restart_after_problem(self, state: ContainerState):
+        """Restart unhealthy or failed addon."""
+        attempts = 0
+        while await self.instance.current_state() == state:
+            if not self.in_progress:
+                _LOGGER.warning(
+                    "Watchdog found addon %s is %s, restarting...",
+                    self.name,
+                    state,
+                )
+                try:
+                    if state == ContainerState.FAILED:
+                        # Ensure failed container is removed before attempting reanimation
+                        if attempts == 0:
+                            with suppress(DockerError):
+                                await self.instance.stop(remove_container=True)
+
+                        await (await self.start())
+                    else:
+                        await (await self.restart())
+                except AddonsError as err:
+                    attempts = attempts + 1
+                    _LOGGER.error("Watchdog restart of addon %s failed!", self.name)
+                    capture_exception(err)
+                else:
+                    break
+
+            if attempts >= WATCHDOG_MAX_ATTEMPTS:
+                _LOGGER.critical(
+                    "Watchdog cannot restart addon %s, failed all %s attempts",
+                    self.name,
+                    attempts,
+                )
+                break
+
+            await asyncio.sleep(WATCHDOG_RETRY_SECONDS)
+
     async def container_state_changed(self, event: DockerContainerStateEvent) -> None:
         """Set addon state from container state."""
         if event.name != self.instance.name:
@@ -1053,4 +1361,4 @@ class Addon(AddonModel):
             ContainerState.STOPPED,
             ContainerState.UNHEALTHY,
         ]:
-            await self._restart_after_problem(self, event.state)
+            await self._restart_after_problem(event.state)

supervisor/addons/configuration.py

@@ -0,0 +1,11 @@
+"""Confgiuration Objects for Addon Config."""
+
+from dataclasses import dataclass
+
+
+@dataclass(slots=True)
+class FolderMapping:
+    """Represent folder mapping configuration."""
+
+    path: str | None
+    read_only: bool

supervisor/addons/const.py

@@ -1,19 +1,36 @@
 """Add-on static data."""
 from datetime import timedelta
-from enum import Enum
+from enum import StrEnum
 
 from ..jobs.const import JobCondition
 
 
-class AddonBackupMode(str, Enum):
+class AddonBackupMode(StrEnum):
     """Backup mode of an Add-on."""
 
     HOT = "hot"
     COLD = "cold"
 
 
+class MappingType(StrEnum):
+    """Mapping type of an Add-on Folder."""
+
+    DATA = "data"
+    CONFIG = "config"
+    SSL = "ssl"
+    ADDONS = "addons"
+    BACKUP = "backup"
+    SHARE = "share"
+    MEDIA = "media"
+    HOMEASSISTANT_CONFIG = "homeassistant_config"
+    ALL_ADDON_CONFIGS = "all_addon_configs"
+    ADDON_CONFIG = "addon_config"
+
+
 ATTR_BACKUP = "backup"
 ATTR_CODENOTARY = "codenotary"
+ATTR_READ_ONLY = "read_only"
+ATTR_PATH = "path"
 WATCHDOG_RETRY_SECONDS = 10
 WATCHDOG_MAX_ATTEMPTS = 5
 WATCHDOG_THROTTLE_PERIOD = timedelta(minutes=30)

supervisor/addons/manager.py

@@ -0,0 +1,374 @@
+"""Supervisor add-on manager."""
+import asyncio
+from collections.abc import Awaitable
+from contextlib import suppress
+import logging
+import tarfile
+from typing import Union
+
+from ..const import AddonBoot, AddonStartup, AddonState
+from ..coresys import CoreSys, CoreSysAttributes
+from ..exceptions import (
+    AddonConfigurationError,
+    AddonsError,
+    AddonsJobError,
+    AddonsNotSupportedError,
+    CoreDNSError,
+    DockerAPIError,
+    DockerError,
+    DockerNotFound,
+    HassioError,
+    HomeAssistantAPIError,
+)
+from ..jobs.decorator import Job, JobCondition
+from ..resolution.const import ContextType, IssueType, SuggestionType
+from ..store.addon import AddonStore
+from ..utils import check_exception_chain
+from ..utils.sentry import capture_exception
+from .addon import Addon
+from .const import ADDON_UPDATE_CONDITIONS
+from .data import AddonsData
+
+_LOGGER: logging.Logger = logging.getLogger(__name__)
+
+AnyAddon = Union[Addon, AddonStore]
+
+
+class AddonManager(CoreSysAttributes):
+    """Manage add-ons inside Supervisor."""
+
+    def __init__(self, coresys: CoreSys):
+        """Initialize Docker base wrapper."""
+        self.coresys: CoreSys = coresys
+        self.data: AddonsData = AddonsData(coresys)
+        self.local: dict[str, Addon] = {}
+        self.store: dict[str, AddonStore] = {}
+
+    @property
+    def all(self) -> list[AnyAddon]:
+        """Return a list of all add-ons."""
+        addons: dict[str, AnyAddon] = {**self.store, **self.local}
+        return list(addons.values())
+
+    @property
+    def installed(self) -> list[Addon]:
+        """Return a list of all installed add-ons."""
+        return list(self.local.values())
+
+    def get(self, addon_slug: str, local_only: bool = False) -> AnyAddon | None:
+        """Return an add-on from slug.
+
+        Prio:
+          1 - Local
+          2 - Store
+        """
+        if addon_slug in self.local:
+            return self.local[addon_slug]
+        if not local_only:
+            return self.store.get(addon_slug)
+        return None
+
+    def from_token(self, token: str) -> Addon | None:
+        """Return an add-on from Supervisor token."""
+        for addon in self.installed:
+            if token == addon.supervisor_token:
+                return addon
+        return None
+
+    async def load(self) -> None:
+        """Start up add-on management."""
+        tasks = []
+        for slug in self.data.system:
+            addon = self.local[slug] = Addon(self.coresys, slug)
+            tasks.append(self.sys_create_task(addon.load()))
+
+        # Run initial tasks
+        _LOGGER.info("Found %d installed add-ons", len(tasks))
+        if tasks:
+            await asyncio.wait(tasks)
+
+        # Sync DNS
+        await self.sync_dns()
+
+    async def boot(self, stage: AddonStartup) -> None:
+        """Boot add-ons with mode auto."""
+        tasks: list[Addon] = []
+        for addon in self.installed:
+            if addon.boot != AddonBoot.AUTO or addon.startup != stage:
+                continue
+            tasks.append(addon)
+
+        # Evaluate add-ons which need to be started
+        _LOGGER.info("Phase '%s' starting %d add-ons", stage, len(tasks))
+        if not tasks:
+            return
+
+        # Start Add-ons sequential
+        # avoid issue on slow IO
+        # Config.wait_boot is deprecated. Until addons update with healthchecks,
+        # add a sleep task for it to keep the same minimum amount of wait time
+        wait_boot: list[Awaitable[None]] = [asyncio.sleep(self.sys_config.wait_boot)]
+        for addon in tasks:
+            try:
+                if start_task := await addon.start():
+                    wait_boot.append(start_task)
+            except AddonsError as err:
+                # Check if there is an system/user issue
+                if check_exception_chain(
+                    err, (DockerAPIError, DockerNotFound, AddonConfigurationError)
+                ):
+                    addon.boot = AddonBoot.MANUAL
+                    addon.save_persist()
+            except HassioError:
+                pass  # These are already handled
+            else:
+                continue
+
+            _LOGGER.warning("Can't start Add-on %s", addon.slug)
+
+        # Ignore exceptions from waiting for addon startup, addon errors handled elsewhere
+        await asyncio.gather(*wait_boot, return_exceptions=True)
+
+    async def shutdown(self, stage: AddonStartup) -> None:
+        """Shutdown addons."""
+        tasks: list[Addon] = []
+        for addon in self.installed:
+            if addon.state != AddonState.STARTED or addon.startup != stage:
+                continue
+            tasks.append(addon)
+
+        # Evaluate add-ons which need to be stopped
+        _LOGGER.info("Phase '%s' stopping %d add-ons", stage, len(tasks))
+        if not tasks:
+            return
+
+        # Stop Add-ons sequential
+        # avoid issue on slow IO
+        for addon in tasks:
+            try:
+                await addon.stop()
+            except Exception as err:  # pylint: disable=broad-except
+                _LOGGER.warning("Can't stop Add-on %s: %s", addon.slug, err)
+                capture_exception(err)
+
+    @Job(
+        name="addon_manager_install",
+        conditions=ADDON_UPDATE_CONDITIONS,
+        on_condition=AddonsJobError,
+    )
+    async def install(self, slug: str) -> None:
+        """Install an add-on."""
+        self.sys_jobs.current.reference = slug
+
+        if slug in self.local:
+            raise AddonsError(f"Add-on {slug} is already installed", _LOGGER.warning)
+        store = self.store.get(slug)
+
+        if not store:
+            raise AddonsError(f"Add-on {slug} does not exist", _LOGGER.error)
+
+        store.validate_availability()
+
+        await Addon(self.coresys, slug).install()
+
+        _LOGGER.info("Add-on '%s' successfully installed", slug)
+
+    async def uninstall(self, slug: str) -> None:
+        """Remove an add-on."""
+        if slug not in self.local:
+            _LOGGER.warning("Add-on %s is not installed", slug)
+            return
+
+        await self.local[slug].uninstall()
+
+        _LOGGER.info("Add-on '%s' successfully removed", slug)
+
+    @Job(
+        name="addon_manager_update",
+        conditions=ADDON_UPDATE_CONDITIONS,
+        on_condition=AddonsJobError,
+    )
+    async def update(
+        self, slug: str, backup: bool | None = False
+    ) -> asyncio.Task | None:
+        """Update add-on.
+
+        Returns a Task that completes when addon has state 'started' (see addon.start)
+        if addon is started after update. Else nothing is returned.
+        """
+        self.sys_jobs.current.reference = slug
+
+        if slug not in self.local:
+            raise AddonsError(f"Add-on {slug} is not installed", _LOGGER.error)
+        addon = self.local[slug]
+
+        if addon.is_detached:
+            raise AddonsError(
+                f"Add-on {slug} is not available inside store", _LOGGER.error
+            )
+        store = self.store[slug]
+
+        if addon.version == store.version:
+            raise AddonsError(f"No update available for add-on {slug}", _LOGGER.warning)
+
+        # Check if available, Maybe something have changed
+        store.validate_availability()
+
+        if backup:
+            await self.sys_backups.do_backup_partial(
+                name=f"addon_{addon.slug}_{addon.version}",
+                homeassistant=False,
+                addons=[addon.slug],
+            )
+
+        return await addon.update()
+
+    @Job(
+        name="addon_manager_rebuild",
+        conditions=[
+            JobCondition.FREE_SPACE,
+            JobCondition.INTERNET_HOST,
+            JobCondition.HEALTHY,
+        ],
+        on_condition=AddonsJobError,
+    )
+    async def rebuild(self, slug: str) -> asyncio.Task | None:
+        """Perform a rebuild of local build add-on.
+
+        Returns a Task that completes when addon has state 'started' (see addon.start)
+        if addon is started after rebuild. Else nothing is returned.
+        """
+        self.sys_jobs.current.reference = slug
+
+        if slug not in self.local:
+            raise AddonsError(f"Add-on {slug} is not installed", _LOGGER.error)
+        addon = self.local[slug]
+
+        if addon.is_detached:
+            raise AddonsError(
+                f"Add-on {slug} is not available inside store", _LOGGER.error
+            )
+        store = self.store[slug]
+
+        # Check if a rebuild is possible now
+        if addon.version != store.version:
+            raise AddonsError(
+                "Version changed, use Update instead Rebuild", _LOGGER.error
+            )
+        if not addon.need_build:
+            raise AddonsNotSupportedError(
+                "Can't rebuild a image based add-on", _LOGGER.error
+            )
+
+        return await addon.rebuild()
+
+    @Job(
+        name="addon_manager_restore",
+        conditions=[
+            JobCondition.FREE_SPACE,
+            JobCondition.INTERNET_HOST,
+            JobCondition.HEALTHY,
+        ],
+        on_condition=AddonsJobError,
+    )
+    async def restore(
+        self, slug: str, tar_file: tarfile.TarFile
+    ) -> asyncio.Task | None:
+        """Restore state of an add-on.
+
+        Returns a Task that completes when addon has state 'started' (see addon.start)
+        if addon is started after restore. Else nothing is returned.
+        """
+        self.sys_jobs.current.reference = slug
+
+        if slug not in self.local:
+            _LOGGER.debug("Add-on %s is not local available for restore", slug)
+            addon = Addon(self.coresys, slug)
+            had_ingress = False
+        else:
+            _LOGGER.debug("Add-on %s is local available for restore", slug)
+            addon = self.local[slug]
+            had_ingress = addon.ingress_panel
+
+        wait_for_start = await addon.restore(tar_file)
+
+        # Check if new
+        if slug not in self.local:
+            _LOGGER.info("Detect new Add-on after restore %s", slug)
+            self.local[slug] = addon
+
+        # Update ingress
+        if had_ingress != addon.ingress_panel:
+            await self.sys_ingress.reload()
+            with suppress(HomeAssistantAPIError):
+                await self.sys_ingress.update_hass_panel(addon)
+
+        return wait_for_start
+
+    @Job(
+        name="addon_manager_repair",
+        conditions=[JobCondition.FREE_SPACE, JobCondition.INTERNET_HOST],
+    )
+    async def repair(self) -> None:
+        """Repair local add-ons."""
+        needs_repair: list[Addon] = []
+
+        # Evaluate Add-ons to repair
+        for addon in self.installed:
+            if await addon.instance.exists():
+                continue
+            needs_repair.append(addon)
+
+        _LOGGER.info("Found %d add-ons to repair", len(needs_repair))
+        if not needs_repair:
+            return
+
+        for addon in needs_repair:
+            _LOGGER.info("Repairing for add-on: %s", addon.slug)
+            with suppress(DockerError, KeyError):
+                # Need pull a image again
+                if not addon.need_build:
+                    await addon.instance.install(addon.version, addon.image)
+                    continue
+
+                # Need local lookup
+                if addon.need_build and not addon.is_detached:
+                    store = self.store[addon.slug]
+                    # If this add-on is available for rebuild
+                    if addon.version == store.version:
+                        await addon.instance.install(addon.version, addon.image)
+                        continue
+
+            _LOGGER.error("Can't repair %s", addon.slug)
+            with suppress(AddonsError):
+                await self.uninstall(addon.slug)
+
+    async def sync_dns(self) -> None:
+        """Sync add-ons DNS names."""
+        # Update hosts
+        add_host_coros: list[Awaitable[None]] = []
+        for addon in self.installed:
+            try:
+                if not await addon.instance.is_running():
+                    continue
+            except DockerError as err:
+                _LOGGER.warning("Add-on %s is corrupt: %s", addon.slug, err)
+                self.sys_resolution.create_issue(
+                    IssueType.CORRUPT_DOCKER,
+                    ContextType.ADDON,
+                    reference=addon.slug,
+                    suggestions=[SuggestionType.EXECUTE_REPAIR],
+                )
+                capture_exception(err)
+            else:
+                add_host_coros.append(
+                    self.sys_plugins.dns.add_host(
+                        ipv4=addon.ip_address, names=[addon.hostname], write=False
+                    )
+                )
+
+        await asyncio.gather(*add_host_coros)
+
+        # Write hosts files
+        with suppress(CoreDNSError):
+            await self.sys_plugins.dns.write_hosts()

supervisor/addons/model.py

@@ -1,6 +1,7 @@
 """Init file for Supervisor add-ons."""
 from abc import ABC, abstractmethod
-from collections.abc import Awaitable, Callable
+from collections import defaultdict
+from collections.abc import Callable
 from contextlib import suppress
 import logging
 from pathlib import Path
@@ -64,6 +65,7 @@ from ..const import (
     ATTR_TIMEOUT,
     ATTR_TMPFS,
     ATTR_TRANSLATIONS,
+    ATTR_TYPE,
     ATTR_UART,
     ATTR_UDEV,
     ATTR_URL,
@@ -79,24 +81,37 @@ from ..const import (
     AddonStage,
     AddonStartup,
 )
-from ..coresys import CoreSys, CoreSysAttributes
+from ..coresys import CoreSys
 from ..docker.const import Capabilities
 from ..exceptions import AddonsNotSupportedError
-from .const import ATTR_BACKUP, ATTR_CODENOTARY, AddonBackupMode
+from ..jobs.const import JOB_GROUP_ADDON
+from ..jobs.job_group import JobGroup
+from ..utils import version_is_new_enough
+from .configuration import FolderMapping
+from .const import (
+    ATTR_BACKUP,
+    ATTR_CODENOTARY,
+    ATTR_PATH,
+    ATTR_READ_ONLY,
+    AddonBackupMode,
+    MappingType,
+)
 from .options import AddonOptions, UiOptions
-from .validate import RE_SERVICE, RE_VOLUME
+from .validate import RE_SERVICE
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 Data = dict[str, Any]
 
 
-class AddonModel(CoreSysAttributes, ABC):
+class AddonModel(JobGroup, ABC):
     """Add-on Data layout."""
 
     def __init__(self, coresys: CoreSys, slug: str):
         """Initialize data holder."""
-        self.coresys: CoreSys = coresys
+        super().__init__(
+            coresys, JOB_GROUP_ADDON.format_map(defaultdict(str, slug=slug)), slug
+        )
         self.slug: str = slug
 
     @property
@@ -532,14 +547,13 @@ class AddonModel(CoreSysAttributes, ABC):
         return ATTR_IMAGE not in self.data
 
     @property
-    def map_volumes(self) -> dict[str, bool]:
-        """Return a dict of {volume: read-only} from add-on."""
+    def map_volumes(self) -> dict[MappingType, FolderMapping]:
+        """Return a dict of {MappingType: FolderMapping} from add-on."""
         volumes = {}
         for volume in self.data[ATTR_MAP]:
-            result = RE_VOLUME.match(volume)
-            if not result:
-                continue
-            volumes[result.group(1)] = result.group(2) != "rw"
+            volumes[MappingType(volume[ATTR_TYPE])] = FolderMapping(
+                volume.get(ATTR_PATH), volume[ATTR_READ_ONLY]
+            )
 
         return volumes
 
@@ -640,7 +654,9 @@ class AddonModel(CoreSysAttributes, ABC):
         # Home Assistant
         version: AwesomeVersion | None = config.get(ATTR_HOMEASSISTANT)
         with suppress(AwesomeVersionException, TypeError):
-            if self.sys_homeassistant.version < version:
+            if version and not version_is_new_enough(
+                self.sys_homeassistant.version, version
+            ):
                 raise AddonsNotSupportedError(
                     f"Add-on {self.slug} not supported on this system, requires Home Assistant version {version} or greater",
                     logger,
@@ -664,19 +680,3 @@ class AddonModel(CoreSysAttributes, ABC):
 
         # local build
         return f"{config[ATTR_REPOSITORY]}/{self.sys_arch.default}-addon-{config[ATTR_SLUG]}"
-
-    def install(self) -> Awaitable[None]:
-        """Install this add-on."""
-        return self.sys_addons.install(self.slug)
-
-    def uninstall(self) -> Awaitable[None]:
-        """Uninstall this add-on."""
-        return self.sys_addons.uninstall(self.slug)
-
-    def update(self, backup: bool | None = False) -> Awaitable[Awaitable[None] | None]:
-        """Update this add-on."""
-        return self.sys_addons.update(self.slug, backup=backup)
-
-    def rebuild(self) -> Awaitable[Awaitable[None] | None]:
-        """Rebuild this add-on."""
-        return self.sys_addons.rebuild(self.slug)

supervisor/addons/validate.py

@@ -81,6 +81,7 @@ from ..const import (
     ATTR_TIMEOUT,
     ATTR_TMPFS,
     ATTR_TRANSLATIONS,
+    ATTR_TYPE,
     ATTR_UART,
     ATTR_UDEV,
     ATTR_URL,
@@ -109,12 +110,22 @@ from ..validate import (
     uuid_match,
     version_tag,
 )
-from .const import ATTR_BACKUP, ATTR_CODENOTARY, RE_SLUG, AddonBackupMode
+from .const import (
+    ATTR_BACKUP,
+    ATTR_CODENOTARY,
+    ATTR_PATH,
+    ATTR_READ_ONLY,
+    RE_SLUG,
+    AddonBackupMode,
+    MappingType,
+)
 from .options import RE_SCHEMA_ELEMENT
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
-RE_VOLUME = re.compile(r"^(config|ssl|addons|backup|share|media)(?::(rw|ro))?$")
+RE_VOLUME = re.compile(
+    r"^(data|config|ssl|addons|backup|share|media|homeassistant_config|all_addon_configs|addon_config)(?::(rw|ro))?$"
+)
 RE_SERVICE = re.compile(r"^(?P<service>mqtt|mysql):(?P<rights>provide|want|need)$")
 
 
@@ -143,6 +154,9 @@ RE_MACHINE = re.compile(
     r"|raspberrypi3"
     r"|raspberrypi4-64"
     r"|raspberrypi4"
+    r"|raspberrypi5-64"
+    r"|yellow"
+    r"|green"
     r"|tinker"
     r")$"
 )
@@ -175,6 +189,20 @@ def _warn_addon_config(config: dict[str, Any]):
             name,
         )
 
+    invalid_services: list[str] = []
+    for service in config.get(ATTR_DISCOVERY, []):
+        try:
+            valid_discovery_service(service)
+        except vol.Invalid:
+            invalid_services.append(service)
+
+    if invalid_services:
+        _LOGGER.warning(
+            "Add-on lists the following unknown services for discovery: %s. Please report this to the maintainer of %s",
+            ", ".join(invalid_services),
+            name,
+        )
+
     return config
 
 
@@ -196,9 +224,9 @@ def _migrate_addon_config(protocol=False):
                     name,
                 )
             if value == "before":
-                config[ATTR_STARTUP] = AddonStartup.SERVICES.value
+                config[ATTR_STARTUP] = AddonStartup.SERVICES
             elif value == "after":
-                config[ATTR_STARTUP] = AddonStartup.APPLICATION.value
+                config[ATTR_STARTUP] = AddonStartup.APPLICATION
 
         # UART 2021-01-20
         if "auto_uart" in config:
@@ -244,6 +272,48 @@ def _migrate_addon_config(protocol=False):
                     name,
                 )
 
+        # 2023-11 "map" entries can also be dict to allow path configuration
+        volumes = []
+        for entry in config.get(ATTR_MAP, []):
+            if isinstance(entry, dict):
+                volumes.append(entry)
+            if isinstance(entry, str):
+                result = RE_VOLUME.match(entry)
+                if not result:
+                    continue
+                volumes.append(
+                    {
+                        ATTR_TYPE: result.group(1),
+                        ATTR_READ_ONLY: result.group(2) != "rw",
+                    }
+                )
+
+        if volumes:
+            config[ATTR_MAP] = volumes
+
+        # 2023-10 "config" became "homeassistant" so /config can be used for addon's public config
+        if any(volume[ATTR_TYPE] == MappingType.CONFIG for volume in volumes):
+            if any(
+                volume
+                and volume[ATTR_TYPE]
+                in {MappingType.ADDON_CONFIG, MappingType.HOMEASSISTANT_CONFIG}
+                for volume in volumes
+            ):
+                _LOGGER.warning(
+                    "Add-on config using incompatible map options, '%s' and '%s' are ignored if '%s' is included. Please report this to the maintainer of %s",
+                    MappingType.ADDON_CONFIG,
+                    MappingType.HOMEASSISTANT_CONFIG,
+                    MappingType.CONFIG,
+                    name,
+                )
+            else:
+                _LOGGER.debug(
+                    "Add-on config using deprecated map option '%s' instead of '%s'. Please report this to the maintainer of %s",
+                    MappingType.CONFIG,
+                    MappingType.HOMEASSISTANT_CONFIG,
+                    name,
+                )
+
         return config
 
     return _migrate
@@ -292,7 +362,15 @@ _SCHEMA_ADDON_CONFIG = vol.Schema(
         vol.Optional(ATTR_DEVICES): [str],
         vol.Optional(ATTR_UDEV, default=False): vol.Boolean(),
         vol.Optional(ATTR_TMPFS, default=False): vol.Boolean(),
-        vol.Optional(ATTR_MAP, default=list): [vol.Match(RE_VOLUME)],
+        vol.Optional(ATTR_MAP, default=list): [
+            vol.Schema(
+                {
+                    vol.Required(ATTR_TYPE): vol.Coerce(MappingType),
+                    vol.Optional(ATTR_READ_ONLY, default=True): bool,
+                    vol.Optional(ATTR_PATH): str,
+                }
+            )
+        ],
         vol.Optional(ATTR_ENVIRONMENT): {vol.Match(r"\w*"): str},
         vol.Optional(ATTR_PRIVILEGED): [vol.Coerce(Capabilities)],
         vol.Optional(ATTR_APPARMOR, default=True): vol.Boolean(),
@@ -313,7 +391,7 @@ _SCHEMA_ADDON_CONFIG = vol.Schema(
         vol.Optional(ATTR_DOCKER_API, default=False): vol.Boolean(),
         vol.Optional(ATTR_AUTH_API, default=False): vol.Boolean(),
         vol.Optional(ATTR_SERVICES): [vol.Match(RE_SERVICE)],
-        vol.Optional(ATTR_DISCOVERY): [valid_discovery_service],
+        vol.Optional(ATTR_DISCOVERY): [str],
         vol.Optional(ATTR_BACKUP_EXCLUDE): [str],
         vol.Optional(ATTR_BACKUP_PRE): str,
         vol.Optional(ATTR_BACKUP_POST): str,

supervisor/api/__init__.py

@@ -5,6 +5,7 @@ from pathlib import Path
 from typing import Any
 
 from aiohttp import web
+from aiohttp_fast_url_dispatcher import FastUrlDispatcher, attach_fast_url_dispatcher
 
 from ..const import AddonState
 from ..coresys import CoreSys, CoreSysAttributes
@@ -64,9 +65,10 @@ class RestAPI(CoreSysAttributes):
                 "max_field_size": MAX_LINE_SIZE,
             },
         )
+        attach_fast_url_dispatcher(self.webapp, FastUrlDispatcher())
 
         # service stuff
-        self._runner: web.AppRunner = web.AppRunner(self.webapp)
+        self._runner: web.AppRunner = web.AppRunner(self.webapp, shutdown_timeout=5)
         self._site: web.TCPSite | None = None
 
     async def load(self) -> None:
@@ -186,6 +188,8 @@ class RestAPI(CoreSysAttributes):
         # Boards endpoints
         self.webapp.add_routes(
             [
+                web.get("/os/boards/green", api_os.boards_green_info),
+                web.post("/os/boards/green", api_os.boards_green_options),
                 web.get("/os/boards/yellow", api_os.boards_yellow_info),
                 web.post("/os/boards/yellow", api_os.boards_yellow_options),
                 web.get("/os/boards/{board}", api_os.boards_other_info),
@@ -485,6 +489,8 @@ class RestAPI(CoreSysAttributes):
                 web.get("/backups/info", api_backups.info),
                 web.post("/backups/options", api_backups.options),
                 web.post("/backups/reload", api_backups.reload),
+                web.post("/backups/freeze", api_backups.freeze),
+                web.post("/backups/thaw", api_backups.thaw),
                 web.post("/backups/new/full", api_backups.backup_full),
                 web.post("/backups/new/partial", api_backups.backup_partial),
                 web.post("/backups/new/upload", api_backups.upload),
@@ -667,9 +673,7 @@ class RestAPI(CoreSysAttributes):
     async def start(self) -> None:
         """Run RESTful API webserver."""
         await self._runner.setup()
-        self._site = web.TCPSite(
-            self._runner, host="0.0.0.0", port=80, shutdown_timeout=5
-        )
+        self._site = web.TCPSite(self._runner, host="0.0.0.0", port=80)
 
         try:
             await self._site.start()

supervisor/api/addons.py

@@ -8,8 +8,8 @@ from aiohttp import web
 import voluptuous as vol
 from voluptuous.humanize import humanize_error
 
-from ..addons import AnyAddon
 from ..addons.addon import Addon
+from ..addons.manager import AnyAddon
 from ..addons.utils import rating_security
 from ..const import (
     ATTR_ADDONS,
@@ -388,7 +388,7 @@ class APIAddons(CoreSysAttributes):
     def uninstall(self, request: web.Request) -> Awaitable[None]:
         """Uninstall add-on."""
         addon = self._extract_addon(request)
-        return asyncio.shield(addon.uninstall())
+        return asyncio.shield(self.sys_addons.uninstall(addon.slug))
 
     @api_process
     async def start(self, request: web.Request) -> None:
@@ -414,7 +414,7 @@ class APIAddons(CoreSysAttributes):
     async def rebuild(self, request: web.Request) -> None:
         """Rebuild local build add-on."""
         addon = self._extract_addon(request)
-        if start_task := await asyncio.shield(addon.rebuild()):
+        if start_task := await asyncio.shield(self.sys_addons.rebuild(addon.slug)):
             await start_task
 
     @api_process_raw(CONTENT_TYPE_BINARY)

supervisor/api/audio.py

@@ -1,11 +1,11 @@
 """Init file for Supervisor Audio RESTful API."""
 import asyncio
 from collections.abc import Awaitable
+from dataclasses import asdict
 import logging
 from typing import Any
 
 from aiohttp import web
-import attr
 import voluptuous as vol
 
 from ..const import (
@@ -76,15 +76,11 @@ class APIAudio(CoreSysAttributes):
             ATTR_UPDATE_AVAILABLE: self.sys_plugins.audio.need_update,
             ATTR_HOST: str(self.sys_docker.network.audio),
             ATTR_AUDIO: {
-                ATTR_CARD: [attr.asdict(card) for card in self.sys_host.sound.cards],
-                ATTR_INPUT: [
-                    attr.asdict(stream) for stream in self.sys_host.sound.inputs
-                ],
-                ATTR_OUTPUT: [
-                    attr.asdict(stream) for stream in self.sys_host.sound.outputs
-                ],
+                ATTR_CARD: [asdict(card) for card in self.sys_host.sound.cards],
+                ATTR_INPUT: [asdict(stream) for stream in self.sys_host.sound.inputs],
+                ATTR_OUTPUT: [asdict(stream) for stream in self.sys_host.sound.outputs],
                 ATTR_APPLICATION: [
-                    attr.asdict(stream) for stream in self.sys_host.sound.applications
+                    asdict(stream) for stream in self.sys_host.sound.applications
                 ],
             },
         }

supervisor/api/auth.py

@@ -11,6 +11,7 @@ from ..addons.addon import Addon
 from ..const import ATTR_PASSWORD, ATTR_USERNAME, REQUEST_FROM
 from ..coresys import CoreSysAttributes
 from ..exceptions import APIForbidden
+from ..utils.json import json_loads
 from .const import CONTENT_TYPE_JSON, CONTENT_TYPE_URL
 from .utils import api_process, api_validate
 
@@ -67,7 +68,7 @@ class APIAuth(CoreSysAttributes):
 
         # Json
         if request.headers.get(CONTENT_TYPE) == CONTENT_TYPE_JSON:
-            data = await request.json()
+            data = await request.json(loads=json_loads)
             return await self._process_dict(request, addon, data)
 
         # URL encoded

supervisor/api/backups.py

@@ -1,5 +1,6 @@
 """Backups RESTful API."""
 import asyncio
+import errno
 import logging
 from pathlib import Path
 import re
@@ -20,6 +21,7 @@ from ..const import (
     ATTR_DAYS_UNTIL_STALE,
     ATTR_FOLDERS,
     ATTR_HOMEASSISTANT,
+    ATTR_HOMEASSISTANT_EXCLUDE_DATABASE,
     ATTR_LOCATON,
     ATTR_NAME,
     ATTR_PASSWORD,
@@ -28,12 +30,14 @@ from ..const import (
     ATTR_SIZE,
     ATTR_SLUG,
     ATTR_SUPERVISOR_VERSION,
+    ATTR_TIMEOUT,
     ATTR_TYPE,
     ATTR_VERSION,
 )
 from ..coresys import CoreSysAttributes
 from ..exceptions import APIError
 from ..mounts.const import MountUsage
+from ..resolution.const import UnhealthyReason
 from .const import CONTENT_TYPE_TAR
 from .utils import api_process, api_validate
 
@@ -63,6 +67,7 @@ SCHEMA_BACKUP_FULL = vol.Schema(
         vol.Optional(ATTR_PASSWORD): vol.Maybe(str),
         vol.Optional(ATTR_COMPRESSED): vol.Maybe(vol.Boolean()),
         vol.Optional(ATTR_LOCATON): vol.Maybe(str),
+        vol.Optional(ATTR_HOMEASSISTANT_EXCLUDE_DATABASE): vol.Boolean(),
     }
 )
 
@@ -80,6 +85,12 @@ SCHEMA_OPTIONS = vol.Schema(
     }
 )
 
+SCHEMA_FREEZE = vol.Schema(
+    {
+        vol.Optional(ATTR_TIMEOUT): vol.All(int, vol.Range(min=1)),
+    }
+)
+
 
 class APIBackups(CoreSysAttributes):
     """Handle RESTful API for backups functions."""
@@ -142,7 +153,7 @@ class APIBackups(CoreSysAttributes):
         self.sys_backups.save_data()
 
     @api_process
-    async def reload(self, request):
+    async def reload(self, _):
         """Reload backup list."""
         await asyncio.shield(self.sys_backups.reload())
         return True
@@ -177,6 +188,7 @@ class APIBackups(CoreSysAttributes):
             ATTR_ADDONS: data_addons,
             ATTR_REPOSITORIES: backup.repositories,
             ATTR_FOLDERS: backup.folders,
+            ATTR_HOMEASSISTANT_EXCLUDE_DATABASE: backup.homeassistant_exclude_database,
         }
 
     def _location_to_mount(self, body: dict[str, Any]) -> dict[str, Any]:
@@ -233,6 +245,17 @@ class APIBackups(CoreSysAttributes):
 
         return await asyncio.shield(self.sys_backups.do_restore_partial(backup, **body))
 
+    @api_process
+    async def freeze(self, request):
+        """Initiate manual freeze for external backup."""
+        body = await api_validate(SCHEMA_FREEZE, request)
+        await asyncio.shield(self.sys_backups.freeze_all(**body))
+
+    @api_process
+    async def thaw(self, request):
+        """Begin thaw after manual freeze."""
+        await self.sys_backups.thaw_all()
+
     @api_process
     async def remove(self, request):
         """Remove a backup."""
@@ -267,6 +290,8 @@ class APIBackups(CoreSysAttributes):
                         backup.write(chunk)
 
             except OSError as err:
+                if err.errno == errno.EBADMSG:
+                    self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
                 _LOGGER.error("Can't write new backup file: %s", err)
                 return False
 

supervisor/api/const.py

@@ -23,7 +23,6 @@ ATTR_CONNECTION_BUS = "connection_bus"
 ATTR_DATA_DISK = "data_disk"
 ATTR_DEVICE = "device"
 ATTR_DEV_PATH = "dev_path"
-ATTR_DISK_LED = "disk_led"
 ATTR_DISKS = "disks"
 ATTR_DRIVES = "drives"
 ATTR_DT_SYNCHRONIZED = "dt_synchronized"
@@ -31,8 +30,8 @@ ATTR_DT_UTC = "dt_utc"
 ATTR_EJECTABLE = "ejectable"
 ATTR_FALLBACK = "fallback"
 ATTR_FILESYSTEMS = "filesystems"
-ATTR_HEARTBEAT_LED = "heartbeat_led"
 ATTR_IDENTIFIERS = "identifiers"
+ATTR_JOBS = "jobs"
 ATTR_LLMNR = "llmnr"
 ATTR_LLMNR_HOSTNAME = "llmnr_hostname"
 ATTR_MDNS = "mdns"
@@ -40,7 +39,6 @@ ATTR_MODEL = "model"
 ATTR_MOUNTS = "mounts"
 ATTR_MOUNT_POINTS = "mount_points"
 ATTR_PANEL_PATH = "panel_path"
-ATTR_POWER_LED = "power_led"
 ATTR_REMOVABLE = "removable"
 ATTR_REVISION = "revision"
 ATTR_SEAT = "seat"
@@ -48,6 +46,7 @@ ATTR_SIGNED = "signed"
 ATTR_STARTUP_TIME = "startup_time"
 ATTR_SUBSYSTEM = "subsystem"
 ATTR_SYSFS = "sysfs"
+ATTR_SYSTEM_HEALTH_LED = "system_health_led"
 ATTR_TIME_DETECTED = "time_detected"
 ATTR_UPDATE_TYPE = "update_type"
 ATTR_USE_NTP = "use_ntp"

supervisor/api/discovery.py

@@ -1,6 +1,9 @@
 """Init file for Supervisor network RESTful API."""
+import logging
+
 import voluptuous as vol
 
+from ..addons.addon import Addon
 from ..const import (
     ATTR_ADDON,
     ATTR_CONFIG,
@@ -9,15 +12,18 @@ from ..const import (
     ATTR_SERVICES,
     ATTR_UUID,
     REQUEST_FROM,
+    AddonState,
 )
 from ..coresys import CoreSysAttributes
 from ..discovery.validate import valid_discovery_service
 from ..exceptions import APIError, APIForbidden
 from .utils import api_process, api_validate, require_home_assistant
 
+_LOGGER: logging.Logger = logging.getLogger(__name__)
+
 SCHEMA_DISCOVERY = vol.Schema(
     {
-        vol.Required(ATTR_SERVICE): valid_discovery_service,
+        vol.Required(ATTR_SERVICE): str,
         vol.Optional(ATTR_CONFIG): vol.Maybe(dict),
     }
 )
@@ -36,19 +42,19 @@ class APIDiscovery(CoreSysAttributes):
     @api_process
     @require_home_assistant
     async def list(self, request):
-        """Show register services."""
-
+        """Show registered and available services."""
         # Get available discovery
-        discovery = []
-        for message in self.sys_discovery.list_messages:
-            discovery.append(
-                {
-                    ATTR_ADDON: message.addon,
-                    ATTR_SERVICE: message.service,
-                    ATTR_UUID: message.uuid,
-                    ATTR_CONFIG: message.config,
-                }
-            )
+        discovery = [
+            {
+                ATTR_ADDON: message.addon,
+                ATTR_SERVICE: message.service,
+                ATTR_UUID: message.uuid,
+                ATTR_CONFIG: message.config,
+            }
+            for message in self.sys_discovery.list_messages
+            if (addon := self.sys_addons.get(message.addon, local_only=True))
+            and addon.state == AddonState.STARTED
+        ]
 
         # Get available services/add-ons
         services = {}
@@ -62,11 +68,28 @@ class APIDiscovery(CoreSysAttributes):
     async def set_discovery(self, request):
         """Write data into a discovery pipeline."""
         body = await api_validate(SCHEMA_DISCOVERY, request)
-        addon = request[REQUEST_FROM]
+        addon: Addon = request[REQUEST_FROM]
+        service = body[ATTR_SERVICE]
+
+        try:
+            valid_discovery_service(service)
+        except vol.Invalid:
+            _LOGGER.warning(
+                "Received discovery message for unknown service %s from addon %s. Please report this to the maintainer of the add-on",
+                service,
+                addon.name,
+            )
 
         # Access?
         if body[ATTR_SERVICE] not in addon.discovery:
-            raise APIForbidden("Can't use discovery!")
+            _LOGGER.error(
+                "Add-on %s attempted to send discovery for service %s which is not listed in its config. Please report this to the maintainer of the add-on",
+                addon.name,
+                service,
+            )
+            raise APIForbidden(
+                "Add-ons must list services they provide via discovery in their config!"
+            )
 
         # Process discovery message
         message = self.sys_discovery.send(addon, **body)

supervisor/api/homeassistant.py

@@ -12,6 +12,7 @@ from ..const import (
     ATTR_AUDIO_INPUT,
     ATTR_AUDIO_OUTPUT,
     ATTR_BACKUP,
+    ATTR_BACKUPS_EXCLUDE_DATABASE,
     ATTR_BLK_READ,
     ATTR_BLK_WRITE,
     ATTR_BOOT,
@@ -51,6 +52,7 @@ SCHEMA_OPTIONS = vol.Schema(
         vol.Optional(ATTR_REFRESH_TOKEN): vol.Maybe(str),
         vol.Optional(ATTR_AUDIO_OUTPUT): vol.Maybe(str),
         vol.Optional(ATTR_AUDIO_INPUT): vol.Maybe(str),
+        vol.Optional(ATTR_BACKUPS_EXCLUDE_DATABASE): vol.Boolean(),
     }
 )
 
@@ -82,6 +84,7 @@ class APIHomeAssistant(CoreSysAttributes):
             ATTR_WATCHDOG: self.sys_homeassistant.watchdog,
             ATTR_AUDIO_INPUT: self.sys_homeassistant.audio_input,
             ATTR_AUDIO_OUTPUT: self.sys_homeassistant.audio_output,
+            ATTR_BACKUPS_EXCLUDE_DATABASE: self.sys_homeassistant.backups_exclude_database,
         }
 
     @api_process
@@ -113,6 +116,11 @@ class APIHomeAssistant(CoreSysAttributes):
         if ATTR_AUDIO_OUTPUT in body:
             self.sys_homeassistant.audio_output = body[ATTR_AUDIO_OUTPUT]
 
+        if ATTR_BACKUPS_EXCLUDE_DATABASE in body:
+            self.sys_homeassistant.backups_exclude_database = body[
+                ATTR_BACKUPS_EXCLUDE_DATABASE
+            ]
+
         self.sys_homeassistant.save_data()
 
     @api_process

supervisor/api/ingress.py

@@ -21,11 +21,18 @@ from ..const import (
     ATTR_ICON,
     ATTR_PANELS,
     ATTR_SESSION,
+    ATTR_SESSION_DATA_USER_ID,
     ATTR_TITLE,
+    HEADER_REMOTE_USER_DISPLAY_NAME,
+    HEADER_REMOTE_USER_ID,
+    HEADER_REMOTE_USER_NAME,
     HEADER_TOKEN,
     HEADER_TOKEN_OLD,
+    IngressSessionData,
+    IngressSessionDataUser,
 )
 from ..coresys import CoreSysAttributes
+from ..exceptions import HomeAssistantAPIError
 from .const import COOKIE_INGRESS
 from .utils import api_process, api_validate, require_home_assistant
 
@@ -33,10 +40,46 @@ _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 VALIDATE_SESSION_DATA = vol.Schema({ATTR_SESSION: str})
 
+"""Expected optional payload of create session request"""
+SCHEMA_INGRESS_CREATE_SESSION_DATA = vol.Schema(
+    {
+        vol.Optional(ATTR_SESSION_DATA_USER_ID): str,
+    }
+)
+
+
+# from https://github.com/aio-libs/aiohttp/blob/8ae650bee4add9f131d49b96a0a150311ea58cd1/aiohttp/helpers.py#L1059C1-L1079C1
+def must_be_empty_body(method: str, code: int) -> bool:
+    """Check if a request must return an empty body."""
+    return (
+        status_code_must_be_empty_body(code)
+        or method_must_be_empty_body(method)
+        or (200 <= code < 300 and method.upper() == hdrs.METH_CONNECT)
+    )
+
+
+def method_must_be_empty_body(method: str) -> bool:
+    """Check if a method must return an empty body."""
+    # https://datatracker.ietf.org/doc/html/rfc9112#section-6.3-2.1
+    # https://datatracker.ietf.org/doc/html/rfc9112#section-6.3-2.2
+    return method.upper() == hdrs.METH_HEAD
+
+
+def status_code_must_be_empty_body(code: int) -> bool:
+    """Check if a status code must return an empty body."""
+    # https://datatracker.ietf.org/doc/html/rfc9112#section-6.3-2.1
+    return code in {204, 304} or 100 <= code < 200
+
 
 class APIIngress(CoreSysAttributes):
     """Ingress view to handle add-on webui routing."""
 
+    _list_of_users: list[IngressSessionDataUser]
+
+    def __init__(self) -> None:
+        """Initialize APIIngress."""
+        self._list_of_users = []
+
     def _extract_addon(self, request: web.Request) -> Addon:
         """Return addon, throw an exception it it doesn't exist."""
         token = request.match_info.get("token")
@@ -71,7 +114,19 @@ class APIIngress(CoreSysAttributes):
     @require_home_assistant
     async def create_session(self, request: web.Request) -> dict[str, Any]:
         """Create a new session."""
-        session = self.sys_ingress.create_session()
+        schema_ingress_config_session_data = await api_validate(
+            SCHEMA_INGRESS_CREATE_SESSION_DATA, request
+        )
+        data: IngressSessionData | None = None
+
+        if ATTR_SESSION_DATA_USER_ID in schema_ingress_config_session_data:
+            user = await self._find_user_by_id(
+                schema_ingress_config_session_data[ATTR_SESSION_DATA_USER_ID]
+            )
+            if user:
+                data = IngressSessionData(user)
+
+        session = self.sys_ingress.create_session(data)
         return {ATTR_SESSION: session}
 
     @api_process
@@ -99,13 +154,14 @@ class APIIngress(CoreSysAttributes):
         # Process requests
         addon = self._extract_addon(request)
         path = request.match_info.get("path")
+        session_data = self.sys_ingress.get_session_data(session)
         try:
             # Websocket
             if _is_websocket(request):
-                return await self._handle_websocket(request, addon, path)
+                return await self._handle_websocket(request, addon, path, session_data)
 
             # Request
-            return await self._handle_request(request, addon, path)
+            return await self._handle_request(request, addon, path, session_data)
 
         except aiohttp.ClientError as err:
             _LOGGER.error("Ingress error: %s", err)
@@ -113,7 +169,11 @@ class APIIngress(CoreSysAttributes):
         raise HTTPBadGateway()
 
     async def _handle_websocket(
-        self, request: web.Request, addon: Addon, path: str
+        self,
+        request: web.Request,
+        addon: Addon,
+        path: str,
+        session_data: IngressSessionData | None,
     ) -> web.WebSocketResponse:
         """Ingress route for websocket."""
         if hdrs.SEC_WEBSOCKET_PROTOCOL in request.headers:
@@ -131,7 +191,7 @@ class APIIngress(CoreSysAttributes):
 
         # Preparing
         url = self._create_url(addon, path)
-        source_header = _init_header(request, addon)
+        source_header = _init_header(request, addon, session_data)
 
         # Support GET query
         if request.query_string:
@@ -157,11 +217,15 @@ class APIIngress(CoreSysAttributes):
         return ws_server
 
     async def _handle_request(
-        self, request: web.Request, addon: Addon, path: str
+        self,
+        request: web.Request,
+        addon: Addon,
+        path: str,
+        session_data: IngressSessionData | None,
     ) -> web.Response | web.StreamResponse:
         """Ingress route for request."""
         url = self._create_url(addon, path)
-        source_header = _init_header(request, addon)
+        source_header = _init_header(request, addon, session_data)
 
         # Passing the raw stream breaks requests for some webservers
         # since we just need it for POST requests really, for all other methods
@@ -184,10 +248,18 @@ class APIIngress(CoreSysAttributes):
             skip_auto_headers={hdrs.CONTENT_TYPE},
         ) as result:
             headers = _response_header(result)
-
+            # Avoid parsing content_type in simple cases for better performance
+            if maybe_content_type := result.headers.get(hdrs.CONTENT_TYPE):
+                content_type = (maybe_content_type.partition(";"))[0].strip()
+            else:
+                content_type = result.content_type
             # Simple request
             if (
-                hdrs.CONTENT_LENGTH in result.headers
+                # empty body responses should not be streamed,
+                # otherwise aiohttp < 3.9.0 may generate
+                # an invalid "0\r\n\r\n" chunk instead of an empty response.
+                must_be_empty_body(request.method, result.status)
+                or hdrs.CONTENT_LENGTH in result.headers
                 and int(result.headers.get(hdrs.CONTENT_LENGTH, 0)) < 4_194_000
             ):
                 # Return Response
@@ -195,13 +267,13 @@ class APIIngress(CoreSysAttributes):
                 return web.Response(
                     headers=headers,
                     status=result.status,
-                    content_type=result.content_type,
+                    content_type=content_type,
                     body=body,
                 )
 
             # Stream response
             response = web.StreamResponse(status=result.status, headers=headers)
-            response.content_type = result.content_type
+            response.content_type = content_type
 
             try:
                 await response.prepare(request)
@@ -217,11 +289,35 @@ class APIIngress(CoreSysAttributes):
 
             return response
 
+    async def _find_user_by_id(self, user_id: str) -> IngressSessionDataUser | None:
+        """Find user object by the user's ID."""
+        try:
+            list_of_users = await self.sys_homeassistant.get_users()
+        except (HomeAssistantAPIError, TypeError) as err:
+            _LOGGER.error(
+                "%s error occurred while requesting list of users: %s", type(err), err
+            )
+            return None
 
-def _init_header(request: web.Request, addon: str) -> CIMultiDict | dict[str, str]:
+        if list_of_users is not None:
+            self._list_of_users = list_of_users
+
+        return next((user for user in self._list_of_users if user.id == user_id), None)
+
+
+def _init_header(
+    request: web.Request, addon: Addon, session_data: IngressSessionData | None
+) -> CIMultiDict | dict[str, str]:
     """Create initial header."""
     headers = {}
 
+    if session_data is not None:
+        headers[HEADER_REMOTE_USER_ID] = session_data.user.id
+        if session_data.user.username is not None:
+            headers[HEADER_REMOTE_USER_NAME] = session_data.user.username
+        if session_data.user.display_name is not None:
+            headers[HEADER_REMOTE_USER_DISPLAY_NAME] = session_data.user.display_name
+
     # filter flags
     for name, value in request.headers.items():
         if name in (
@@ -234,6 +330,9 @@ def _init_header(request: web.Request, addon: str) -> CIMultiDict | dict[str, st
             hdrs.SEC_WEBSOCKET_KEY,
             istr(HEADER_TOKEN),
             istr(HEADER_TOKEN_OLD),
+            istr(HEADER_REMOTE_USER_ID),
+            istr(HEADER_REMOTE_USER_NAME),
+            istr(HEADER_REMOTE_USER_DISPLAY_NAME),
         ):
             continue
         headers[name] = value

supervisor/api/jobs.py

@@ -6,7 +6,9 @@ from aiohttp import web
 import voluptuous as vol
 
 from ..coresys import CoreSysAttributes
+from ..jobs import SupervisorJob
 from ..jobs.const import ATTR_IGNORE_CONDITIONS, JobCondition
+from .const import ATTR_JOBS
 from .utils import api_process, api_validate
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
@@ -19,11 +21,45 @@ SCHEMA_OPTIONS = vol.Schema(
 class APIJobs(CoreSysAttributes):
     """Handle RESTful API for OS functions."""
 
+    def _list_jobs(self) -> list[dict[str, Any]]:
+        """Return current job tree."""
+        jobs_by_parent: dict[str | None, list[SupervisorJob]] = {}
+        for job in self.sys_jobs.jobs:
+            if job.internal:
+                continue
+
+            if job.parent_id not in jobs_by_parent:
+                jobs_by_parent[job.parent_id] = [job]
+            else:
+                jobs_by_parent[job.parent_id].append(job)
+
+        job_list: list[dict[str, Any]] = []
+        queue: list[tuple[list[dict[str, Any]], SupervisorJob]] = [
+            (job_list, job) for job in jobs_by_parent.get(None, [])
+        ]
+
+        while queue:
+            (current_list, current_job) = queue.pop(0)
+            child_jobs: list[dict[str, Any]] = []
+
+            # We remove parent_id and instead use that info to represent jobs as a tree
+            job_dict = current_job.as_dict() | {"child_jobs": child_jobs}
+            job_dict.pop("parent_id")
+            current_list.append(job_dict)
+
+            if current_job.uuid in jobs_by_parent:
+                queue.extend(
+                    [(child_jobs, job) for job in jobs_by_parent.get(current_job.uuid)]
+                )
+
+        return job_list
+
     @api_process
     async def info(self, request: web.Request) -> dict[str, Any]:
         """Return JobManager information."""
         return {
             ATTR_IGNORE_CONDITIONS: self.sys_jobs.ignore_conditions,
+            ATTR_JOBS: self._list_jobs(),
         }
 
     @api_process

supervisor/api/middleware/security.py

@@ -19,6 +19,7 @@ from ...const import (
     CoreState,
 )
 from ...coresys import CoreSys, CoreSysAttributes
+from ...utils import version_is_new_enough
 from ..utils import api_return_error, excract_supervisor_token
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
@@ -195,7 +196,7 @@ class SecurityMiddleware(CoreSysAttributes):
             CoreState.FREEZE,
         ):
             return api_return_error(
-                message=f"System is not ready with state: {self.sys_core.state.value}"
+                message=f"System is not ready with state: {self.sys_core.state}"
             )
 
         return await handler(request)
@@ -273,9 +274,8 @@ class SecurityMiddleware(CoreSysAttributes):
     @middleware
     async def core_proxy(self, request: Request, handler: RequestHandler) -> Response:
         """Validate user from Core API proxy."""
-        if (
-            request[REQUEST_FROM] != self.sys_homeassistant
-            or self.sys_homeassistant.version >= _CORE_VERSION
+        if request[REQUEST_FROM] != self.sys_homeassistant or version_is_new_enough(
+            self.sys_homeassistant.version, _CORE_VERSION
         ):
             return await handler(request)
 

supervisor/api/network.py

@@ -1,11 +1,11 @@
 """REST API for network."""
 import asyncio
 from collections.abc import Awaitable
+from dataclasses import replace
 from ipaddress import ip_address, ip_interface
 from typing import Any
 
 from aiohttp import web
-import attr
 import voluptuous as vol
 
 from ..const import (
@@ -43,8 +43,7 @@ from ..const import (
 )
 from ..coresys import CoreSysAttributes
 from ..exceptions import APIError, HostNetworkNotFound
-from ..host.const import AuthMethod, InterfaceType, WifiMode
-from ..host.network import (
+from ..host.configuration import (
     AccessPoint,
     Interface,
     InterfaceMethod,
@@ -52,6 +51,7 @@ from ..host.network import (
     VlanConfig,
     WifiConfig,
 )
+from ..host.const import AuthMethod, InterfaceType, WifiMode
 from .utils import api_process, api_validate
 
 _SCHEMA_IP_CONFIG = vol.Schema(
@@ -121,6 +121,7 @@ def interface_struct(interface: Interface) -> dict[str, Any]:
         ATTR_ENABLED: interface.enabled,
         ATTR_CONNECTED: interface.connected,
         ATTR_PRIMARY: interface.primary,
+        ATTR_MAC: interface.mac,
         ATTR_IPV4: ipconfig_struct(interface.ipv4) if interface.ipv4 else None,
         ATTR_IPV6: ipconfig_struct(interface.ipv6) if interface.ipv6 else None,
         ATTR_WIFI: wifi_struct(interface.wifi) if interface.wifi else None,
@@ -196,19 +197,19 @@ class APINetwork(CoreSysAttributes):
         # Apply config
         for key, config in body.items():
             if key == ATTR_IPV4:
-                interface.ipv4 = attr.evolve(
+                interface.ipv4 = replace(
                     interface.ipv4
                     or IpConfig(InterfaceMethod.STATIC, [], None, [], None),
                     **config,
                 )
             elif key == ATTR_IPV6:
-                interface.ipv6 = attr.evolve(
+                interface.ipv6 = replace(
                     interface.ipv6
                     or IpConfig(InterfaceMethod.STATIC, [], None, [], None),
                     **config,
                 )
             elif key == ATTR_WIFI:
-                interface.wifi = attr.evolve(
+                interface.wifi = replace(
                     interface.wifi
                     or WifiConfig(
                         WifiMode.INFRASTRUCTURE, "", AuthMethod.OPEN, None, None
@@ -276,6 +277,8 @@ class APINetwork(CoreSysAttributes):
             )
 
         vlan_interface = Interface(
+            "",
+            "",
             "",
             True,
             True,

supervisor/api/os.py

@@ -8,11 +8,15 @@ from aiohttp import web
 import voluptuous as vol
 
 from ..const import (
+    ATTR_ACTIVITY_LED,
     ATTR_BOARD,
     ATTR_BOOT,
     ATTR_DEVICES,
+    ATTR_DISK_LED,
+    ATTR_HEARTBEAT_LED,
     ATTR_ID,
     ATTR_NAME,
+    ATTR_POWER_LED,
     ATTR_SERIAL,
     ATTR_SIZE,
     ATTR_UPDATE_AVAILABLE,
@@ -27,21 +31,19 @@ from .const import (
     ATTR_DATA_DISK,
     ATTR_DEV_PATH,
     ATTR_DEVICE,
-    ATTR_DISK_LED,
     ATTR_DISKS,
-    ATTR_HEARTBEAT_LED,
     ATTR_MODEL,
-    ATTR_POWER_LED,
+    ATTR_SYSTEM_HEALTH_LED,
     ATTR_VENDOR,
 )
 from .utils import api_process, api_validate
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
+# pylint: disable=no-value-for-parameter
 SCHEMA_VERSION = vol.Schema({vol.Optional(ATTR_VERSION): version_tag})
 SCHEMA_DISK = vol.Schema({vol.Required(ATTR_DEVICE): str})
 
-# pylint: disable=no-value-for-parameter
 SCHEMA_YELLOW_OPTIONS = vol.Schema(
     {
         vol.Optional(ATTR_DISK_LED): vol.Boolean(),
@@ -49,6 +51,14 @@ SCHEMA_YELLOW_OPTIONS = vol.Schema(
         vol.Optional(ATTR_POWER_LED): vol.Boolean(),
     }
 )
+SCHEMA_GREEN_OPTIONS = vol.Schema(
+    {
+        vol.Optional(ATTR_ACTIVITY_LED): vol.Boolean(),
+        vol.Optional(ATTR_POWER_LED): vol.Boolean(),
+        vol.Optional(ATTR_SYSTEM_HEALTH_LED): vol.Boolean(),
+    }
+)
+# pylint: enable=no-value-for-parameter
 
 
 class APIOS(CoreSysAttributes):
@@ -105,6 +115,31 @@ class APIOS(CoreSysAttributes):
             ],
         }
 
+    @api_process
+    async def boards_green_info(self, request: web.Request) -> dict[str, Any]:
+        """Get green board settings."""
+        return {
+            ATTR_ACTIVITY_LED: self.sys_dbus.agent.board.green.activity_led,
+            ATTR_POWER_LED: self.sys_dbus.agent.board.green.power_led,
+            ATTR_SYSTEM_HEALTH_LED: self.sys_dbus.agent.board.green.user_led,
+        }
+
+    @api_process
+    async def boards_green_options(self, request: web.Request) -> None:
+        """Update green board settings."""
+        body = await api_validate(SCHEMA_GREEN_OPTIONS, request)
+
+        if ATTR_ACTIVITY_LED in body:
+            self.sys_dbus.agent.board.green.activity_led = body[ATTR_ACTIVITY_LED]
+
+        if ATTR_POWER_LED in body:
+            self.sys_dbus.agent.board.green.power_led = body[ATTR_POWER_LED]
+
+        if ATTR_SYSTEM_HEALTH_LED in body:
+            self.sys_dbus.agent.board.green.user_led = body[ATTR_SYSTEM_HEALTH_LED]
+
+        self.sys_dbus.agent.board.green.save_data()
+
     @api_process
     async def boards_yellow_info(self, request: web.Request) -> dict[str, Any]:
         """Get yellow board settings."""
@@ -128,6 +163,7 @@ class APIOS(CoreSysAttributes):
         if ATTR_POWER_LED in body:
             self.sys_dbus.agent.board.yellow.power_led = body[ATTR_POWER_LED]
 
+        self.sys_dbus.agent.board.yellow.save_data()
         self.sys_resolution.create_issue(
             IssueType.REBOOT_REQUIRED,
             ContextType.SYSTEM,

supervisor/api/proxy.py

@@ -6,7 +6,10 @@ import logging
 import aiohttp
 from aiohttp import web
 from aiohttp.client_exceptions import ClientConnectorError
+from aiohttp.client_ws import ClientWebSocketResponse
 from aiohttp.hdrs import AUTHORIZATION, CONTENT_TYPE
+from aiohttp.http import WSMessage
+from aiohttp.http_websocket import WSMsgType
 from aiohttp.web_exceptions import HTTPBadGateway, HTTPUnauthorized
 
 from ..coresys import CoreSysAttributes
@@ -18,6 +21,13 @@ _LOGGER: logging.Logger = logging.getLogger(__name__)
 FORWARD_HEADERS = ("X-Speech-Content",)
 HEADER_HA_ACCESS = "X-Ha-Access"
 
+# Maximum message size for websocket messages from Home Assistant.
+# Since these are coming from core we want the largest possible size
+# that is not likely to cause a memory problem as most modern browsers
+# support large messages.
+# https://github.com/home-assistant/supervisor/issues/4392
+MAX_MESSAGE_SIZE_FROM_CORE = 64 * 1024 * 1024
+
 
 class APIProxy(CoreSysAttributes):
     """API Proxy for Home Assistant."""
@@ -67,7 +77,7 @@ class APIProxy(CoreSysAttributes):
             _LOGGER.error("Error on API for request %s", path)
         except aiohttp.ClientError as err:
             _LOGGER.error("Client error on API %s request %s", path, err)
-        except asyncio.TimeoutError:
+        except TimeoutError:
             _LOGGER.error("Client timeout error on API request %s", path)
 
         raise HTTPBadGateway()
@@ -107,12 +117,14 @@ class APIProxy(CoreSysAttributes):
                 body=data, status=client.status, content_type=client.content_type
             )
 
-    async def _websocket_client(self):
+    async def _websocket_client(self) -> ClientWebSocketResponse:
         """Initialize a WebSocket API connection."""
         url = f"{self.sys_homeassistant.api_url}/api/websocket"
 
         try:
-            client = await self.sys_websession.ws_connect(url, heartbeat=30, ssl=False)
+            client = await self.sys_websession.ws_connect(
+                url, heartbeat=30, ssl=False, max_msg_size=MAX_MESSAGE_SIZE_FROM_CORE
+            )
 
             # Handle authentication
             data = await client.receive_json()
@@ -158,6 +170,25 @@ class APIProxy(CoreSysAttributes):
 
         raise APIError()
 
+    async def _proxy_message(
+        self,
+        read_task: asyncio.Task,
+        target: web.WebSocketResponse | ClientWebSocketResponse,
+    ) -> None:
+        """Proxy a message from client to server or vice versa."""
+        if read_task.exception():
+            raise read_task.exception()
+
+        msg: WSMessage = read_task.result()
+        if msg.type == WSMsgType.TEXT:
+            return await target.send_str(msg.data)
+        if msg.type == WSMsgType.BINARY:
+            return await target.send_bytes(msg.data)
+
+        raise TypeError(
+            f"Cannot proxy websocket message of unsupported type: {msg.type}"
+        )
+
     async def websocket(self, request: web.Request):
         """Initialize a WebSocket API connection."""
         if not await self.sys_homeassistant.api.check_api_state():
@@ -205,13 +236,13 @@ class APIProxy(CoreSysAttributes):
 
         _LOGGER.info("Home Assistant WebSocket API request running")
         try:
-            client_read = None
-            server_read = None
+            client_read: asyncio.Task | None = None
+            server_read: asyncio.Task | None = None
             while not server.closed and not client.closed:
                 if not client_read:
-                    client_read = self.sys_create_task(client.receive_str())
+                    client_read = self.sys_create_task(client.receive())
                 if not server_read:
-                    server_read = self.sys_create_task(server.receive_str())
+                    server_read = self.sys_create_task(server.receive())
 
                 # wait until data need to be processed
                 await asyncio.wait(
@@ -220,14 +251,12 @@ class APIProxy(CoreSysAttributes):
 
                 # server
                 if server_read.done() and not client.closed:
-                    server_read.exception()
-                    await client.send_str(server_read.result())
+                    await self._proxy_message(server_read, client)
                     server_read = None
 
                 # client
                 if client_read.done() and not server.closed:
-                    client_read.exception()
-                    await server.send_str(client_read.result())
+                    await self._proxy_message(client_read, server)
                     client_read = None
 
         except asyncio.CancelledError:
@@ -237,9 +266,9 @@ class APIProxy(CoreSysAttributes):
             _LOGGER.info("Home Assistant WebSocket API error: %s", err)
 
         finally:
-            if client_read:
+            if client_read and not client_read.done():
                 client_read.cancel()
-            if server_read:
+            if server_read and not server_read.done():
                 server_read.cancel()
 
             # close connections

supervisor/api/store.py

@@ -6,7 +6,7 @@ from typing import Any
 from aiohttp import web
 import voluptuous as vol
 
-from ..addons import AnyAddon
+from ..addons.manager import AnyAddon
 from ..addons.utils import rating_security
 from ..api.const import ATTR_SIGNED
 from ..api.utils import api_process, api_process_raw, api_validate
@@ -186,18 +186,20 @@ class APIStore(CoreSysAttributes):
         }
 
     @api_process
-    async def addons_list(self, request: web.Request) -> list[dict[str, Any]]:
+    async def addons_list(self, request: web.Request) -> dict[str, Any]:
         """Return all store add-ons."""
-        return [
-            self._generate_addon_information(self.sys_addons.store[addon])
-            for addon in self.sys_addons.store
-        ]
+        return {
+            ATTR_ADDONS: [
+                self._generate_addon_information(self.sys_addons.store[addon])
+                for addon in self.sys_addons.store
+            ]
+        }
 
     @api_process
     def addons_addon_install(self, request: web.Request) -> Awaitable[None]:
         """Install add-on."""
         addon = self._extract_addon(request)
-        return asyncio.shield(addon.install())
+        return asyncio.shield(self.sys_addons.install(addon.slug))
 
     @api_process
     async def addons_addon_update(self, request: web.Request) -> None:
@@ -209,7 +211,7 @@ class APIStore(CoreSysAttributes):
         body = await api_validate(SCHEMA_UPDATE, request)
 
         if start_task := await asyncio.shield(
-            addon.update(backup=body.get(ATTR_BACKUP))
+            self.sys_addons.update(addon.slug, backup=body.get(ATTR_BACKUP))
         ):
             await start_task
 

supervisor/api/utils.py

@@ -22,7 +22,7 @@ from ..const import (
 from ..coresys import CoreSys
 from ..exceptions import APIError, APIForbidden, DockerAPIError, HassioError
 from ..utils import check_exception_chain, get_message_from_exception_chain
-from ..utils.json import JSONEncoder
+from ..utils.json import json_dumps, json_loads as json_loads_util
 from ..utils.log_format import format_message
 from .const import CONTENT_TYPE_BINARY
 
@@ -48,7 +48,7 @@ def json_loads(data: Any) -> dict[str, Any]:
     if not data:
         return {}
     try:
-        return json.loads(data)
+        return json_loads_util(data)
     except json.JSONDecodeError as err:
         raise APIError("Invalid json") from err
 
@@ -130,7 +130,7 @@ def api_return_error(
             JSON_MESSAGE: message or "Unknown error, see supervisor",
         },
         status=400,
-        dumps=lambda x: json.dumps(x, cls=JSONEncoder),
+        dumps=json_dumps,
     )
 
 
@@ -138,7 +138,7 @@ def api_return_ok(data: dict[str, Any] | None = None) -> web.Response:
     """Return an API ok answer."""
     return web.json_response(
         {JSON_RESULT: RESULT_OK, JSON_DATA: data or {}},
-        dumps=lambda x: json.dumps(x, cls=JSONEncoder),
+        dumps=json_dumps,
     )
 
 

supervisor/arch.py

@@ -28,6 +28,7 @@ class CpuArch(CoreSysAttributes):
         """Initialize CPU Architecture handler."""
         self.coresys = coresys
         self._supported_arch: list[str] = []
+        self._supported_set: set[str] = set()
         self._default_arch: str
 
     @property
@@ -70,9 +71,11 @@ class CpuArch(CoreSysAttributes):
         if native_support not in self._supported_arch:
             self._supported_arch.append(native_support)
 
+        self._supported_set = set(self._supported_arch)
+
     def is_supported(self, arch_list: list[str]) -> bool:
         """Return True if there is a supported arch by this platform."""
-        return not set(self.supported).isdisjoint(set(arch_list))
+        return not self._supported_set.isdisjoint(arch_list)
 
     def match(self, arch_list: list[str]) -> str:
         """Return best match for this CPU/Platform."""

supervisor/backups/backup.py

@@ -1,4 +1,5 @@
 """Representation of a backup file."""
+import asyncio
 from base64 import b64decode, b64encode
 from collections.abc import Awaitable
 from datetime import timedelta
@@ -18,13 +19,14 @@ from securetar import SecureTarFile, atomic_contents_add, secure_path
 import voluptuous as vol
 from voluptuous.humanize import humanize_error
 
-from ..addons import Addon
+from ..addons.manager import Addon
 from ..const import (
     ATTR_ADDONS,
     ATTR_COMPRESSED,
     ATTR_CRYPTO,
     ATTR_DATE,
     ATTR_DOCKER,
+    ATTR_EXCLUDE_DATABASE,
     ATTR_FOLDERS,
     ATTR_HOMEASSISTANT,
     ATTR_NAME,
@@ -129,7 +131,14 @@ class Backup(CoreSysAttributes):
         """Return backup Home Assistant version."""
         if self.homeassistant is None:
             return None
-        return self._data[ATTR_HOMEASSISTANT][ATTR_VERSION]
+        return self.homeassistant[ATTR_VERSION]
+
+    @property
+    def homeassistant_exclude_database(self) -> bool:
+        """Return whether database was excluded from Home Assistant backup."""
+        if self.homeassistant is None:
+            return None
+        return self.homeassistant[ATTR_EXCLUDE_DATABASE]
 
     @property
     def homeassistant(self):
@@ -183,7 +192,15 @@ class Backup(CoreSysAttributes):
             days=self.sys_backups.days_until_stale
         )
 
-    def new(self, slug, name, date, sys_type, password=None, compressed=True):
+    def new(
+        self,
+        slug: str,
+        name: str,
+        date: str,
+        sys_type: BackupType,
+        password: str | None = None,
+        compressed: bool = True,
+    ):
         """Initialize a new backup."""
         # Init metadata
         self._data[ATTR_VERSION] = 2
@@ -288,7 +305,7 @@ class Backup(CoreSysAttributes):
 
     async def __aenter__(self):
         """Async context to open a backup."""
-        self._tmp = TemporaryDirectory(dir=str(self.sys_config.path_tmp))
+        self._tmp = TemporaryDirectory(dir=str(self.tarfile.parent))
 
         # create a backup
         if not self.tarfile.is_file():
@@ -298,7 +315,11 @@ class Backup(CoreSysAttributes):
         def _extract_backup():
             """Extract a backup."""
             with tarfile.open(self.tarfile, "r:") as tar:
-                tar.extractall(path=self._tmp.name, members=secure_path(tar))
+                tar.extractall(
+                    path=self._tmp.name,
+                    members=secure_path(tar),
+                    filter="fully_trusted",
+                )
 
         await self.sys_run_in_executor(_extract_backup)
 
@@ -332,14 +353,14 @@ class Backup(CoreSysAttributes):
         finally:
             self._tmp.cleanup()
 
-    async def store_addons(self, addon_list: list[str]) -> list[Awaitable[None]]:
+    async def store_addons(self, addon_list: list[str]) -> list[asyncio.Task]:
         """Add a list of add-ons into backup.
 
-        For each addon that needs to be started after backup, returns a task which
+        For each addon that needs to be started after backup, returns a Task which
         completes when that addon has state 'started' (see addon.start).
         """
 
-        async def _addon_save(addon: Addon) -> Awaitable[None] | None:
+        async def _addon_save(addon: Addon) -> asyncio.Task | None:
             """Task to store an add-on into backup."""
             tar_name = f"{addon.slug}.tar{'.gz' if self.compressed else ''}"
             addon_file = SecureTarFile(
@@ -371,7 +392,7 @@ class Backup(CoreSysAttributes):
 
         # Save Add-ons sequential
         # avoid issue on slow IO
-        start_tasks: list[Awaitable[None]] = []
+        start_tasks: list[asyncio.Task] = []
         for addon in addon_list:
             try:
                 if start_task := await _addon_save(addon):
@@ -381,10 +402,12 @@ class Backup(CoreSysAttributes):
 
         return start_tasks
 
-    async def restore_addons(self, addon_list: list[str]) -> list[Awaitable[None]]:
+    async def restore_addons(
+        self, addon_list: list[str]
+    ) -> tuple[bool, list[asyncio.Task]]:
         """Restore a list add-on from backup."""
 
-        async def _addon_restore(addon_slug: str) -> Awaitable[None] | None:
+        async def _addon_restore(addon_slug: str) -> tuple[bool, asyncio.Task | None]:
             """Task to restore an add-on into backup."""
             tar_name = f"{addon_slug}.tar{'.gz' if self.compressed else ''}"
             addon_file = SecureTarFile(
@@ -398,30 +421,36 @@ class Backup(CoreSysAttributes):
             # If exists inside backup
             if not addon_file.path.exists():
                 _LOGGER.error("Can't find backup %s", addon_slug)
-                return
+                return (False, None)
 
             # Perform a restore
             try:
-                return await self.sys_addons.restore(addon_slug, addon_file)
+                return (True, await self.sys_addons.restore(addon_slug, addon_file))
             except AddonsError:
                 _LOGGER.error("Can't restore backup %s", addon_slug)
+                return (False, None)
 
         # Save Add-ons sequential
         # avoid issue on slow IO
-        start_tasks: list[Awaitable[None]] = []
+        start_tasks: list[asyncio.Task] = []
+        success = True
         for slug in addon_list:
             try:
-                if start_task := await _addon_restore(slug):
-                    start_tasks.append(start_task)
+                addon_success, start_task = await _addon_restore(slug)
             except Exception as err:  # pylint: disable=broad-except
                 _LOGGER.warning("Can't restore Add-on %s: %s", slug, err)
+                success = False
+            else:
+                success = success and addon_success
+                if start_task:
+                    start_tasks.append(start_task)
 
-        return start_tasks
+        return (success, start_tasks)
 
     async def store_folders(self, folder_list: list[str]):
         """Backup Supervisor data into backup."""
 
-        def _folder_save(name: str):
+        async def _folder_save(name: str):
             """Take backup of a folder."""
             slug_name = name.replace("/", "_")
             tar_name = Path(
@@ -434,39 +463,43 @@ class Backup(CoreSysAttributes):
                 _LOGGER.warning("Can't find backup folder %s", name)
                 return
 
-            # Take backup
-            _LOGGER.info("Backing up folder %s", name)
-            with SecureTarFile(
-                tar_name, "w", key=self._key, gzip=self.compressed, bufsize=BUF_SIZE
-            ) as tar_file:
-                atomic_contents_add(
-                    tar_file,
-                    origin_dir,
-                    excludes=[
-                        bound.bind_mount.local_where.as_posix()
-                        for bound in self.sys_mounts.bound_mounts
-                        if bound.bind_mount.local_where
-                    ],
-                    arcname=".",
-                )
+            def _save() -> None:
+                # Take backup
+                _LOGGER.info("Backing up folder %s", name)
+                with SecureTarFile(
+                    tar_name, "w", key=self._key, gzip=self.compressed, bufsize=BUF_SIZE
+                ) as tar_file:
+                    atomic_contents_add(
+                        tar_file,
+                        origin_dir,
+                        excludes=[
+                            bound.bind_mount.local_where.as_posix()
+                            for bound in self.sys_mounts.bound_mounts
+                            if bound.bind_mount.local_where
+                        ],
+                        arcname=".",
+                    )
 
-            _LOGGER.info("Backup folder %s done", name)
+                _LOGGER.info("Backup folder %s done", name)
+
+            await self.sys_run_in_executor(_save)
             self._data[ATTR_FOLDERS].append(name)
 
         # Save folder sequential
         # avoid issue on slow IO
         for folder in folder_list:
             try:
-                await self.sys_run_in_executor(_folder_save, folder)
+                await _folder_save(folder)
             except (tarfile.TarError, OSError) as err:
                 raise BackupError(
                     f"Can't backup folder {folder}: {str(err)}", _LOGGER.error
                 ) from err
 
-    async def restore_folders(self, folder_list: list[str]):
+    async def restore_folders(self, folder_list: list[str]) -> bool:
         """Backup Supervisor data into backup."""
+        success = True
 
-        async def _folder_restore(name: str) -> None:
+        async def _folder_restore(name: str) -> bool:
             """Intenal function to restore a folder."""
             slug_name = name.replace("/", "_")
             tar_name = Path(
@@ -477,14 +510,26 @@ class Backup(CoreSysAttributes):
             # Check if exists inside backup
             if not tar_name.exists():
                 _LOGGER.warning("Can't find restore folder %s", name)
-                return
+                return False
+
+            # Unmount any mounts within folder
+            bind_mounts = [
+                bound.bind_mount
+                for bound in self.sys_mounts.bound_mounts
+                if bound.bind_mount.local_where
+                and bound.bind_mount.local_where.is_relative_to(origin_dir)
+            ]
+            if bind_mounts:
+                await asyncio.gather(
+                    *[bind_mount.unmount() for bind_mount in bind_mounts]
+                )
 
             # Clean old stuff
             if origin_dir.is_dir():
                 await remove_folder(origin_dir, content_only=True)
 
             # Perform a restore
-            def _restore() -> None:
+            def _restore() -> bool:
                 try:
                     _LOGGER.info("Restore folder %s", name)
                     with SecureTarFile(
@@ -494,24 +539,39 @@ class Backup(CoreSysAttributes):
                         gzip=self.compressed,
                         bufsize=BUF_SIZE,
                     ) as tar_file:
-                        tar_file.extractall(path=origin_dir, members=tar_file)
+                        tar_file.extractall(
+                            path=origin_dir, members=tar_file, filter="fully_trusted"
+                        )
                     _LOGGER.info("Restore folder %s done", name)
                 except (tarfile.TarError, OSError) as err:
                     _LOGGER.warning("Can't restore folder %s: %s", name, err)
+                    return False
+                return True
 
-            await self.sys_run_in_executor(_restore)
+            try:
+                return await self.sys_run_in_executor(_restore)
+            finally:
+                if bind_mounts:
+                    await asyncio.gather(
+                        *[bind_mount.mount() for bind_mount in bind_mounts]
+                    )
 
         # Restore folder sequential
         # avoid issue on slow IO
         for folder in folder_list:
             try:
-                await _folder_restore(folder)
+                success = success and await _folder_restore(folder)
             except Exception as err:  # pylint: disable=broad-except
                 _LOGGER.warning("Can't restore folder %s: %s", folder, err)
+                success = False
+        return success
 
-    async def store_homeassistant(self):
-        """Backup Home Assitant Core configuration folder."""
-        self._data[ATTR_HOMEASSISTANT] = {ATTR_VERSION: self.sys_homeassistant.version}
+    async def store_homeassistant(self, exclude_database: bool = False):
+        """Backup Home Assistant Core configuration folder."""
+        self._data[ATTR_HOMEASSISTANT] = {
+            ATTR_VERSION: self.sys_homeassistant.version,
+            ATTR_EXCLUDE_DATABASE: exclude_database,
+        }
 
         # Backup Home Assistant Core config directory
         tar_name = Path(
@@ -521,13 +581,13 @@ class Backup(CoreSysAttributes):
             tar_name, "w", key=self._key, gzip=self.compressed, bufsize=BUF_SIZE
         )
 
-        await self.sys_homeassistant.backup(homeassistant_file)
+        await self.sys_homeassistant.backup(homeassistant_file, exclude_database)
 
         # Store size
         self.homeassistant[ATTR_SIZE] = homeassistant_file.size
 
     async def restore_homeassistant(self) -> Awaitable[None]:
-        """Restore Home Assitant Core configuration folder."""
+        """Restore Home Assistant Core configuration folder."""
         await self.sys_homeassistant.core.stop()
 
         # Restore Home Assistant Core config directory
@@ -538,7 +598,9 @@ class Backup(CoreSysAttributes):
             tar_name, "r", key=self._key, gzip=self.compressed, bufsize=BUF_SIZE
         )
 
-        await self.sys_homeassistant.restore(homeassistant_file)
+        await self.sys_homeassistant.restore(
+            homeassistant_file, self.homeassistant_exclude_database
+        )
 
         # Generate restore task
         async def _core_update():
@@ -561,12 +623,12 @@ class Backup(CoreSysAttributes):
         """Store repository list into backup."""
         self.repositories = self.sys_store.repository_urls
 
-    async def restore_repositories(self, replace: bool = False):
+    def restore_repositories(self, replace: bool = False) -> Awaitable[None]:
         """Restore repositories from backup.
 
         Return a coroutine.
         """
-        await self.sys_store.update_repositories(
+        return self.sys_store.update_repositories(
             self.repositories, add_with_errors=True, replace=replace
         )
 

supervisor/backups/const.py

@@ -1,11 +1,38 @@
 """Backup consts."""
-from enum import Enum
+from enum import StrEnum
 
 BUF_SIZE = 2**20 * 4  # 4MB
+DEFAULT_FREEZE_TIMEOUT = 600
 
 
-class BackupType(str, Enum):
+class BackupType(StrEnum):
     """Backup type enum."""
 
     FULL = "full"
     PARTIAL = "partial"
+
+
+class BackupJobStage(StrEnum):
+    """Backup job stage enum."""
+
+    ADDON_REPOSITORIES = "addon_repositories"
+    ADDONS = "addons"
+    DOCKER_CONFIG = "docker_config"
+    FINISHING_FILE = "finishing_file"
+    FOLDERS = "folders"
+    HOME_ASSISTANT = "home_assistant"
+    AWAIT_ADDON_RESTARTS = "await_addon_restarts"
+
+
+class RestoreJobStage(StrEnum):
+    """Restore job stage enum."""
+
+    ADDON_REPOSITORIES = "addon_repositories"
+    ADDONS = "addons"
+    AWAIT_ADDON_RESTARTS = "await_addon_restarts"
+    AWAIT_HOME_ASSISTANT_RESTART = "await_home_assistant_restart"
+    CHECK_HOME_ASSISTANT = "check_home_assistant"
+    DOCKER_CONFIG = "docker_config"
+    FOLDERS = "folders"
+    HOME_ASSISTANT = "home_assistant"
+    REMOVE_DELTA_ADDONS = "remove_delta_addons"

supervisor/backups/manager.py

@@ -3,6 +3,7 @@ from __future__ import annotations
 
 import asyncio
 from collections.abc import Awaitable, Iterable
+import errno
 import logging
 from pathlib import Path
 
@@ -13,44 +14,35 @@ from ..const import (
     FOLDER_HOMEASSISTANT,
     CoreState,
 )
-from ..coresys import CoreSysAttributes
 from ..dbus.const import UnitActiveState
-from ..exceptions import AddonsError
-from ..jobs.decorator import Job, JobCondition
+from ..exceptions import AddonsError, BackupError, BackupInvalidError, BackupJobError
+from ..jobs.const import JOB_GROUP_BACKUP_MANAGER, JobCondition, JobExecutionLimit
+from ..jobs.decorator import Job
+from ..jobs.job_group import JobGroup
 from ..mounts.mount import Mount
+from ..resolution.const import UnhealthyReason
 from ..utils.common import FileConfiguration
 from ..utils.dt import utcnow
 from ..utils.sentinel import DEFAULT
 from ..utils.sentry import capture_exception
 from .backup import Backup
-from .const import BackupType
+from .const import DEFAULT_FREEZE_TIMEOUT, BackupJobStage, BackupType, RestoreJobStage
 from .utils import create_slug
 from .validate import ALL_FOLDERS, SCHEMA_BACKUPS_CONFIG
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 
-def _list_backup_files(path: Path) -> Iterable[Path]:
-    """Return iterable of backup files, suppress and log OSError for network mounts."""
-    try:
-        # is_dir does a stat syscall which raises if the mount is down
-        if path.is_dir():
-            return path.glob("*.tar")
-    except OSError as err:
-        _LOGGER.error("Could not list backups from %s: %s", path.as_posix(), err)
-
-    return []
-
-
-class BackupManager(FileConfiguration, CoreSysAttributes):
+class BackupManager(FileConfiguration, JobGroup):
     """Manage backups."""
 
     def __init__(self, coresys):
         """Initialize a backup manager."""
         super().__init__(FILE_HASSIO_BACKUPS, SCHEMA_BACKUPS_CONFIG)
-        self.coresys = coresys
-        self._backups = {}
-        self.lock = asyncio.Lock()
+        super(FileConfiguration, self).__init__(coresys, JOB_GROUP_BACKUP_MANAGER)
+        self._backups: dict[str, Backup] = {}
+        self._thaw_task: Awaitable[None] | None = None
+        self._thaw_event: asyncio.Event = asyncio.Event()
 
     @property
     def list_backups(self) -> set[Backup]:
@@ -76,7 +68,7 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             if mount.state == UnitActiveState.ACTIVE
         ]
 
-    def get(self, slug):
+    def get(self, slug: str) -> Backup:
         """Return backup object."""
         return self._backups.get(slug)
 
@@ -90,6 +82,46 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
 
         return self.sys_config.path_backup
 
+    def _change_stage(
+        self,
+        stage: BackupJobStage | RestoreJobStage,
+        backup: Backup | None = None,
+    ):
+        """Change the stage of the current job during backup/restore.
+
+        Must be called from an existing backup/restore job.
+        """
+        job_name = self.sys_jobs.current.name
+        if "restore" in job_name:
+            action = "Restore"
+        elif "freeze" in job_name:
+            action = "Freeze"
+        elif "thaw" in job_name:
+            action = "Thaw"
+        else:
+            action = "Backup"
+
+        _LOGGER.info(
+            "%s %sstarting stage %s",
+            action,
+            f"{backup.slug} " if backup else "",
+            stage,
+        )
+        self.sys_jobs.current.stage = stage
+
+    def _list_backup_files(self, path: Path) -> Iterable[Path]:
+        """Return iterable of backup files, suppress and log OSError for network mounts."""
+        try:
+            # is_dir does a stat syscall which raises if the mount is down
+            if path.is_dir():
+                return path.glob("*.tar")
+        except OSError as err:
+            if err.errno == errno.EBADMSG and path == self.sys_config.path_backup:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
+            _LOGGER.error("Could not list backups from %s: %s", path.as_posix(), err)
+
+        return []
+
     def _create_backup(
         self,
         name: str,
@@ -98,7 +130,10 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         compressed: bool = True,
         location: Mount | type[DEFAULT] | None = DEFAULT,
     ) -> Backup:
-        """Initialize a new backup object from name."""
+        """Initialize a new backup object from name.
+
+        Must be called from an existing backup job.
+        """
         date_str = utcnow().isoformat()
         slug = create_slug(name, date_str)
         tar_file = Path(self._get_base_path(location), f"{slug}.tar")
@@ -107,19 +142,24 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         backup = Backup(self.coresys, tar_file)
         backup.new(slug, name, date_str, sys_type, password, compressed)
 
+        # Add backup ID to job
+        self.sys_jobs.current.reference = backup.slug
+
+        self._change_stage(BackupJobStage.ADDON_REPOSITORIES, backup)
         backup.store_repositories()
+        self._change_stage(BackupJobStage.DOCKER_CONFIG, backup)
         backup.store_dockerconfig()
 
         return backup
 
-    def load(self):
+    def load(self) -> Awaitable[None]:
         """Load exists backups data.
 
         Return a coroutine.
         """
         return self.reload()
 
-    async def reload(self):
+    async def reload(self) -> None:
         """Load exists backups."""
         self._backups = {}
 
@@ -132,14 +172,14 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         tasks = [
             self.sys_create_task(_load_backup(tar_file))
             for path in self.backup_locations
-            for tar_file in _list_backup_files(path)
+            for tar_file in self._list_backup_files(path)
         ]
 
         _LOGGER.info("Found %d backup files", len(tasks))
         if tasks:
             await asyncio.wait(tasks)
 
-    def remove(self, backup):
+    def remove(self, backup: Backup) -> bool:
         """Remove a backup."""
         try:
             backup.tarfile.unlink()
@@ -147,12 +187,17 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             _LOGGER.info("Removed backup file %s", backup.slug)
 
         except OSError as err:
+            if (
+                err.errno == errno.EBADMSG
+                and backup.tarfile.parent == self.sys_config.path_backup
+            ):
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             _LOGGER.error("Can't remove backup %s: %s", backup.slug, err)
             return False
 
         return True
 
-    async def import_backup(self, tar_file):
+    async def import_backup(self, tar_file: Path) -> Backup | None:
         """Check backup tarfile and import it."""
         backup = Backup(self.coresys, tar_file)
 
@@ -171,6 +216,8 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             backup.tarfile.rename(tar_origin)
 
         except OSError as err:
+            if err.errno == errno.EBADMSG:
+                self.sys_resolution.unhealthy = UnhealthyReason.OSERROR_BAD_MESSAGE
             _LOGGER.error("Can't move backup file to storage: %s", err)
             return None
 
@@ -189,26 +236,39 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         addon_list: list[Addon],
         folder_list: list[str],
         homeassistant: bool,
-    ):
+        homeassistant_exclude_database: bool | None,
+    ) -> Backup | None:
+        """Create a backup.
+
+        Must be called from an existing backup job.
+        """
         addon_start_tasks: list[Awaitable[None]] | None = None
+
         try:
             self.sys_core.state = CoreState.FREEZE
 
             async with backup:
                 # Backup add-ons
                 if addon_list:
-                    _LOGGER.info("Backing up %s store Add-ons", backup.slug)
+                    self._change_stage(BackupJobStage.ADDONS, backup)
                     addon_start_tasks = await backup.store_addons(addon_list)
 
                 # HomeAssistant Folder is for v1
                 if homeassistant:
-                    await backup.store_homeassistant()
+                    self._change_stage(BackupJobStage.HOME_ASSISTANT, backup)
+                    await backup.store_homeassistant(
+                        self.sys_homeassistant.backups_exclude_database
+                        if homeassistant_exclude_database is None
+                        else homeassistant_exclude_database
+                    )
 
                 # Backup folders
                 if folder_list:
-                    _LOGGER.info("Backing up %s store folders", backup.slug)
+                    self._change_stage(BackupJobStage.FOLDERS, backup)
                     await backup.store_folders(folder_list)
 
+                self._change_stage(BackupJobStage.FINISHING_FILE, backup)
+
         except Exception as err:  # pylint: disable=broad-except
             _LOGGER.exception("Backup %s error", backup.slug)
             capture_exception(err)
@@ -217,6 +277,7 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             self._backups[backup.slug] = backup
 
             if addon_start_tasks:
+                self._change_stage(BackupJobStage.AWAIT_ADDON_RESTARTS, backup)
                 # Ignore exceptions from waiting for addon startup, addon errors handled elsewhere
                 await asyncio.gather(*addon_start_tasks, return_exceptions=True)
 
@@ -224,33 +285,48 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         finally:
             self.sys_core.state = CoreState.RUNNING
 
-    @Job(conditions=[JobCondition.FREE_SPACE, JobCondition.RUNNING])
+    @Job(
+        name="backup_manager_full_backup",
+        conditions=[JobCondition.RUNNING],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
+    )
     async def do_backup_full(
         self,
-        name="",
-        password=None,
-        compressed=True,
+        name: str = "",
+        password: str | None = None,
+        compressed: bool = True,
         location: Mount | type[DEFAULT] | None = DEFAULT,
-    ):
+        homeassistant_exclude_database: bool | None = None,
+    ) -> Backup | None:
         """Create a full backup."""
-        if self.lock.locked():
-            _LOGGER.error("A backup/restore process is already running")
-            return None
+        if self._get_base_path(location) == self.sys_config.path_backup:
+            await Job.check_conditions(
+                self, {JobCondition.FREE_SPACE}, "BackupManager.do_backup_full"
+            )
 
         backup = self._create_backup(
             name, BackupType.FULL, password, compressed, location
         )
 
         _LOGGER.info("Creating new full backup with slug %s", backup.slug)
-        async with self.lock:
-            backup = await self._do_backup(
-                backup, self.sys_addons.installed, ALL_FOLDERS, True
-            )
-            if backup:
-                _LOGGER.info("Creating full backup with slug %s completed", backup.slug)
-            return backup
+        backup = await self._do_backup(
+            backup,
+            self.sys_addons.installed,
+            ALL_FOLDERS,
+            True,
+            homeassistant_exclude_database,
+        )
+        if backup:
+            _LOGGER.info("Creating full backup with slug %s completed", backup.slug)
+        return backup
 
-    @Job(conditions=[JobCondition.FREE_SPACE, JobCondition.RUNNING])
+    @Job(
+        name="backup_manager_partial_backup",
+        conditions=[JobCondition.RUNNING],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
+    )
     async def do_backup_partial(
         self,
         name: str = "",
@@ -260,11 +336,13 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         homeassistant: bool = False,
         compressed: bool = True,
         location: Mount | type[DEFAULT] | None = DEFAULT,
-    ):
+        homeassistant_exclude_database: bool | None = None,
+    ) -> Backup | None:
         """Create a partial backup."""
-        if self.lock.locked():
-            _LOGGER.error("A backup/restore process is already running")
-            return None
+        if self._get_base_path(location) == self.sys_config.path_backup:
+            await Job.check_conditions(
+                self, {JobCondition.FREE_SPACE}, "BackupManager.do_backup_partial"
+            )
 
         addons = addons or []
         folders = folders or []
@@ -282,21 +360,20 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         )
 
         _LOGGER.info("Creating new partial backup with slug %s", backup.slug)
-        async with self.lock:
-            addon_list = []
-            for addon_slug in addons:
-                addon = self.sys_addons.get(addon_slug)
-                if addon and addon.is_installed:
-                    addon_list.append(addon)
-                    continue
-                _LOGGER.warning("Add-on %s not found/installed", addon_slug)
+        addon_list = []
+        for addon_slug in addons:
+            addon = self.sys_addons.get(addon_slug)
+            if addon and addon.is_installed:
+                addon_list.append(addon)
+                continue
+            _LOGGER.warning("Add-on %s not found/installed", addon_slug)
 
-            backup = await self._do_backup(backup, addon_list, folders, homeassistant)
-            if backup:
-                _LOGGER.info(
-                    "Creating partial backup with slug %s completed", backup.slug
-                )
-            return backup
+        backup = await self._do_backup(
+            backup, addon_list, folders, homeassistant, homeassistant_exclude_database
+        )
+        if backup:
+            _LOGGER.info("Creating partial backup with slug %s completed", backup.slug)
+        return backup
 
     async def _do_restore(
         self,
@@ -305,28 +382,34 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         folder_list: list[str],
         homeassistant: bool,
         replace: bool,
-    ):
+    ) -> bool:
+        """Restore from a backup.
+
+        Must be called from an existing restore job.
+        """
         addon_start_tasks: list[Awaitable[None]] | None = None
+        success = True
+
         try:
             task_hass: asyncio.Task | None = None
             async with backup:
                 # Restore docker config
-                _LOGGER.info("Restoring %s Docker config", backup.slug)
+                self._change_stage(RestoreJobStage.DOCKER_CONFIG, backup)
                 backup.restore_dockerconfig(replace)
 
                 # Process folders
                 if folder_list:
-                    _LOGGER.info("Restoring %s folders", backup.slug)
-                    await backup.restore_folders(folder_list)
+                    self._change_stage(RestoreJobStage.FOLDERS, backup)
+                    success = await backup.restore_folders(folder_list)
 
                 # Process Home-Assistant
                 if homeassistant:
-                    _LOGGER.info("Restoring %s Home Assistant Core", backup.slug)
+                    self._change_stage(RestoreJobStage.HOME_ASSISTANT, backup)
                     task_hass = await backup.restore_homeassistant()
 
                 # Delete delta add-ons
                 if replace:
-                    _LOGGER.info("Removing Add-ons not in the backup %s", backup.slug)
+                    self._change_stage(RestoreJobStage.REMOVE_DELTA_ADDONS, backup)
                     for addon in self.sys_addons.installed:
                         if addon.slug in backup.addon_list:
                             continue
@@ -334,97 +417,123 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
                         # Remove Add-on because it's not a part of the new env
                         # Do it sequential avoid issue on slow IO
                         try:
-                            await addon.uninstall()
+                            await self.sys_addons.uninstall(addon.slug)
                         except AddonsError:
                             _LOGGER.warning("Can't uninstall Add-on %s", addon.slug)
+                            success = False
 
                 if addon_list:
-                    _LOGGER.info("Restoring %s Repositories", backup.slug)
+                    self._change_stage(RestoreJobStage.ADDON_REPOSITORIES, backup)
                     await backup.restore_repositories(replace)
 
-                    _LOGGER.info("Restoring %s Add-ons", backup.slug)
-                    addon_start_tasks = await backup.restore_addons(addon_list)
+                    self._change_stage(RestoreJobStage.ADDONS, backup)
+                    restore_success, addon_start_tasks = await backup.restore_addons(
+                        addon_list
+                    )
+                    success = success and restore_success
 
                 # Wait for Home Assistant Core update/downgrade
                 if task_hass:
-                    _LOGGER.info("Restore %s wait for Home-Assistant", backup.slug)
+                    self._change_stage(
+                        RestoreJobStage.AWAIT_HOME_ASSISTANT_RESTART, backup
+                    )
                     await task_hass
-
+        except BackupError:
+            raise
         except Exception as err:  # pylint: disable=broad-except
             _LOGGER.exception("Restore %s error", backup.slug)
             capture_exception(err)
-            return False
+            raise BackupError(
+                f"Restore {backup.slug} error, check logs for details"
+            ) from err
         else:
             if addon_start_tasks:
-                # Ignore exceptions from waiting for addon startup, addon errors handled elsewhere
-                await asyncio.gather(*addon_start_tasks, return_exceptions=True)
+                self._change_stage(RestoreJobStage.AWAIT_ADDON_RESTARTS, backup)
+                # Failure to resume addons post restore is still a restore failure
+                if any(
+                    await asyncio.gather(*addon_start_tasks, return_exceptions=True)
+                ):
+                    return False
 
-            return True
+            return success
         finally:
-            # Do we need start Home Assistant Core?
-            if not await self.sys_homeassistant.core.is_running():
-                await self.sys_homeassistant.core.start()
+            # Leave Home Assistant alone if it wasn't part of the restore
+            if homeassistant:
+                self._change_stage(RestoreJobStage.CHECK_HOME_ASSISTANT, backup)
 
-            # Check If we can access to API / otherwise restart
-            if not await self.sys_homeassistant.api.check_api_state():
-                _LOGGER.warning("Need restart HomeAssistant for API")
-                await self.sys_homeassistant.core.restart()
+                # Do we need start Home Assistant Core?
+                if not await self.sys_homeassistant.core.is_running():
+                    await self.sys_homeassistant.core.start()
+
+                # Check If we can access to API / otherwise restart
+                if not await self.sys_homeassistant.api.check_api_state():
+                    _LOGGER.warning("Need restart HomeAssistant for API")
+                    await self.sys_homeassistant.core.restart()
 
     @Job(
+        name="backup_manager_full_restore",
         conditions=[
             JobCondition.FREE_SPACE,
             JobCondition.HEALTHY,
             JobCondition.INTERNET_HOST,
             JobCondition.INTERNET_SYSTEM,
             JobCondition.RUNNING,
-        ]
+        ],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
     )
-    async def do_restore_full(self, backup: Backup, password=None):
+    async def do_restore_full(
+        self, backup: Backup, password: str | None = None
+    ) -> bool:
         """Restore a backup."""
-        if self.lock.locked():
-            _LOGGER.error("A backup/restore process is already running")
-            return False
+        # Add backup ID to job
+        self.sys_jobs.current.reference = backup.slug
 
         if backup.sys_type != BackupType.FULL:
-            _LOGGER.error("%s is only a partial backup!", backup.slug)
-            return False
+            raise BackupInvalidError(
+                f"{backup.slug} is only a partial backup!", _LOGGER.error
+            )
 
         if backup.protected and not backup.set_password(password):
-            _LOGGER.error("Invalid password for backup %s", backup.slug)
-            return False
+            raise BackupInvalidError(
+                f"Invalid password for backup {backup.slug}", _LOGGER.error
+            )
 
         if backup.supervisor_version > self.sys_supervisor.version:
-            _LOGGER.error(
-                "Backup was made on supervisor version %s, can't restore on %s. Must update supervisor first.",
-                backup.supervisor_version,
-                self.sys_supervisor.version,
+            raise BackupInvalidError(
+                f"Backup was made on supervisor version {backup.supervisor_version}, "
+                f"can't restore on {self.sys_supervisor.version}. Must update supervisor first.",
+                _LOGGER.error,
             )
-            return False
 
         _LOGGER.info("Full-Restore %s start", backup.slug)
-        async with self.lock:
-            self.sys_core.state = CoreState.FREEZE
+        self.sys_core.state = CoreState.FREEZE
 
+        try:
             # Stop Home-Assistant / Add-ons
             await self.sys_core.shutdown()
 
             success = await self._do_restore(
                 backup, backup.addon_list, backup.folders, True, True
             )
-
+        finally:
             self.sys_core.state = CoreState.RUNNING
 
-            if success:
-                _LOGGER.info("Full-Restore %s done", backup.slug)
+        if success:
+            _LOGGER.info("Full-Restore %s done", backup.slug)
+        return success
 
     @Job(
+        name="backup_manager_partial_restore",
         conditions=[
             JobCondition.FREE_SPACE,
             JobCondition.HEALTHY,
             JobCondition.INTERNET_HOST,
             JobCondition.INTERNET_SYSTEM,
             JobCondition.RUNNING,
-        ]
+        ],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
     )
     async def do_restore_partial(
         self,
@@ -433,11 +542,10 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
         addons: list[str] | None = None,
         folders: list[Path] | None = None,
         password: str | None = None,
-    ):
+    ) -> bool:
         """Restore a backup."""
-        if self.lock.locked():
-            _LOGGER.error("A backup/restore process is already running")
-            return False
+        # Add backup ID to job
+        self.sys_jobs.current.reference = backup.slug
 
         addon_list = addons or []
         folder_list = folders or []
@@ -448,30 +556,118 @@ class BackupManager(FileConfiguration, CoreSysAttributes):
             homeassistant = True
 
         if backup.protected and not backup.set_password(password):
-            _LOGGER.error("Invalid password for backup %s", backup.slug)
-            return False
+            raise BackupInvalidError(
+                f"Invalid password for backup {backup.slug}", _LOGGER.error
+            )
 
         if backup.homeassistant is None and homeassistant:
-            _LOGGER.error("No Home Assistant Core data inside the backup")
-            return False
+            raise BackupInvalidError(
+                "No Home Assistant Core data inside the backup", _LOGGER.error
+            )
 
         if backup.supervisor_version > self.sys_supervisor.version:
-            _LOGGER.error(
-                "Backup was made on supervisor version %s, can't restore on %s. Must update supervisor first.",
-                backup.supervisor_version,
-                self.sys_supervisor.version,
+            raise BackupInvalidError(
+                f"Backup was made on supervisor version {backup.supervisor_version}, "
+                f"can't restore on {self.sys_supervisor.version}. Must update supervisor first.",
+                _LOGGER.error,
             )
-            return False
 
         _LOGGER.info("Partial-Restore %s start", backup.slug)
-        async with self.lock:
-            self.sys_core.state = CoreState.FREEZE
+        self.sys_core.state = CoreState.FREEZE
 
+        try:
             success = await self._do_restore(
                 backup, addon_list, folder_list, homeassistant, False
             )
-
+        finally:
             self.sys_core.state = CoreState.RUNNING
 
-            if success:
-                _LOGGER.info("Partial-Restore %s done", backup.slug)
+        if success:
+            _LOGGER.info("Partial-Restore %s done", backup.slug)
+        return success
+
+    @Job(
+        name="backup_manager_freeze_all",
+        conditions=[JobCondition.RUNNING],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
+    )
+    async def freeze_all(self, timeout: float = DEFAULT_FREEZE_TIMEOUT) -> None:
+        """Freeze system to prepare for an external backup such as an image snapshot."""
+        self.sys_core.state = CoreState.FREEZE
+
+        # Determine running addons
+        installed = self.sys_addons.installed.copy()
+        is_running: list[bool] = await asyncio.gather(
+            *[addon.is_running() for addon in installed]
+        )
+        running_addons = [
+            installed[ind] for ind in range(len(installed)) if is_running[ind]
+        ]
+
+        # Create thaw task first to ensure we eventually undo freezes even if the below fails
+        self._thaw_task = asyncio.shield(
+            self.sys_create_task(self._thaw_all(running_addons, timeout))
+        )
+
+        # Tell Home Assistant to freeze for a backup
+        self._change_stage(BackupJobStage.HOME_ASSISTANT)
+        await self.sys_homeassistant.begin_backup()
+
+        # Run all pre-backup tasks for addons
+        self._change_stage(BackupJobStage.ADDONS)
+        await asyncio.gather(*[addon.begin_backup() for addon in running_addons])
+
+    @Job(
+        name="backup_manager_thaw_all",
+        conditions=[JobCondition.FROZEN],
+        on_condition=BackupJobError,
+    )
+    async def _thaw_all(
+        self, running_addons: list[Addon], timeout: float = DEFAULT_FREEZE_TIMEOUT
+    ) -> None:
+        """Thaw system after user signal or timeout."""
+        try:
+            try:
+                await asyncio.wait_for(self._thaw_event.wait(), timeout)
+            except TimeoutError:
+                _LOGGER.warning(
+                    "Timeout waiting for signal to thaw after manual freeze, beginning thaw now"
+                )
+
+            self._change_stage(BackupJobStage.HOME_ASSISTANT)
+            await self.sys_homeassistant.end_backup()
+
+            self._change_stage(BackupJobStage.ADDONS)
+            addon_start_tasks: list[asyncio.Task] = [
+                task
+                for task in await asyncio.gather(
+                    *[addon.end_backup() for addon in running_addons]
+                )
+                if task
+            ]
+        finally:
+            self.sys_core.state = CoreState.RUNNING
+            self._thaw_event.clear()
+            self._thaw_task = None
+
+        if addon_start_tasks:
+            self._change_stage(BackupJobStage.AWAIT_ADDON_RESTARTS)
+            await asyncio.gather(*addon_start_tasks, return_exceptions=True)
+
+    @Job(
+        name="backup_manager_signal_thaw",
+        conditions=[JobCondition.FROZEN],
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=BackupJobError,
+        internal=True,
+    )
+    async def thaw_all(self) -> None:
+        """Signal thaw task to begin unfreezing the system."""
+        if not self._thaw_task:
+            raise BackupError(
+                "Freeze was not initiated by freeze API, cannot thaw this way"
+            )
+
+        self._thaw_event.set()
+        await self._thaw_task

supervisor/backups/validate.py

@@ -14,6 +14,7 @@ from ..const import (
     ATTR_DATE,
     ATTR_DAYS_UNTIL_STALE,
     ATTR_DOCKER,
+    ATTR_EXCLUDE_DATABASE,
     ATTR_FOLDERS,
     ATTR_HOMEASSISTANT,
     ATTR_NAME,
@@ -103,6 +104,9 @@ SCHEMA_BACKUP = vol.Schema(
                     {
                         vol.Required(ATTR_VERSION): version_tag,
                         vol.Optional(ATTR_SIZE, default=0): vol.Coerce(float),
+                        vol.Optional(
+                            ATTR_EXCLUDE_DATABASE, default=False
+                        ): vol.Boolean(),
                     },
                     extra=vol.REMOVE_EXTRA,
                 )

supervisor/bootstrap.py

@@ -6,7 +6,7 @@ import signal
 
 from colorlog import ColoredFormatter
 
-from .addons import AddonManager
+from .addons.manager import AddonManager
 from .api import RestAPI
 from .arch import CpuArch
 from .auth import Auth
@@ -221,6 +221,14 @@ def initialize_system(coresys: CoreSys) -> None:
         )
         config.path_emergency.mkdir()
 
+    # Addon Configs folder
+    if not config.path_addon_configs.is_dir():
+        _LOGGER.debug(
+            "Creating Supervisor add-on configs folder at '%s'",
+            config.path_addon_configs,
+        )
+        config.path_addon_configs.mkdir()
+
 
 def migrate_system_env(coresys: CoreSys) -> None:
     """Cleanup some stuff after update."""

supervisor/config.py

@@ -1,5 +1,5 @@
 """Bootstrap Supervisor."""
-from datetime import datetime
+from datetime import UTC, datetime
 import logging
 import os
 from pathlib import Path, PurePath
@@ -48,8 +48,9 @@ MEDIA_DATA = PurePath("media")
 MOUNTS_FOLDER = PurePath("mounts")
 MOUNTS_CREDENTIALS = PurePath(".mounts_credentials")
 EMERGENCY_DATA = PurePath("emergency")
+ADDON_CONFIGS = PurePath("addon_configs")
 
-DEFAULT_BOOT_TIME = datetime.utcfromtimestamp(0).isoformat()
+DEFAULT_BOOT_TIME = datetime.fromtimestamp(0, UTC).isoformat()
 
 # We filter out UTC because it's the system default fallback
 # Core also not respect the cotnainer timezone and reset timezones
@@ -153,7 +154,7 @@ class CoreConfig(FileConfiguration):
 
     def modify_log_level(self) -> None:
         """Change log level."""
-        lvl = getattr(logging, str(self.logging.value).upper())
+        lvl = getattr(logging, self.logging.value.upper())
         logging.getLogger("supervisor").setLevel(lvl)
 
     @property
@@ -163,7 +164,7 @@ class CoreConfig(FileConfiguration):
 
         boot_time = parse_datetime(boot_str)
         if not boot_time:
-            return datetime.utcfromtimestamp(1)
+            return datetime.fromtimestamp(1, UTC)
         return boot_time
 
     @last_boot.setter
@@ -231,6 +232,16 @@ class CoreConfig(FileConfiguration):
         """Return root add-on data folder external for Docker."""
         return PurePath(self.path_extern_supervisor, ADDONS_DATA)
 
+    @property
+    def path_addon_configs(self) -> Path:
+        """Return root Add-on configs folder."""
+        return self.path_supervisor / ADDON_CONFIGS
+
+    @property
+    def path_extern_addon_configs(self) -> PurePath:
+        """Return root Add-on configs folder external for Docker."""
+        return PurePath(self.path_extern_supervisor, ADDON_CONFIGS)
+
     @property
     def path_audio(self) -> Path:
         """Return root audio data folder."""

supervisor/const.py

@@ -1,8 +1,10 @@
 """Constants file for Supervisor."""
-from enum import Enum
+from dataclasses import dataclass
+from enum import StrEnum
 from ipaddress import ip_network
 from pathlib import Path
 from sys import version_info as systemversion
+from typing import Self
 
 from aiohttp import __version__ as aiohttpversion
 
@@ -18,6 +20,7 @@ SUPERVISOR_DATA = Path("/data")
 FILE_HASSIO_ADDONS = Path(SUPERVISOR_DATA, "addons.json")
 FILE_HASSIO_AUTH = Path(SUPERVISOR_DATA, "auth.json")
 FILE_HASSIO_BACKUPS = Path(SUPERVISOR_DATA, "backups.json")
+FILE_HASSIO_BOARD = Path(SUPERVISOR_DATA, "board.json")
 FILE_HASSIO_CONFIG = Path(SUPERVISOR_DATA, "config.json")
 FILE_HASSIO_DISCOVERY = Path(SUPERVISOR_DATA, "discovery.json")
 FILE_HASSIO_DOCKER = Path(SUPERVISOR_DATA, "docker.json")
@@ -69,6 +72,9 @@ JSON_RESULT = "result"
 RESULT_ERROR = "error"
 RESULT_OK = "ok"
 
+HEADER_REMOTE_USER_ID = "X-Remote-User-Id"
+HEADER_REMOTE_USER_NAME = "X-Remote-User-Name"
+HEADER_REMOTE_USER_DISPLAY_NAME = "X-Remote-User-Display-Name"
 HEADER_TOKEN_OLD = "X-Hassio-Key"
 HEADER_TOKEN = "X-Supervisor-Token"
 
@@ -84,6 +90,7 @@ REQUEST_FROM = "HASSIO_FROM"
 ATTR_ACCESS_TOKEN = "access_token"
 ATTR_ACCESSPOINTS = "accesspoints"
 ATTR_ACTIVE = "active"
+ATTR_ACTIVITY_LED = "activity_led"
 ATTR_ADDON = "addon"
 ATTR_ADDONS = "addons"
 ATTR_ADDONS_CUSTOM_LIST = "addons_custom_list"
@@ -109,6 +116,7 @@ ATTR_BACKUP_EXCLUDE = "backup_exclude"
 ATTR_BACKUP_POST = "backup_post"
 ATTR_BACKUP_PRE = "backup_pre"
 ATTR_BACKUPS = "backups"
+ATTR_BACKUPS_EXCLUDE_DATABASE = "backups_exclude_database"
 ATTR_BLK_READ = "blk_read"
 ATTR_BLK_WRITE = "blk_write"
 ATTR_BOARD = "board"
@@ -148,9 +156,11 @@ ATTR_DIAGNOSTICS = "diagnostics"
 ATTR_DISCOVERY = "discovery"
 ATTR_DISK = "disk"
 ATTR_DISK_FREE = "disk_free"
+ATTR_DISK_LED = "disk_led"
 ATTR_DISK_LIFE_TIME = "disk_life_time"
 ATTR_DISK_TOTAL = "disk_total"
 ATTR_DISK_USED = "disk_used"
+ATTR_DISPLAYNAME = "displayname"
 ATTR_DNS = "dns"
 ATTR_DOCKER = "docker"
 ATTR_DOCKER_API = "docker_api"
@@ -160,6 +170,7 @@ ATTR_ENABLE = "enable"
 ATTR_ENABLED = "enabled"
 ATTR_ENVIRONMENT = "environment"
 ATTR_EVENT = "event"
+ATTR_EXCLUDE_DATABASE = "exclude_database"
 ATTR_FEATURES = "features"
 ATTR_FILENAME = "filename"
 ATTR_FLAGS = "flags"
@@ -173,7 +184,9 @@ ATTR_HASSIO_API = "hassio_api"
 ATTR_HASSIO_ROLE = "hassio_role"
 ATTR_HASSOS = "hassos"
 ATTR_HEALTHY = "healthy"
+ATTR_HEARTBEAT_LED = "heartbeat_led"
 ATTR_HOMEASSISTANT = "homeassistant"
+ATTR_HOMEASSISTANT_EXCLUDE_DATABASE = "homeassistant_exclude_database"
 ATTR_HOMEASSISTANT_API = "homeassistant_api"
 ATTR_HOST = "host"
 ATTR_HOST_DBUS = "host_dbus"
@@ -248,6 +261,7 @@ ATTR_PLUGINS = "plugins"
 ATTR_PORT = "port"
 ATTR_PORTS = "ports"
 ATTR_PORTS_DESCRIPTION = "ports_description"
+ATTR_POWER_LED = "power_led"
 ATTR_PREFIX = "prefix"
 ATTR_PRIMARY = "primary"
 ATTR_PRIORITY = "priority"
@@ -271,6 +285,9 @@ ATTR_SERVERS = "servers"
 ATTR_SERVICE = "service"
 ATTR_SERVICES = "services"
 ATTR_SESSION = "session"
+ATTR_SESSION_DATA = "session_data"
+ATTR_SESSION_DATA_USER = "user"
+ATTR_SESSION_DATA_USER_ID = "user_id"
 ATTR_SIGNAL = "signal"
 ATTR_SIZE = "size"
 ATTR_SLUG = "slug"
@@ -308,6 +325,7 @@ ATTR_UPDATE_KEY = "update_key"
 ATTR_URL = "url"
 ATTR_USB = "usb"
 ATTR_USER = "user"
+ATTR_USER_LED = "user_led"
 ATTR_USERNAME = "username"
 ATTR_UUID = "uuid"
 ATTR_VALID = "valid"
@@ -327,14 +345,6 @@ PROVIDE_SERVICE = "provide"
 NEED_SERVICE = "need"
 WANT_SERVICE = "want"
 
-
-MAP_CONFIG = "config"
-MAP_SSL = "ssl"
-MAP_ADDONS = "addons"
-MAP_BACKUP = "backup"
-MAP_SHARE = "share"
-MAP_MEDIA = "media"
-
 ARCH_ARMHF = "armhf"
 ARCH_ARMV7 = "armv7"
 ARCH_AARCH64 = "aarch64"
@@ -367,14 +377,14 @@ ROLE_ADMIN = "admin"
 ROLE_ALL = [ROLE_DEFAULT, ROLE_HOMEASSISTANT, ROLE_BACKUP, ROLE_MANAGER, ROLE_ADMIN]
 
 
-class AddonBoot(str, Enum):
+class AddonBoot(StrEnum):
     """Boot mode for the add-on."""
 
     AUTO = "auto"
     MANUAL = "manual"
 
 
-class AddonStartup(str, Enum):
+class AddonStartup(StrEnum):
     """Startup types of Add-on."""
 
     INITIALIZE = "initialize"
@@ -384,7 +394,7 @@ class AddonStartup(str, Enum):
     ONCE = "once"
 
 
-class AddonStage(str, Enum):
+class AddonStage(StrEnum):
     """Stage types of add-on."""
 
     STABLE = "stable"
@@ -392,7 +402,7 @@ class AddonStage(str, Enum):
     DEPRECATED = "deprecated"
 
 
-class AddonState(str, Enum):
+class AddonState(StrEnum):
     """State of add-on."""
 
     STARTUP = "startup"
@@ -402,7 +412,7 @@ class AddonState(str, Enum):
     ERROR = "error"
 
 
-class UpdateChannel(str, Enum):
+class UpdateChannel(StrEnum):
     """Core supported update channels."""
 
     STABLE = "stable"
@@ -410,7 +420,7 @@ class UpdateChannel(str, Enum):
     DEV = "dev"
 
 
-class CoreState(str, Enum):
+class CoreState(StrEnum):
     """Represent current loading state."""
 
     INITIALIZE = "initialize"
@@ -423,7 +433,7 @@ class CoreState(str, Enum):
     CLOSE = "close"
 
 
-class LogLevel(str, Enum):
+class LogLevel(StrEnum):
     """Logging level of system."""
 
     DEBUG = "debug"
@@ -433,7 +443,7 @@ class LogLevel(str, Enum):
     CRITICAL = "critical"
 
 
-class HostFeature(str, Enum):
+class HostFeature(StrEnum):
     """Host feature."""
 
     HASSOS = "hassos"
@@ -445,15 +455,16 @@ class HostFeature(str, Enum):
     TIMEDATE = "timedate"
 
 
-class BusEvent(str, Enum):
+class BusEvent(StrEnum):
     """Bus event type."""
 
     HARDWARE_NEW_DEVICE = "hardware_new_device"
     HARDWARE_REMOVE_DEVICE = "hardware_remove_device"
     DOCKER_CONTAINER_STATE_CHANGE = "docker_container_state_change"
+    SUPERVISOR_STATE_CHANGE = "supervisor_state_change"
 
 
-class CpuArch(str, Enum):
+class CpuArch(StrEnum):
     """Supported CPU architectures."""
 
     ARMV7 = "armv7"
@@ -461,3 +472,52 @@ class CpuArch(str, Enum):
     AARCH64 = "aarch64"
     I386 = "i386"
     AMD64 = "amd64"
+
+
+@dataclass
+class IngressSessionDataUser:
+    """Format of an IngressSessionDataUser object."""
+
+    id: str
+    display_name: str | None = None
+    username: str | None = None
+
+    def to_dict(self) -> dict[str, str | None]:
+        """Get dictionary representation."""
+        return {
+            ATTR_ID: self.id,
+            ATTR_DISPLAYNAME: self.display_name,
+            ATTR_USERNAME: self.username,
+        }
+
+    @classmethod
+    def from_dict(cls, data: dict[str, str | None]) -> Self:
+        """Return object from dictionary representation."""
+        return cls(
+            id=data[ATTR_ID],
+            display_name=data.get(ATTR_DISPLAYNAME),
+            username=data.get(ATTR_USERNAME),
+        )
+
+
+@dataclass
+class IngressSessionData:
+    """Format of an IngressSessionData object."""
+
+    user: IngressSessionDataUser
+
+    def to_dict(self) -> dict[str, dict[str, str | None]]:
+        """Get dictionary representation."""
+        return {ATTR_USER: self.user.to_dict()}
+
+    @classmethod
+    def from_dict(cls, data: dict[str, dict[str, str | None]]) -> Self:
+        """Return object from dictionary representation."""
+        return cls(user=IngressSessionDataUser.from_dict(data[ATTR_USER]))
+
+
+STARTING_STATES = [
+    CoreState.INITIALIZE,
+    CoreState.STARTUP,
+    CoreState.SETUP,
+]

supervisor/core.py

@@ -7,7 +7,14 @@ import logging
 
 import async_timeout
 
-from .const import RUN_SUPERVISOR_STATE, AddonStartup, CoreState
+from .const import (
+    ATTR_STARTUP,
+    RUN_SUPERVISOR_STATE,
+    STARTING_STATES,
+    AddonStartup,
+    BusEvent,
+    CoreState,
+)
 from .coresys import CoreSys, CoreSysAttributes
 from .exceptions import (
     HassioError,
@@ -21,7 +28,7 @@ from .homeassistant.core import LANDINGPAGE
 from .resolution.const import ContextType, IssueType, SuggestionType, UnhealthyReason
 from .utils.dt import utcnow
 from .utils.sentry import capture_exception
-from .utils.whoami import retrieve_whoami
+from .utils.whoami import WhoamiData, retrieve_whoami
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
@@ -56,16 +63,23 @@ class Core(CoreSysAttributes):
         if self._state == new_state:
             return
         try:
-            RUN_SUPERVISOR_STATE.write_text(new_state.value, encoding="utf-8")
+            RUN_SUPERVISOR_STATE.write_text(new_state, encoding="utf-8")
         except OSError as err:
             _LOGGER.warning(
                 "Can't update the Supervisor state to %s: %s", new_state, err
             )
         finally:
             self._state = new_state
-            self.sys_homeassistant.websocket.supervisor_update_event(
-                "info", {"state": new_state}
-            )
+
+            # Don't attempt to notify anyone on CLOSE as we're about to stop the event loop
+            if new_state != CoreState.CLOSE:
+                self.sys_bus.fire_event(BusEvent.SUPERVISOR_STATE_CHANGE, new_state)
+
+                # These will be received by HA after startup has completed which won't make sense
+                if new_state not in STARTING_STATES:
+                    self.sys_homeassistant.websocket.supervisor_update_event(
+                        "info", {"state": new_state}
+                    )
 
     async def connect(self):
         """Connect Supervisor container."""
@@ -119,12 +133,12 @@ class Core(CoreSysAttributes):
             self._adjust_system_datetime(),
             # Load mounts
             self.sys_mounts.load(),
-            # Start docker monitoring
+            # Load Docker manager
             self.sys_docker.load(),
-            # Load Plugins container
-            self.sys_plugins.load(),
             # load last available data
             self.sys_updater.load(),
+            # Load Plugins container
+            self.sys_plugins.load(),
             # Load Home Assistant
             self.sys_homeassistant.load(),
             # Load CPU/Arch
@@ -236,7 +250,7 @@ class Core(CoreSysAttributes):
                 except HomeAssistantError as err:
                     capture_exception(err)
             else:
-                _LOGGER.info("Skiping start of Home Assistant")
+                _LOGGER.info("Skipping start of Home Assistant")
 
             # Core is not running
             if self.sys_homeassistant.core.error_state:
@@ -266,7 +280,9 @@ class Core(CoreSysAttributes):
             self.sys_create_task(self.sys_resolution.healthcheck())
 
             self.state = CoreState.RUNNING
-            self.sys_homeassistant.websocket.supervisor_update_event("supervisor", {})
+            self.sys_homeassistant.websocket.supervisor_update_event(
+                "supervisor", {ATTR_STARTUP: "complete"}
+            )
             _LOGGER.info("Supervisor is up and running")
 
     async def stop(self):
@@ -293,7 +309,7 @@ class Core(CoreSysAttributes):
                         )
                     ]
                 )
-        except asyncio.TimeoutError:
+        except TimeoutError:
             _LOGGER.warning("Stage 1: Force Shutdown!")
 
         # Stage 2
@@ -310,7 +326,7 @@ class Core(CoreSysAttributes):
                         )
                     ]
                 )
-        except asyncio.TimeoutError:
+        except TimeoutError:
             _LOGGER.warning("Stage 2: Force Shutdown!")
 
         self.state = CoreState.CLOSE
@@ -347,6 +363,13 @@ class Core(CoreSysAttributes):
         self.sys_config.last_boot = self.sys_hardware.helper.last_boot
         self.sys_config.save_data()
 
+    async def _retrieve_whoami(self, with_ssl: bool) -> WhoamiData | None:
+        try:
+            return await retrieve_whoami(self.sys_websession, with_ssl)
+        except WhoamiSSLError:
+            _LOGGER.info("Whoami service SSL error")
+            return None
+
     async def _adjust_system_datetime(self):
         """Adjust system time/date on startup."""
         # If no timezone is detect or set
@@ -359,21 +382,15 @@ class Core(CoreSysAttributes):
 
         # Get Timezone data
         try:
-            data = await retrieve_whoami(self.sys_websession)
-        except WhoamiSSLError:
-            pass
+            data = await self._retrieve_whoami(True)
+
+            # SSL Date Issue & possible time drift
+            if not data:
+                data = await self._retrieve_whoami(False)
         except WhoamiError as err:
             _LOGGER.warning("Can't adjust Time/Date settings: %s", err)
             return
 
-        # SSL Date Issue & possible time drift
-        if not data:
-            try:
-                data = await retrieve_whoami(self.sys_websession, with_ssl=False)
-            except WhoamiError as err:
-                _LOGGER.error("Can't adjust Time/Date settings: %s", err)
-                return
-
         self.sys_config.timezone = self.sys_config.timezone or data.timezone
 
         # Calculate if system time is out of sync

supervisor/coresys.py

@@ -3,7 +3,9 @@ from __future__ import annotations
 
 import asyncio
 from collections.abc import Callable, Coroutine
+from contextvars import Context, copy_context
 from datetime import datetime
+from functools import partial
 import logging
 import os
 from types import MappingProxyType
@@ -16,7 +18,7 @@ from .const import ENV_SUPERVISOR_DEV, SERVER_SOFTWARE
 from .utils.dt import UTC, get_time_zone
 
 if TYPE_CHECKING:
-    from .addons import AddonManager
+    from .addons.manager import AddonManager
     from .api import RestAPI
     from .arch import CpuArch
     from .auth import Auth
@@ -98,6 +100,9 @@ class CoreSys:
             {aiohttp.hdrs.USER_AGENT: SERVER_SOFTWARE}
         )
 
+        # Task factory attributes
+        self._set_task_context: list[Callable[[Context], Context]] = []
+
     @property
     def dev(self) -> bool:
         """Return True if we run dev mode."""
@@ -519,15 +524,33 @@ class CoreSys:
         """Return now in local timezone."""
         return datetime.now(get_time_zone(self.timezone) or UTC)
 
+    def add_set_task_context_callback(
+        self, callback: Callable[[Context], Context]
+    ) -> None:
+        """Add callback used to modify context prior to creating a task.
+
+        Only used for tasks created via CoreSys.create_task. Callback can modify the provided
+        context using context.run (ex. `context.run(var.set, "new_value")`). Callback should
+        return the context to be provided to task.
+        """
+        self._set_task_context.append(callback)
+
     def run_in_executor(
-        self, funct: Callable[..., T], *args: Any
+        self, funct: Callable[..., T], *args: tuple[Any], **kwargs: dict[str, Any]
     ) -> Coroutine[Any, Any, T]:
         """Add an job to the executor pool."""
+        if kwargs:
+            funct = partial(funct, **kwargs)
+
         return self.loop.run_in_executor(None, funct, *args)
 
     def create_task(self, coroutine: Coroutine) -> asyncio.Task:
         """Create an async task."""
-        return self.loop.create_task(coroutine)
+        context = copy_context()
+        for callback in self._set_task_context:
+            context = callback(context)
+
+        return self.loop.create_task(coroutine, context=context)
 
 
 class CoreSysAttributes:
@@ -700,10 +723,10 @@ class CoreSysAttributes:
         return self.coresys.now()
 
     def sys_run_in_executor(
-        self, funct: Callable[..., T], *args: Any
+        self, funct: Callable[..., T], *args: tuple[Any], **kwargs: dict[str, Any]
     ) -> Coroutine[Any, Any, T]:
-        """Add an job to the executor pool."""
-        return self.coresys.run_in_executor(funct, *args)
+        """Add a job to the executor pool."""
+        return self.coresys.run_in_executor(funct, *args, **kwargs)
 
     def sys_create_task(self, coroutine: Coroutine) -> asyncio.Task:
         """Create an async task."""

supervisor/data/arch.json

@@ -5,7 +5,9 @@
   "raspberrypi3-64": ["aarch64", "armv7", "armhf"],
   "raspberrypi4": ["armv7", "armhf"],
   "raspberrypi4-64": ["aarch64", "armv7", "armhf"],
+  "raspberrypi5-64": ["aarch64", "armv7", "armhf"],
   "yellow": ["aarch64", "armv7", "armhf"],
+  "green": ["aarch64", "armv7", "armhf"],
   "tinker": ["armv7", "armhf"],
   "odroid-c2": ["aarch64", "armv7", "armhf"],
   "odroid-c4": ["aarch64", "armv7", "armhf"],

supervisor/dbus/agent/__init__.py

@@ -6,7 +6,7 @@ from typing import Any
 from awesomeversion import AwesomeVersion
 from dbus_fast.aio.message_bus import MessageBus
 
-from ...exceptions import DBusError, DBusInterfaceError
+from ...exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..const import (
     DBUS_ATTR_DIAGNOSTICS,
     DBUS_ATTR_VERSION,
@@ -99,7 +99,7 @@ class OSAgent(DBusInterfaceProxy):
             await asyncio.gather(*[dbus.connect(bus) for dbus in self.all])
         except DBusError:
             _LOGGER.warning("Can't connect to OS-Agent")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No OS-Agent support on the host. Some Host functions have been disabled."
             )

supervisor/dbus/agent/boards/__init__.py

@@ -11,7 +11,8 @@ from ...const import (
     DBUS_OBJECT_HAOS_BOARDS,
 )
 from ...interface import DBusInterfaceProxy, dbus_property
-from .const import BOARD_NAME_SUPERVISED, BOARD_NAME_YELLOW
+from .const import BOARD_NAME_GREEN, BOARD_NAME_SUPERVISED, BOARD_NAME_YELLOW
+from .green import Green
 from .interface import BoardProxy
 from .supervised import Supervised
 from .yellow import Yellow
@@ -39,6 +40,14 @@ class BoardManager(DBusInterfaceProxy):
         """Get board name."""
         return self.properties[DBUS_ATTR_BOARD]
 
+    @property
+    def green(self) -> Green:
+        """Get Green board."""
+        if self.board != BOARD_NAME_GREEN:
+            raise BoardInvalidError("Green board is not in use", _LOGGER.error)
+
+        return self._board_proxy
+
     @property
     def supervised(self) -> Supervised:
         """Get Supervised board."""
@@ -61,6 +70,8 @@ class BoardManager(DBusInterfaceProxy):
 
         if self.board == BOARD_NAME_YELLOW:
             self._board_proxy = Yellow()
+        elif self.board == BOARD_NAME_GREEN:
+            self._board_proxy = Green()
         elif self.board == BOARD_NAME_SUPERVISED:
             self._board_proxy = Supervised()
 

supervisor/dbus/agent/boards/const.py

@@ -1,4 +1,5 @@
 """Constants for boards."""
 
+BOARD_NAME_GREEN = "Green"
 BOARD_NAME_SUPERVISED = "Supervised"
 BOARD_NAME_YELLOW = "Yellow"

supervisor/dbus/agent/boards/green.py

@@ -0,0 +1,65 @@
+"""Green board management."""
+
+import asyncio
+
+from dbus_fast.aio.message_bus import MessageBus
+
+from ....const import ATTR_ACTIVITY_LED, ATTR_POWER_LED, ATTR_USER_LED
+from ...const import DBUS_ATTR_ACTIVITY_LED, DBUS_ATTR_POWER_LED, DBUS_ATTR_USER_LED
+from ...interface import dbus_property
+from .const import BOARD_NAME_GREEN
+from .interface import BoardProxy
+from .validate import SCHEMA_GREEN_BOARD
+
+
+class Green(BoardProxy):
+    """Green board manager object."""
+
+    def __init__(self) -> None:
+        """Initialize properties."""
+        super().__init__(BOARD_NAME_GREEN, SCHEMA_GREEN_BOARD)
+
+    @property
+    @dbus_property
+    def activity_led(self) -> bool:
+        """Get activity LED enabled."""
+        return self.properties[DBUS_ATTR_ACTIVITY_LED]
+
+    @activity_led.setter
+    def activity_led(self, enabled: bool) -> None:
+        """Enable/disable activity LED."""
+        self._data[ATTR_ACTIVITY_LED] = enabled
+        asyncio.create_task(self.dbus.Boards.Green.set_activity_led(enabled))
+
+    @property
+    @dbus_property
+    def power_led(self) -> bool:
+        """Get power LED enabled."""
+        return self.properties[DBUS_ATTR_POWER_LED]
+
+    @power_led.setter
+    def power_led(self, enabled: bool) -> None:
+        """Enable/disable power LED."""
+        self._data[ATTR_POWER_LED] = enabled
+        asyncio.create_task(self.dbus.Boards.Green.set_power_led(enabled))
+
+    @property
+    @dbus_property
+    def user_led(self) -> bool:
+        """Get user LED enabled."""
+        return self.properties[DBUS_ATTR_USER_LED]
+
+    @user_led.setter
+    def user_led(self, enabled: bool) -> None:
+        """Enable/disable disk LED."""
+        self._data[ATTR_USER_LED] = enabled
+        asyncio.create_task(self.dbus.Boards.Green.set_user_led(enabled))
+
+    async def connect(self, bus: MessageBus) -> None:
+        """Connect to D-Bus."""
+        await super().connect(bus)
+
+        # Set LEDs based on settings on connect
+        self.activity_led = self._data[ATTR_ACTIVITY_LED]
+        self.power_led = self._data[ATTR_POWER_LED]
+        self.user_led = self._data[ATTR_USER_LED]

supervisor/dbus/agent/boards/interface.py

@@ -1,17 +1,23 @@
 """Board dbus proxy interface."""
 
+from voluptuous import Schema
+
+from ....const import FILE_HASSIO_BOARD
+from ....utils.common import FileConfiguration
 from ...const import DBUS_IFACE_HAOS_BOARDS, DBUS_NAME_HAOS, DBUS_OBJECT_HAOS_BOARDS
 from ...interface import DBusInterfaceProxy
+from .validate import SCHEMA_BASE_BOARD
 
 
-class BoardProxy(DBusInterfaceProxy):
+class BoardProxy(FileConfiguration, DBusInterfaceProxy):
     """DBus interface proxy for os board."""
 
     bus_name: str = DBUS_NAME_HAOS
 
-    def __init__(self, name: str) -> None:
+    def __init__(self, name: str, file_schema: Schema | None = None) -> None:
         """Initialize properties."""
-        super().__init__()
+        super().__init__(FILE_HASSIO_BOARD, file_schema or SCHEMA_BASE_BOARD)
+        super(FileConfiguration, self).__init__()
 
         self._name: str = name
         self.object_path: str = f"{DBUS_OBJECT_HAOS_BOARDS}/{name}"

supervisor/dbus/agent/boards/validate.py

@@ -0,0 +1,32 @@
+"""Validation for board config."""
+
+import voluptuous as vol
+
+from ....const import (
+    ATTR_ACTIVITY_LED,
+    ATTR_DISK_LED,
+    ATTR_HEARTBEAT_LED,
+    ATTR_POWER_LED,
+    ATTR_USER_LED,
+)
+
+# pylint: disable=no-value-for-parameter
+SCHEMA_BASE_BOARD = vol.Schema({}, extra=vol.REMOVE_EXTRA)
+
+SCHEMA_GREEN_BOARD = vol.Schema(
+    {
+        vol.Optional(ATTR_ACTIVITY_LED, default=True): vol.Boolean(),
+        vol.Optional(ATTR_POWER_LED, default=True): vol.Boolean(),
+        vol.Optional(ATTR_USER_LED, default=True): vol.Boolean(),
+    },
+    extra=vol.REMOVE_EXTRA,
+)
+
+SCHEMA_YELLOW_BOARD = vol.Schema(
+    {
+        vol.Optional(ATTR_DISK_LED, default=True): vol.Boolean(),
+        vol.Optional(ATTR_HEARTBEAT_LED, default=True): vol.Boolean(),
+        vol.Optional(ATTR_POWER_LED, default=True): vol.Boolean(),
+    },
+    extra=vol.REMOVE_EXTRA,
+)

supervisor/dbus/agent/boards/yellow.py

@@ -2,10 +2,14 @@
 
 import asyncio
 
+from dbus_fast.aio.message_bus import MessageBus
+
+from ....const import ATTR_DISK_LED, ATTR_HEARTBEAT_LED, ATTR_POWER_LED
 from ...const import DBUS_ATTR_DISK_LED, DBUS_ATTR_HEARTBEAT_LED, DBUS_ATTR_POWER_LED
 from ...interface import dbus_property
 from .const import BOARD_NAME_YELLOW
 from .interface import BoardProxy
+from .validate import SCHEMA_YELLOW_BOARD
 
 
 class Yellow(BoardProxy):
@@ -13,7 +17,7 @@ class Yellow(BoardProxy):
 
     def __init__(self) -> None:
         """Initialize properties."""
-        super().__init__(BOARD_NAME_YELLOW)
+        super().__init__(BOARD_NAME_YELLOW, SCHEMA_YELLOW_BOARD)
 
     @property
     @dbus_property
@@ -24,6 +28,7 @@ class Yellow(BoardProxy):
     @heartbeat_led.setter
     def heartbeat_led(self, enabled: bool) -> None:
         """Enable/disable heartbeat LED."""
+        self._data[ATTR_HEARTBEAT_LED] = enabled
         asyncio.create_task(self.dbus.Boards.Yellow.set_heartbeat_led(enabled))
 
     @property
@@ -35,6 +40,7 @@ class Yellow(BoardProxy):
     @power_led.setter
     def power_led(self, enabled: bool) -> None:
         """Enable/disable power LED."""
+        self._data[ATTR_POWER_LED] = enabled
         asyncio.create_task(self.dbus.Boards.Yellow.set_power_led(enabled))
 
     @property
@@ -46,4 +52,14 @@ class Yellow(BoardProxy):
     @disk_led.setter
     def disk_led(self, enabled: bool) -> None:
         """Enable/disable disk LED."""
+        self._data[ATTR_DISK_LED] = enabled
         asyncio.create_task(self.dbus.Boards.Yellow.set_disk_led(enabled))
+
+    async def connect(self, bus: MessageBus) -> None:
+        """Connect to D-Bus."""
+        await super().connect(bus)
+
+        # Set LEDs based on settings on connect
+        self.disk_led = self._data[ATTR_DISK_LED]
+        self.heartbeat_led = self._data[ATTR_HEARTBEAT_LED]
+        self.power_led = self._data[ATTR_POWER_LED]

supervisor/dbus/const.py

@@ -1,5 +1,5 @@
 """Constants for DBUS."""
-from enum import Enum, IntEnum
+from enum import IntEnum, StrEnum
 from socket import AF_INET, AF_INET6
 
 DBUS_NAME_HAOS = "io.hass.os"
@@ -64,6 +64,7 @@ DBUS_OBJECT_UDISKS2 = "/org/freedesktop/UDisks2/Manager"
 DBUS_ATTR_ACTIVE_ACCESSPOINT = "ActiveAccessPoint"
 DBUS_ATTR_ACTIVE_CONNECTION = "ActiveConnection"
 DBUS_ATTR_ACTIVE_CONNECTIONS = "ActiveConnections"
+DBUS_ATTR_ACTIVITY_LED = "ActivityLED"
 DBUS_ATTR_ADDRESS_DATA = "AddressData"
 DBUS_ATTR_BITRATE = "Bitrate"
 DBUS_ATTR_BOARD = "Board"
@@ -144,6 +145,7 @@ DBUS_ATTR_OPERATION = "Operation"
 DBUS_ATTR_OPTIONS = "Options"
 DBUS_ATTR_PARSER_VERSION = "ParserVersion"
 DBUS_ATTR_PARTITIONS = "Partitions"
+DBUS_ATTR_PATH = "Path"
 DBUS_ATTR_POWER_LED = "PowerLED"
 DBUS_ATTR_PRIMARY_CONNECTION = "PrimaryConnection"
 DBUS_ATTR_READ_ONLY = "ReadOnly"
@@ -168,6 +170,7 @@ DBUS_ATTR_TIMEUSEC = "TimeUSec"
 DBUS_ATTR_TIMEZONE = "Timezone"
 DBUS_ATTR_TRANSACTION_STATISTICS = "TransactionStatistics"
 DBUS_ATTR_TYPE = "Type"
+DBUS_ATTR_USER_LED = "UserLED"
 DBUS_ATTR_USERSPACE_TIMESTAMP_MONOTONIC = "UserspaceTimestampMonotonic"
 DBUS_ATTR_UUID_UPPERCASE = "UUID"
 DBUS_ATTR_UUID = "Uuid"
@@ -180,7 +183,7 @@ DBUS_ATTR_WWN = "WWN"
 DBUS_ERR_SYSTEMD_NO_SUCH_UNIT = "org.freedesktop.systemd1.NoSuchUnit"
 
 
-class RaucState(str, Enum):
+class RaucState(StrEnum):
     """Rauc slot states."""
 
     GOOD = "good"
@@ -188,7 +191,7 @@ class RaucState(str, Enum):
     ACTIVE = "active"
 
 
-class InterfaceMethod(str, Enum):
+class InterfaceMethod(StrEnum):
     """Interface method simple."""
 
     AUTO = "auto"
@@ -197,14 +200,14 @@ class InterfaceMethod(str, Enum):
     LINK_LOCAL = "link-local"
 
 
-class ConnectionType(str, Enum):
+class ConnectionType(StrEnum):
     """Connection type."""
 
     ETHERNET = "802-3-ethernet"
     WIRELESS = "802-11-wireless"
 
 
-class ConnectionStateType(int, Enum):
+class ConnectionStateType(IntEnum):
     """Connection states.
 
     https://developer.gnome.org/NetworkManager/stable/nm-dbus-types.html#NMActiveConnectionState
@@ -217,7 +220,7 @@ class ConnectionStateType(int, Enum):
     DEACTIVATED = 4
 
 
-class ConnectionStateFlags(int, Enum):
+class ConnectionStateFlags(IntEnum):
     """Connection state flags.
 
     https://developer-old.gnome.org/NetworkManager/stable/nm-dbus-types.html#NMActivationStateFlags
@@ -234,7 +237,7 @@ class ConnectionStateFlags(int, Enum):
     EXTERNAL = 0x80
 
 
-class ConnectivityState(int, Enum):
+class ConnectivityState(IntEnum):
     """Network connectvity.
 
     https://developer.gnome.org/NetworkManager/unstable/nm-dbus-types.html#NMConnectivityState
@@ -247,7 +250,7 @@ class ConnectivityState(int, Enum):
     CONNECTIVITY_FULL = 4
 
 
-class DeviceType(int, Enum):
+class DeviceType(IntEnum):
     """Device types.
 
     https://developer.gnome.org/NetworkManager/stable/nm-dbus-types.html#NMDeviceType
@@ -262,7 +265,7 @@ class DeviceType(int, Enum):
     VETH = 20
 
 
-class WirelessMethodType(int, Enum):
+class WirelessMethodType(IntEnum):
     """Device Type."""
 
     UNKNOWN = 0
@@ -279,7 +282,7 @@ class DNSAddressFamily(IntEnum):
     INET6 = AF_INET6
 
 
-class MulticastProtocolEnabled(str, Enum):
+class MulticastProtocolEnabled(StrEnum):
     """Multicast protocol enabled or resolve."""
 
     YES = "yes"
@@ -287,7 +290,7 @@ class MulticastProtocolEnabled(str, Enum):
     RESOLVE = "resolve"
 
 
-class DNSOverTLSEnabled(str, Enum):
+class DNSOverTLSEnabled(StrEnum):
     """DNS over TLS enabled."""
 
     YES = "yes"
@@ -295,7 +298,7 @@ class DNSOverTLSEnabled(str, Enum):
     OPPORTUNISTIC = "opportunistic"
 
 
-class DNSSECValidation(str, Enum):
+class DNSSECValidation(StrEnum):
     """DNSSEC validation enforced."""
 
     YES = "yes"
@@ -303,7 +306,7 @@ class DNSSECValidation(str, Enum):
     ALLOW_DOWNGRADE = "allow-downgrade"
 
 
-class DNSStubListenerEnabled(str, Enum):
+class DNSStubListenerEnabled(StrEnum):
     """DNS stub listener enabled."""
 
     YES = "yes"
@@ -312,7 +315,7 @@ class DNSStubListenerEnabled(str, Enum):
     UDP_ONLY = "udp"
 
 
-class ResolvConfMode(str, Enum):
+class ResolvConfMode(StrEnum):
     """Resolv.conf management mode."""
 
     FOREIGN = "foreign"
@@ -322,7 +325,7 @@ class ResolvConfMode(str, Enum):
     UPLINK = "uplink"
 
 
-class StopUnitMode(str, Enum):
+class StopUnitMode(StrEnum):
     """Mode for stopping the unit."""
 
     REPLACE = "replace"
@@ -331,7 +334,7 @@ class StopUnitMode(str, Enum):
     IGNORE_REQUIREMENTS = "ignore-requirements"
 
 
-class StartUnitMode(str, Enum):
+class StartUnitMode(StrEnum):
     """Mode for starting the unit."""
 
     REPLACE = "replace"
@@ -341,7 +344,7 @@ class StartUnitMode(str, Enum):
     ISOLATE = "isolate"
 
 
-class UnitActiveState(str, Enum):
+class UnitActiveState(StrEnum):
     """Active state of a systemd unit."""
 
     ACTIVE = "active"

supervisor/dbus/hostname.py

@@ -3,7 +3,7 @@ import logging
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from .const import (
     DBUS_ATTR_CHASSIS,
     DBUS_ATTR_DEPLOYMENT,
@@ -39,7 +39,7 @@ class Hostname(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd-hostname")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No hostname support on the host. Hostname functions have been disabled."
             )

supervisor/dbus/logind.py

@@ -3,7 +3,7 @@ import logging
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from .const import DBUS_NAME_LOGIND, DBUS_OBJECT_LOGIND
 from .interface import DBusInterface
 from .utils import dbus_connected
@@ -28,8 +28,8 @@ class Logind(DBusInterface):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd-logind")
-        except DBusInterfaceError:
-            _LOGGER.info("No systemd-logind support on the host.")
+        except (DBusServiceUnkownError, DBusInterfaceError):
+            _LOGGER.warning("No systemd-logind support on the host.")
 
     @dbus_connected
     async def reboot(self) -> None:

supervisor/dbus/network/__init__.py

@@ -9,7 +9,10 @@ from ...exceptions import (
     DBusError,
     DBusFatalError,
     DBusInterfaceError,
+    DBusNoReplyError,
+    DBusServiceUnkownError,
     HostNotSupportedError,
+    NetworkInterfaceNotFound,
 )
 from ...utils.sentry import capture_exception
 from ..const import (
@@ -67,9 +70,9 @@ class NetworkManager(DBusInterfaceProxy):
         return self._settings
 
     @property
-    def interfaces(self) -> dict[str, NetworkInterface]:
-        """Return a dictionary of active interfaces."""
-        return self._interfaces
+    def interfaces(self) -> set[NetworkInterface]:
+        """Return a set of active interfaces."""
+        return set(self._interfaces.values())
 
     @property
     @dbus_property
@@ -83,6 +86,20 @@ class NetworkManager(DBusInterfaceProxy):
         """Return Network Manager version."""
         return AwesomeVersion(self.properties[DBUS_ATTR_VERSION])
 
+    def get(self, name_or_mac: str) -> NetworkInterface:
+        """Get an interface by name or mac address."""
+        if name_or_mac not in self._interfaces:
+            raise NetworkInterfaceNotFound(
+                f"No interface exists with name or mac address '{name_or_mac}'"
+            )
+        return self._interfaces[name_or_mac]
+
+    def __contains__(self, item: NetworkInterface | str) -> bool:
+        """Return true if specified network interface exists."""
+        if isinstance(item, str):
+            return item in self._interfaces
+        return item in self.interfaces
+
     @dbus_connected
     async def activate_connection(
         self, connection_object: str, device_object: str
@@ -128,7 +145,7 @@ class NetworkManager(DBusInterfaceProxy):
             await self.settings.connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to Network Manager")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No Network Manager support on the host. Local network functions have been disabled."
             )
@@ -167,9 +184,9 @@ class NetworkManager(DBusInterfaceProxy):
 
         if changed and (
             DBUS_ATTR_DEVICES not in changed
-            or {
-                intr.object_path for intr in self.interfaces.values() if intr.managed
-            }.issubset(set(changed[DBUS_ATTR_DEVICES]))
+            or {intr.object_path for intr in self.interfaces if intr.managed}.issubset(
+                set(changed[DBUS_ATTR_DEVICES])
+            )
         ):
             # If none of our managed devices were removed then most likely this is just veths changing.
             # We don't care about veths and reprocessing all their changes can swamp a system when
@@ -177,8 +194,8 @@ class NetworkManager(DBusInterfaceProxy):
             # in rare occaisions but we'll catch it on the next host update scheduled task.
             return
 
-        interfaces = {}
-        curr_devices = {intr.object_path: intr for intr in self.interfaces.values()}
+        interfaces: dict[str, NetworkInterface] = {}
+        curr_devices = {intr.object_path: intr for intr in self.interfaces}
         for device in self.properties[DBUS_ATTR_DEVICES]:
             if device in curr_devices and curr_devices[device].is_connected:
                 interface = curr_devices[device]
@@ -195,8 +212,22 @@ class NetworkManager(DBusInterfaceProxy):
                     # try to query it. Ignore those cases.
                     _LOGGER.debug("Can't process %s: %s", device, err)
                     continue
+                except (
+                    DBusNoReplyError,
+                    DBusServiceUnkownError,
+                ) as err:
+                    # This typically means that NetworkManager disappeared. Give up immeaditly.
+                    _LOGGER.error(
+                        "NetworkManager not responding while processing %s: %s. Giving up.",
+                        device,
+                        err,
+                    )
+                    capture_exception(err)
+                    return
                 except Exception as err:  # pylint: disable=broad-except
-                    _LOGGER.exception("Error while processing %s: %s", device, err)
+                    _LOGGER.exception(
+                        "Unkown error while processing %s: %s", device, err
+                    )
                     capture_exception(err)
                     continue
 
@@ -222,6 +253,7 @@ class NetworkManager(DBusInterfaceProxy):
                 interface.primary = False
 
             interfaces[interface.name] = interface
+            interfaces[interface.hw_address] = interface
 
         # Disconnect removed devices
         for device in set(curr_devices.keys()) - set(
@@ -242,7 +274,7 @@ class NetworkManager(DBusInterfaceProxy):
 
     def disconnect(self) -> None:
         """Disconnect from D-Bus."""
-        for intr in self.interfaces.values():
+        for intr in self.interfaces:
             intr.shutdown()
 
         super().disconnect()

supervisor/dbus/network/configuration.py

@@ -1,66 +1,72 @@
-"""NetworkConnection object4s for Network Manager."""
+"""NetworkConnection objects for Network Manager."""
+from dataclasses import dataclass
 from ipaddress import IPv4Address, IPv6Address
 
-import attr
 
-
-@attr.s(slots=True)
+@dataclass(slots=True)
 class DNSConfiguration:
     """DNS configuration Object."""
 
-    nameservers: list[IPv4Address | IPv6Address] = attr.ib()
-    domains: list[str] = attr.ib()
-    interface: str = attr.ib()
-    priority: int = attr.ib()
-    vpn: bool = attr.ib()
+    nameservers: list[IPv4Address | IPv6Address]
+    domains: list[str]
+    interface: str
+    priority: int
+    vpn: bool
 
 
-@attr.s(slots=True)
+@dataclass(slots=True)
 class ConnectionProperties:
     """Connection Properties object for Network Manager."""
 
-    id: str | None = attr.ib()
-    uuid: str | None = attr.ib()
-    type: str | None = attr.ib()
-    interface_name: str | None = attr.ib()
+    id: str | None
+    uuid: str | None
+    type: str | None
+    interface_name: str | None
 
 
-@attr.s(slots=True)
+@dataclass(slots=True)
 class WirelessProperties:
     """Wireless Properties object for Network Manager."""
 
-    ssid: str | None = attr.ib()
-    assigned_mac: str | None = attr.ib()
-    mode: str | None = attr.ib()
-    powersave: int | None = attr.ib()
+    ssid: str | None
+    assigned_mac: str | None
+    mode: str | None
+    powersave: int | None
 
 
-@attr.s(slots=True)
+@dataclass(slots=True)
 class WirelessSecurityProperties:
     """Wireless Security Properties object for Network Manager."""
 
-    auth_alg: str | None = attr.ib()
-    key_mgmt: str | None = attr.ib()
-    psk: str | None = attr.ib()
+    auth_alg: str | None
+    key_mgmt: str | None
+    psk: str | None
 
 
-@attr.s(slots=True)
+@dataclass(slots=True)
 class EthernetProperties:
     """Ethernet properties object for Network Manager."""
 
-    assigned_mac: str | None = attr.ib()
+    assigned_mac: str | None
 
 
-@attr.s(slots=True)
+@dataclass(slots=True)
 class VlanProperties:
     """Ethernet properties object for Network Manager."""
 
-    id: int | None = attr.ib()
-    parent: str | None = attr.ib()
+    id: int | None
+    parent: str | None
 
 
-@attr.s(slots=True)
+@dataclass(slots=True)
 class IpProperties:
     """IP properties object for Network Manager."""
 
-    method: str | None = attr.ib()
+    method: str | None
+
+
+@dataclass(slots=True)
+class MatchProperties:
+    """Match properties object for Network Manager."""
+
+    path: list[str] | None = None

supervisor/dbus/network/connection.py

@@ -121,7 +121,7 @@ class NetworkConnection(DBusInterfaceProxy):
         self._state_flags = {
             flag
             for flag in ConnectionStateFlags
-            if flag.value & self.properties[DBUS_ATTR_STATE_FLAGS]
+            if flag & self.properties[DBUS_ATTR_STATE_FLAGS]
         } or {ConnectionStateFlags.NONE}
 
         # IPv4

supervisor/dbus/network/dns.py

@@ -12,7 +12,7 @@ from ...const import (
     ATTR_PRIORITY,
     ATTR_VPN,
 )
-from ...exceptions import DBusError, DBusInterfaceError
+from ...exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..const import (
     DBUS_ATTR_CONFIGURATION,
     DBUS_ATTR_MODE,
@@ -67,7 +67,7 @@ class NetworkManagerDNS(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to DnsManager")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No DnsManager support on the host. Local DNS functions have been disabled."
             )

supervisor/dbus/network/interface.py

@@ -9,7 +9,9 @@ from ..const import (
     DBUS_ATTR_DEVICE_INTERFACE,
     DBUS_ATTR_DEVICE_TYPE,
     DBUS_ATTR_DRIVER,
+    DBUS_ATTR_HWADDRESS,
     DBUS_ATTR_MANAGED,
+    DBUS_ATTR_PATH,
     DBUS_IFACE_DEVICE,
     DBUS_NAME_NM,
     DBUS_OBJECT_BASE,
@@ -67,6 +69,18 @@ class NetworkInterface(DBusInterfaceProxy):
         """Return interface driver."""
         return self.properties[DBUS_ATTR_MANAGED]
 
+    @property
+    @dbus_property
+    def hw_address(self) -> str:
+        """Return hardware address (i.e. mac address) of device."""
+        return self.properties[DBUS_ATTR_HWADDRESS]
+
+    @property
+    @dbus_property
+    def path(self) -> str:
+        """Return The path of the device as exposed by the udev property ID_PATH."""
+        return self.properties[DBUS_ATTR_PATH]
+
     @property
     def connection(self) -> NetworkConnection | None:
         """Return the connection used for this interface."""
@@ -98,6 +112,18 @@ class NetworkInterface(DBusInterfaceProxy):
 
         self._wireless = wireless
 
+    def __eq__(self, other: object) -> bool:
+        """Is object equal to another."""
+        return (
+            isinstance(other, type(self))
+            and other.bus_name == self.bus_name
+            and other.object_path == self.object_path
+        )
+
+    def __hash__(self) -> int:
+        """Hash of object."""
+        return hash((self.bus_name, self.object_path))
+
     async def connect(self, bus: MessageBus) -> None:
         """Connect to D-Bus."""
         await super().connect(bus)

supervisor/dbus/network/setting/__init__.py

@@ -2,6 +2,7 @@
 import logging
 from typing import Any
 
+from dbus_fast import Variant
 from dbus_fast.aio.message_bus import MessageBus
 
 from ....const import ATTR_METHOD, ATTR_MODE, ATTR_PSK, ATTR_SSID
@@ -12,6 +13,7 @@ from ..configuration import (
     ConnectionProperties,
     EthernetProperties,
     IpProperties,
+    MatchProperties,
     VlanProperties,
     WirelessProperties,
     WirelessSecurityProperties,
@@ -24,6 +26,8 @@ CONF_ATTR_802_WIRELESS_SECURITY = "802-11-wireless-security"
 CONF_ATTR_VLAN = "vlan"
 CONF_ATTR_IPV4 = "ipv4"
 CONF_ATTR_IPV6 = "ipv6"
+CONF_ATTR_MATCH = "match"
+CONF_ATTR_PATH = "path"
 
 ATTR_ID = "id"
 ATTR_UUID = "uuid"
@@ -34,6 +38,7 @@ ATTR_POWERSAVE = "powersave"
 ATTR_AUTH_ALG = "auth-alg"
 ATTR_KEY_MGMT = "key-mgmt"
 ATTR_INTERFACE_NAME = "interface-name"
+ATTR_PATH = "path"
 
 IPV4_6_IGNORE_FIELDS = [
     "addresses",
@@ -47,8 +52,8 @@ _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 
 def _merge_settings_attribute(
-    base_settings: Any,
-    new_settings: Any,
+    base_settings: dict[str, dict[str, Variant]],
+    new_settings: dict[str, dict[str, Variant]],
     attribute: str,
     *,
     ignore_current_value: list[str] = None,
@@ -58,8 +63,7 @@ def _merge_settings_attribute(
         if attribute in base_settings:
             if ignore_current_value:
                 for field in ignore_current_value:
-                    if field in base_settings[attribute]:
-                        del base_settings[attribute][field]
+                    base_settings[attribute].pop(field, None)
 
             base_settings[attribute].update(new_settings[attribute])
         else:
@@ -85,6 +89,7 @@ class NetworkSetting(DBusInterface):
         self._vlan: VlanProperties | None = None
         self._ipv4: IpProperties | None = None
         self._ipv6: IpProperties | None = None
+        self._match: MatchProperties | None = None
 
     @property
     def connection(self) -> ConnectionProperties | None:
@@ -121,19 +126,29 @@ class NetworkSetting(DBusInterface):
         """Return ipv6 properties if any."""
         return self._ipv6
 
+    @property
+    def match(self) -> MatchProperties | None:
+        """Return match properties if any."""
+        return self._match
+
     @dbus_connected
     async def get_settings(self) -> dict[str, Any]:
         """Return connection settings."""
         return await self.dbus.Settings.Connection.call_get_settings()
 
     @dbus_connected
-    async def update(self, settings: Any) -> None:
+    async def update(self, settings: dict[str, dict[str, Variant]]) -> None:
         """Update connection settings."""
-        new_settings = await self.dbus.Settings.Connection.call_get_settings(
-            unpack_variants=False
-        )
+        new_settings: dict[
+            str, dict[str, Variant]
+        ] = await self.dbus.Settings.Connection.call_get_settings(unpack_variants=False)
 
-        _merge_settings_attribute(new_settings, settings, CONF_ATTR_CONNECTION)
+        _merge_settings_attribute(
+            new_settings,
+            settings,
+            CONF_ATTR_CONNECTION,
+            ignore_current_value=[ATTR_INTERFACE_NAME],
+        )
         _merge_settings_attribute(new_settings, settings, CONF_ATTR_802_ETHERNET)
         _merge_settings_attribute(new_settings, settings, CONF_ATTR_802_WIRELESS)
         _merge_settings_attribute(
@@ -152,6 +167,7 @@ class NetworkSetting(DBusInterface):
             CONF_ATTR_IPV6,
             ignore_current_value=IPV4_6_IGNORE_FIELDS,
         )
+        _merge_settings_attribute(new_settings, settings, CONF_ATTR_MATCH)
 
         await self.dbus.Settings.Connection.call_update(new_settings)
 
@@ -217,3 +233,6 @@ class NetworkSetting(DBusInterface):
             self._ipv6 = IpProperties(
                 data[CONF_ATTR_IPV6].get(ATTR_METHOD),
             )
+
+        if CONF_ATTR_MATCH in data:
+            self._match = MatchProperties(data[CONF_ATTR_MATCH].get(ATTR_PATH))

supervisor/dbus/network/setting/generate.py

@@ -2,7 +2,7 @@
 from __future__ import annotations
 
 import socket
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 from uuid import uuid4
 
 from dbus_fast import Variant
@@ -15,17 +15,23 @@ from . import (
     CONF_ATTR_CONNECTION,
     CONF_ATTR_IPV4,
     CONF_ATTR_IPV6,
+    CONF_ATTR_MATCH,
+    CONF_ATTR_PATH,
     CONF_ATTR_VLAN,
 )
+from .. import NetworkManager
 from ....host.const import InterfaceMethod, InterfaceType
 
 if TYPE_CHECKING:
-    from ....host.network import Interface
+    from ....host.configuration import Interface
 
 
 def get_connection_from_interface(
-    interface: Interface, name: str | None = None, uuid: str | None = None
-) -> Any:
+    interface: Interface,
+    network_manager: NetworkManager,
+    name: str | None = None,
+    uuid: str | None = None,
+) -> dict[str, dict[str, Variant]]:
     """Generate message argument for network interface update."""
 
     # Generate/Update ID/name
@@ -39,26 +45,28 @@ def get_connection_from_interface(
     elif interface.type == InterfaceType.WIRELESS:
         iftype = "802-11-wireless"
     else:
-        iftype = interface.type.value
+        iftype = interface.type
 
     # Generate UUID
     if not uuid:
         uuid = str(uuid4())
 
-    connection = {
-        "id": Variant("s", name),
-        "type": Variant("s", iftype),
-        "uuid": Variant("s", uuid),
-        "llmnr": Variant("i", 2),
-        "mdns": Variant("i", 2),
-        "autoconnect": Variant("b", True),
+    conn: dict[str, dict[str, Variant]] = {
+        CONF_ATTR_CONNECTION: {
+            "id": Variant("s", name),
+            "type": Variant("s", iftype),
+            "uuid": Variant("s", uuid),
+            "llmnr": Variant("i", 2),
+            "mdns": Variant("i", 2),
+            "autoconnect": Variant("b", True),
+        },
     }
 
     if interface.type != InterfaceType.VLAN:
-        connection["interface-name"] = Variant("s", interface.name)
-
-    conn = {}
-    conn[CONF_ATTR_CONNECTION] = connection
+        if interface.path:
+            conn[CONF_ATTR_MATCH] = {CONF_ATTR_PATH: Variant("as", [interface.path])}
+        else:
+            conn[CONF_ATTR_CONNECTION]["interface-name"] = Variant("s", interface.name)
 
     ipv4 = {}
     if not interface.ipv4 or interface.ipv4.method == InterfaceMethod.AUTO:
@@ -117,9 +125,15 @@ def get_connection_from_interface(
     if interface.type == InterfaceType.ETHERNET:
         conn[CONF_ATTR_802_ETHERNET] = {ATTR_ASSIGNED_MAC: Variant("s", "preserve")}
     elif interface.type == "vlan":
+        parent = interface.vlan.interface
+        if parent in network_manager and (
+            parent_connection := network_manager.get(parent).connection
+        ):
+            parent = parent_connection.uuid
+
         conn[CONF_ATTR_VLAN] = {
             "id": Variant("u", interface.vlan.id),
-            "parent": Variant("s", interface.vlan.interface),
+            "parent": Variant("s", parent),
         }
     elif interface.type == InterfaceType.WIRELESS:
         wireless = {
@@ -134,8 +148,8 @@ def get_connection_from_interface(
             wireless["security"] = Variant("s", CONF_ATTR_802_WIRELESS_SECURITY)
             wireless_security = {}
             if interface.wifi.auth == "wep":
-                wireless_security["auth-alg"] = Variant("s", "none")
-                wireless_security["key-mgmt"] = Variant("s", "open")
+                wireless_security["auth-alg"] = Variant("s", "open")
+                wireless_security["key-mgmt"] = Variant("s", "none")
             elif interface.wifi.auth == "wpa-psk":
                 wireless_security["auth-alg"] = Variant("s", "open")
                 wireless_security["key-mgmt"] = Variant("s", "wpa-psk")

supervisor/dbus/network/settings.py

@@ -4,7 +4,7 @@ from typing import Any
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ...exceptions import DBusError, DBusInterfaceError
+from ...exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..const import DBUS_NAME_NM, DBUS_OBJECT_SETTINGS
 from ..interface import DBusInterface
 from ..network.setting import NetworkSetting
@@ -28,7 +28,7 @@ class NetworkManagerSettings(DBusInterface):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to Network Manager Settings")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No Network Manager Settings support on the host. Local network functions have been disabled."
             )

supervisor/dbus/rauc.py

@@ -4,7 +4,7 @@ from typing import Any
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..utils.dbus import DBusSignalWrapper
 from .const import (
     DBUS_ATTR_BOOT_SLOT,
@@ -49,7 +49,7 @@ class Rauc(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to rauc")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning("Host has no rauc support. OTA updates have been disabled.")
 
     @property
@@ -95,7 +95,7 @@ class Rauc(DBusInterfaceProxy):
     @dbus_connected
     async def mark(self, state: RaucState, slot_identifier: str) -> tuple[str, str]:
         """Get slot status."""
-        return await self.dbus.Installer.call_mark(state.value, slot_identifier)
+        return await self.dbus.Installer.call_mark(state, slot_identifier)
 
     @dbus_connected
     async def update(self, changed: dict[str, Any] | None = None) -> None:

supervisor/dbus/resolved.py

@@ -5,7 +5,7 @@ import logging
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from .const import (
     DBUS_ATTR_CACHE_STATISTICS,
     DBUS_ATTR_CURRENT_DNS_SERVER,
@@ -59,7 +59,7 @@ class Resolved(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd-resolved.")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "Host has no systemd-resolved support. DNS will not work correctly."
             )

supervisor/dbus/systemd.py

@@ -10,6 +10,7 @@ from ..exceptions import (
     DBusError,
     DBusFatalError,
     DBusInterfaceError,
+    DBusServiceUnkownError,
     DBusSystemdNoSuchUnit,
 )
 from .const import (
@@ -86,7 +87,7 @@ class Systemd(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No systemd support on the host. Host control has been disabled."
             )
@@ -122,25 +123,25 @@ class Systemd(DBusInterfaceProxy):
     @systemd_errors
     async def start_unit(self, unit: str, mode: StartUnitMode) -> str:
         """Start a systemd service unit. Returns object path of job."""
-        return await self.dbus.Manager.call_start_unit(unit, mode.value)
+        return await self.dbus.Manager.call_start_unit(unit, mode)
 
     @dbus_connected
     @systemd_errors
     async def stop_unit(self, unit: str, mode: StopUnitMode) -> str:
         """Stop a systemd service unit. Returns object path of job."""
-        return await self.dbus.Manager.call_stop_unit(unit, mode.value)
+        return await self.dbus.Manager.call_stop_unit(unit, mode)
 
     @dbus_connected
     @systemd_errors
     async def reload_unit(self, unit: str, mode: StartUnitMode) -> str:
         """Reload a systemd service unit. Returns object path of job."""
-        return await self.dbus.Manager.call_reload_or_restart_unit(unit, mode.value)
+        return await self.dbus.Manager.call_reload_or_restart_unit(unit, mode)
 
     @dbus_connected
     @systemd_errors
     async def restart_unit(self, unit: str, mode: StartUnitMode) -> str:
         """Restart a systemd service unit. Returns object path of job."""
-        return await self.dbus.Manager.call_restart_unit(unit, mode.value)
+        return await self.dbus.Manager.call_restart_unit(unit, mode)
 
     @dbus_connected
     async def list_units(
@@ -155,7 +156,7 @@ class Systemd(DBusInterfaceProxy):
     ) -> str:
         """Start a transient unit which is released when stopped or on reboot. Returns object path of job."""
         return await self.dbus.Manager.call_start_transient_unit(
-            unit, mode.value, properties, []
+            unit, mode, properties, []
         )
 
     @dbus_connected

supervisor/dbus/timedate.py

@@ -4,7 +4,7 @@ import logging
 
 from dbus_fast.aio.message_bus import MessageBus
 
-from ..exceptions import DBusError, DBusInterfaceError
+from ..exceptions import DBusError, DBusInterfaceError, DBusServiceUnkownError
 from ..utils.dt import utc_from_timestamp
 from .const import (
     DBUS_ATTR_NTP,
@@ -63,7 +63,7 @@ class TimeDate(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to systemd-timedate")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No timedate support on the host. Time/Date functions have been disabled."
             )

supervisor/dbus/udisks2/__init__.py

@@ -6,7 +6,12 @@ from typing import Any
 from awesomeversion import AwesomeVersion
 from dbus_fast.aio import MessageBus
 
-from ...exceptions import DBusError, DBusInterfaceError, DBusObjectError
+from ...exceptions import (
+    DBusError,
+    DBusInterfaceError,
+    DBusObjectError,
+    DBusServiceUnkownError,
+)
 from ..const import (
     DBUS_ATTR_SUPPORTED_FILESYSTEMS,
     DBUS_ATTR_VERSION,
@@ -45,7 +50,7 @@ class UDisks2(DBusInterfaceProxy):
             await super().connect(bus)
         except DBusError:
             _LOGGER.warning("Can't connect to udisks2")
-        except DBusInterfaceError:
+        except (DBusServiceUnkownError, DBusInterfaceError):
             _LOGGER.warning(
                 "No udisks2 support on the host. Host control has been disabled."
             )

supervisor/dbus/udisks2/block.py

@@ -263,6 +263,4 @@ class UDisks2Block(DBusInterfaceProxy):
     ) -> None:
         """Format block device."""
         options = options.to_dict() if options else {}
-        await self.dbus.Block.call_format(
-            type_.value, options | UDISKS2_DEFAULT_OPTIONS
-        )
+        await self.dbus.Block.call_format(type_, options | UDISKS2_DEFAULT_OPTIONS)

supervisor/dbus/udisks2/const.py

@@ -1,20 +1,20 @@
 """Constants for UDisks2."""
 
-from enum import Enum
+from enum import StrEnum
 
 from dbus_fast import Variant
 
 UDISKS2_DEFAULT_OPTIONS = {"auth.no_user_interaction": Variant("b", True)}
 
 
-class EncryptType(str, Enum):
+class EncryptType(StrEnum):
     """Encryption type."""
 
     LUKS1 = "luks1"
     LUKS2 = "luks2"
 
 
-class EraseMode(str, Enum):
+class EraseMode(StrEnum):
     """Erase mode."""
 
     ZERO = "zero"
@@ -22,7 +22,7 @@ class EraseMode(str, Enum):
     ATA_SECURE_ERASE_ENHANCED = "ata-secure-erase-enhanced"
 
 
-class FormatType(str, Enum):
+class FormatType(StrEnum):
     """Format type."""
 
     EMPTY = "empty"
@@ -31,7 +31,7 @@ class FormatType(str, Enum):
     GPT = "gpt"
 
 
-class PartitionTableType(str, Enum):
+class PartitionTableType(StrEnum):
     """Partition Table type."""
 
     DOS = "dos"

supervisor/dbus/udisks2/data.py

@@ -3,10 +3,9 @@
 from dataclasses import dataclass
 from inspect import get_annotations
 from pathlib import Path
-from typing import Any, TypedDict
+from typing import Any, NotRequired, TypedDict
 
 from dbus_fast import Variant
-from typing_extensions import NotRequired
 
 from .const import EncryptType, EraseMode
 
@@ -167,10 +166,10 @@ class FormatOptions(UDisks2StandardOptions):
             )
             if self.encrypt_passpharase
             else None,
-            "encrypt.type": Variant("s", self.encrypt_type.value)
+            "encrypt.type": Variant("s", self.encrypt_type)
             if self.encrypt_type
             else None,
-            "erase": Variant("s", self.erase.value) if self.erase else None,
+            "erase": Variant("s", self.erase) if self.erase else None,
             "update-partition-type": _optional_variant("b", self.update_partition_type),
             "no-block": _optional_variant("b", self.no_block),
             "dry-run-first": _optional_variant("b", self.dry_run_first),

supervisor/discovery/validate.py

@@ -33,7 +33,7 @@ SCHEMA_DISCOVERY = vol.Schema(
             {
                 vol.Required(ATTR_UUID): uuid_match,
                 vol.Required(ATTR_ADDON): str,
-                vol.Required(ATTR_SERVICE): valid_discovery_service,
+                vol.Required(ATTR_SERVICE): str,
                 vol.Required(ATTR_CONFIG): vol.Maybe(dict),
             },
             extra=vol.REMOVE_EXTRA,

supervisor/docker/addon.py

@@ -1,7 +1,6 @@
 """Init file for Supervisor add-on Docker object."""
 from __future__ import annotations
 
-import asyncio
 from collections.abc import Awaitable
 from contextlib import suppress
 from ipaddress import IPv4Address, ip_address
@@ -16,15 +15,10 @@ from docker.types import Mount
 import requests
 
 from ..addons.build import AddonBuild
+from ..addons.const import MappingType
 from ..bus import EventListener
 from ..const import (
     DOCKER_CPU_RUNTIME_ALLOCATION,
-    MAP_ADDONS,
-    MAP_BACKUP,
-    MAP_CONFIG,
-    MAP_MEDIA,
-    MAP_SHARE,
-    MAP_SSL,
     SECURITY_DISABLE,
     SECURITY_PROFILE,
     SYSTEMD_JOURNAL_PERSISTENT,
@@ -37,14 +31,15 @@ from ..exceptions import (
     CoreDNSError,
     DBusError,
     DockerError,
+    DockerJobError,
     DockerNotFound,
     HardwareNotFound,
 )
 from ..hardware.const import PolicyGroup
 from ..hardware.data import Device
-from ..jobs.decorator import Job, JobCondition, JobExecutionLimit
+from ..jobs.const import JobCondition, JobExecutionLimit
+from ..jobs.decorator import Job
 from ..resolution.const import ContextType, IssueType, SuggestionType
-from ..utils import process_lock
 from ..utils.sentry import capture_exception
 from .const import (
     ENV_TIME,
@@ -74,8 +69,8 @@ class DockerAddon(DockerInterface):
 
     def __init__(self, coresys: CoreSys, addon: Addon):
         """Initialize Docker Home Assistant wrapper."""
-        super().__init__(coresys)
         self.addon: Addon = addon
+        super().__init__(coresys)
 
         self._hw_listener: EventListener | None = None
 
@@ -278,7 +273,7 @@ class DockerAddon(DockerInterface):
         return None
 
     @property
-    def capabilities(self) -> list[str] | None:
+    def capabilities(self) -> list[Capabilities] | None:
         """Generate needed capabilities."""
         capabilities: set[Capabilities] = set(self.addon.privileged)
 
@@ -292,7 +287,7 @@ class DockerAddon(DockerInterface):
 
         # Return None if no capabilities is present
         if capabilities:
-            return [cap.value for cap in capabilities]
+            return list(capabilities)
         return None
 
     @property
@@ -329,76 +324,118 @@ class DockerAddon(DockerInterface):
         """Return mounts for container."""
         addon_mapping = self.addon.map_volumes
 
+        target_data_path = ""
+        if MappingType.DATA in addon_mapping:
+            target_data_path = addon_mapping[MappingType.DATA].path
+
         mounts = [
             MOUNT_DEV,
             Mount(
-                type=MountType.BIND.value,
+                type=MountType.BIND,
                 source=self.addon.path_extern_data.as_posix(),
-                target="/data",
+                target=target_data_path or "/data",
                 read_only=False,
             ),
         ]
 
         # setup config mappings
-        if MAP_CONFIG in addon_mapping:
+        if MappingType.CONFIG in addon_mapping:
             mounts.append(
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_homeassistant.as_posix(),
-                    target="/config",
-                    read_only=addon_mapping[MAP_CONFIG],
+                    target=addon_mapping[MappingType.CONFIG].path or "/config",
+                    read_only=addon_mapping[MappingType.CONFIG].read_only,
                 )
             )
 
-        if MAP_SSL in addon_mapping:
+        else:
+            # Map addon's public config folder if not using deprecated config option
+            if self.addon.addon_config_used:
+                mounts.append(
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.addon.path_extern_config.as_posix(),
+                        target=addon_mapping[MappingType.ADDON_CONFIG].path
+                        or "/config",
+                        read_only=addon_mapping[MappingType.ADDON_CONFIG].read_only,
+                    )
+                )
+
+            # Map Home Assistant config in new way
+            if MappingType.HOMEASSISTANT_CONFIG in addon_mapping:
+                mounts.append(
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.sys_config.path_extern_homeassistant.as_posix(),
+                        target=addon_mapping[MappingType.HOMEASSISTANT_CONFIG].path
+                        or "/homeassistant",
+                        read_only=addon_mapping[
+                            MappingType.HOMEASSISTANT_CONFIG
+                        ].read_only,
+                    )
+                )
+
+        if MappingType.ALL_ADDON_CONFIGS in addon_mapping:
             mounts.append(
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
+                    source=self.sys_config.path_extern_addon_configs.as_posix(),
+                    target=addon_mapping[MappingType.ALL_ADDON_CONFIGS].path
+                    or "/addon_configs",
+                    read_only=addon_mapping[MappingType.ALL_ADDON_CONFIGS].read_only,
+                )
+            )
+
+        if MappingType.SSL in addon_mapping:
+            mounts.append(
+                Mount(
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_ssl.as_posix(),
-                    target="/ssl",
-                    read_only=addon_mapping[MAP_SSL],
+                    target=addon_mapping[MappingType.SSL].path or "/ssl",
+                    read_only=addon_mapping[MappingType.SSL].read_only,
                 )
             )
 
-        if MAP_ADDONS in addon_mapping:
+        if MappingType.ADDONS in addon_mapping:
             mounts.append(
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_addons_local.as_posix(),
-                    target="/addons",
-                    read_only=addon_mapping[MAP_ADDONS],
+                    target=addon_mapping[MappingType.ADDONS].path or "/addons",
+                    read_only=addon_mapping[MappingType.ADDONS].read_only,
                 )
             )
 
-        if MAP_BACKUP in addon_mapping:
+        if MappingType.BACKUP in addon_mapping:
             mounts.append(
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_backup.as_posix(),
-                    target="/backup",
-                    read_only=addon_mapping[MAP_BACKUP],
+                    target=addon_mapping[MappingType.BACKUP].path or "/backup",
+                    read_only=addon_mapping[MappingType.BACKUP].read_only,
                 )
             )
 
-        if MAP_SHARE in addon_mapping:
+        if MappingType.SHARE in addon_mapping:
             mounts.append(
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_share.as_posix(),
-                    target="/share",
-                    read_only=addon_mapping[MAP_SHARE],
-                    propagation=PropagationMode.RSLAVE.value,
+                    target=addon_mapping[MappingType.SHARE].path or "/share",
+                    read_only=addon_mapping[MappingType.SHARE].read_only,
+                    propagation=PropagationMode.RSLAVE,
                 )
             )
 
-        if MAP_MEDIA in addon_mapping:
+        if MappingType.MEDIA in addon_mapping:
             mounts.append(
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_media.as_posix(),
-                    target="/media",
-                    read_only=addon_mapping[MAP_MEDIA],
-                    propagation=PropagationMode.RSLAVE.value,
+                    target=addon_mapping[MappingType.MEDIA].path or "/media",
+                    read_only=addon_mapping[MappingType.MEDIA].read_only,
+                    propagation=PropagationMode.RSLAVE,
                 )
             )
 
@@ -411,7 +448,7 @@ class DockerAddon(DockerInterface):
                     continue
                 mounts.append(
                     Mount(
-                        type=MountType.BIND.value,
+                        type=MountType.BIND,
                         source=gpio_path,
                         target=gpio_path,
                         read_only=False,
@@ -422,7 +459,7 @@ class DockerAddon(DockerInterface):
         if self.addon.with_devicetree:
             mounts.append(
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source="/sys/firmware/devicetree/base",
                     target="/device-tree",
                     read_only=True,
@@ -437,7 +474,7 @@ class DockerAddon(DockerInterface):
         if self.addon.with_kernel_modules:
             mounts.append(
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source="/lib/modules",
                     target="/lib/modules",
                     read_only=True,
@@ -456,19 +493,19 @@ class DockerAddon(DockerInterface):
         if self.addon.with_audio:
             mounts += [
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.addon.path_extern_pulse.as_posix(),
                     target="/etc/pulse/client.conf",
                     read_only=True,
                 ),
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_plugins.audio.path_extern_pulse.as_posix(),
                     target="/run/audio",
                     read_only=True,
                 ),
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_plugins.audio.path_extern_asound.as_posix(),
                     target="/etc/asound.conf",
                     read_only=True,
@@ -479,13 +516,13 @@ class DockerAddon(DockerInterface):
         if self.addon.with_journald:
             mounts += [
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=SYSTEMD_JOURNAL_PERSISTENT.as_posix(),
                     target=SYSTEMD_JOURNAL_PERSISTENT.as_posix(),
                     read_only=True,
                 ),
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=SYSTEMD_JOURNAL_VOLATILE.as_posix(),
                     target=SYSTEMD_JOURNAL_VOLATILE.as_posix(),
                     read_only=True,
@@ -494,28 +531,23 @@ class DockerAddon(DockerInterface):
 
         return mounts
 
-    def _run(self) -> None:
-        """Run Docker image.
-
-        Need run inside executor.
-        """
-        if self._is_running():
-            return
-
+    @Job(
+        name="docker_addon_run",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def run(self) -> None:
+        """Run Docker image."""
         # Security check
         if not self.addon.protected:
             _LOGGER.warning("%s running with disabled protected mode!", self.addon.name)
 
-        # Cleanup
-        self._stop()
-
         # Don't set a hostname if no separate UTS namespace is used
         hostname = None if self.uts_mode else self.addon.hostname
 
         # Create & Run container
         try:
-            docker_container = self.sys_docker.run(
-                self.image,
+            await self._run(
                 tag=str(self.addon.version),
                 name=self.name,
                 hostname=hostname,
@@ -546,14 +578,13 @@ class DockerAddon(DockerInterface):
             )
             raise
 
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Docker add-on %s with version %s", self.image, self.version
         )
 
         # Write data to DNS server
         try:
-            self.sys_plugins.dns.add_host(
+            await self.sys_plugins.dns.add_host(
                 ipv4=self.ip_address, names=[self.addon.hostname]
             )
         except CoreDNSError as err:
@@ -566,13 +597,19 @@ class DockerAddon(DockerInterface):
                 BusEvent.HARDWARE_NEW_DEVICE, self._hardware_events
             )
 
-    def _update(
-        self, version: AwesomeVersion, image: str | None = None, latest: bool = False
+    @Job(
+        name="docker_addon_update",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def update(
+        self,
+        version: AwesomeVersion,
+        image: str | None = None,
+        latest: bool = False,
+        arch: CpuArch | None = None,
     ) -> None:
-        """Update a docker image.
-
-        Need run inside executor.
-        """
+        """Update a docker image."""
         image = image or self.image
 
         _LOGGER.info(
@@ -580,15 +617,20 @@ class DockerAddon(DockerInterface):
         )
 
         # Update docker image
-        self._install(
-            version, image=image, latest=latest, need_build=self.addon.latest_need_build
+        await self.install(
+            version,
+            image=image,
+            latest=latest,
+            arch=arch,
+            need_build=self.addon.latest_need_build,
         )
 
-        # Stop container & cleanup
-        with suppress(DockerError):
-            self._stop()
-
-    def _install(
+    @Job(
+        name="docker_addon_install",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def install(
         self,
         version: AwesomeVersion,
         image: str | None = None,
@@ -597,20 +639,14 @@ class DockerAddon(DockerInterface):
         *,
         need_build: bool | None = None,
     ) -> None:
-        """Pull Docker image or build it.
-
-        Need run inside executor.
-        """
+        """Pull Docker image or build it."""
         if need_build is None and self.addon.need_build or need_build:
-            self._build(version)
+            await self._build(version)
         else:
-            super()._install(version, image, latest, arch)
+            await super().install(version, image, latest, arch)
 
-    def _build(self, version: AwesomeVersion) -> None:
-        """Build a Docker container.
-
-        Need run inside executor.
-        """
+    async def _build(self, version: AwesomeVersion) -> None:
+        """Build a Docker container."""
         build_env = AddonBuild(self.coresys, self.addon)
         if not build_env.is_valid:
             _LOGGER.error("Invalid build environment, can't build this add-on!")
@@ -618,8 +654,10 @@ class DockerAddon(DockerInterface):
 
         _LOGGER.info("Starting build for %s:%s", self.image, version)
         try:
-            image, log = self.sys_docker.images.build(
-                use_config_proxy=False, **build_env.get_docker_args(version)
+            image, log = await self.sys_run_in_executor(
+                self.sys_docker.images.build,
+                use_config_proxy=False,
+                **build_env.get_docker_args(version),
             )
 
             _LOGGER.debug("Build %s:%s done: %s", self.image, version, log)
@@ -642,77 +680,51 @@ class DockerAddon(DockerInterface):
 
         _LOGGER.info("Build %s:%s done", self.image, version)
 
-    @process_lock
+    @Job(
+        name="docker_addon_export_image",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
     def export_image(self, tar_file: Path) -> Awaitable[None]:
         """Export current images into a tar file."""
-        return self.sys_run_in_executor(self._export_image, tar_file)
+        return self.sys_run_in_executor(
+            self.sys_docker.export_image, self.image, self.version, tar_file
+        )
 
-    def _export_image(self, tar_file: Path) -> None:
-        """Export current images into a tar file.
-
-        Need run inside executor.
-        """
-        try:
-            image = self.sys_docker.api.get_image(f"{self.image}:{self.version}")
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            _LOGGER.error("Can't fetch image %s: %s", self.image, err)
-            raise DockerError() from err
-
-        _LOGGER.info("Export image %s to %s", self.image, tar_file)
-        try:
-            with tar_file.open("wb") as write_tar:
-                for chunk in image:
-                    write_tar.write(chunk)
-        except (OSError, requests.RequestException) as err:
-            _LOGGER.error("Can't write tar file %s: %s", tar_file, err)
-            raise DockerError() from err
-
-        _LOGGER.info("Export image %s done", self.image)
-
-    @process_lock
-    def import_image(self, tar_file: Path) -> Awaitable[None]:
+    @Job(
+        name="docker_addon_import_image",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def import_image(self, tar_file: Path) -> None:
         """Import a tar file as image."""
-        return self.sys_run_in_executor(self._import_image, tar_file)
+        docker_image = await self.sys_run_in_executor(
+            self.sys_docker.import_image, tar_file
+        )
+        if docker_image:
+            self._meta = docker_image.attrs
+            _LOGGER.info("Importing image %s and version %s", tar_file, self.version)
 
-    def _import_image(self, tar_file: Path) -> None:
-        """Import a tar file as image.
+            with suppress(DockerError):
+                await self.cleanup()
 
-        Need run inside executor.
-        """
-        try:
-            with tar_file.open("rb") as read_tar:
-                docker_image_list = self.sys_docker.images.load(read_tar)
-
-            if len(docker_image_list) != 1:
-                _LOGGER.warning(
-                    "Unexpected image count %d while importing image from tar",
-                    len(docker_image_list),
-                )
-                return
-            docker_image = docker_image_list[0]
-        except (docker.errors.DockerException, OSError) as err:
-            _LOGGER.error("Can't import image %s: %s", self.image, err)
-            raise DockerError() from err
-
-        self._meta = docker_image.attrs
-        _LOGGER.info("Importing image %s and version %s", tar_file, self.version)
-
-        with suppress(DockerError):
-            self._cleanup()
-
-    @process_lock
-    def write_stdin(self, data: bytes) -> Awaitable[None]:
+    @Job(
+        name="docker_addon_write_stdin",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def write_stdin(self, data: bytes) -> None:
         """Write to add-on stdin."""
-        return self.sys_run_in_executor(self._write_stdin, data)
+        if not await self.is_running():
+            raise DockerError()
+
+        await self.sys_run_in_executor(self._write_stdin, data)
 
     def _write_stdin(self, data: bytes) -> None:
         """Write to add-on stdin.
 
         Need run inside executor.
         """
-        if not self._is_running():
-            raise DockerError()
-
         try:
             # Load needed docker objects
             container = self.sys_docker.containers.get(self.name)
@@ -730,15 +742,17 @@ class DockerAddon(DockerInterface):
             _LOGGER.error("Can't write to %s stdin: %s", self.name, err)
             raise DockerError() from err
 
-    def _stop(self, remove_container=True) -> None:
-        """Stop/remove Docker container.
-
-        Need run inside executor.
-        """
+    @Job(
+        name="docker_addon_stop",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def stop(self, remove_container: bool = True) -> None:
+        """Stop/remove Docker container."""
         # DNS
         if self.ip_address != NO_ADDDRESS:
             try:
-                self.sys_plugins.dns.delete_host(self.addon.hostname)
+                await self.sys_plugins.dns.delete_host(self.addon.hostname)
             except CoreDNSError as err:
                 _LOGGER.warning("Can't update DNS for %s", self.name)
                 capture_exception(err)
@@ -748,9 +762,9 @@ class DockerAddon(DockerInterface):
             self.sys_bus.remove_listener(self._hw_listener)
             self._hw_listener = None
 
-        super()._stop(remove_container)
+        await super().stop(remove_container)
 
-    def _validate_trust(
+    async def _validate_trust(
         self, image_id: str, image: str, version: AwesomeVersion
     ) -> None:
         """Validate trust of content."""
@@ -758,13 +772,14 @@ class DockerAddon(DockerInterface):
             return
 
         checksum = image_id.partition(":")[2]
-        job = asyncio.run_coroutine_threadsafe(
-            self.sys_security.verify_content(self.addon.codenotary, checksum),
-            self.sys_loop,
-        )
-        job.result()
+        return await self.sys_security.verify_content(self.addon.codenotary, checksum)
 
-    @Job(conditions=[JobCondition.OS_AGENT], limit=JobExecutionLimit.SINGLE_WAIT)
+    @Job(
+        name="docker_addon_hardware_events",
+        conditions=[JobCondition.OS_AGENT],
+        limit=JobExecutionLimit.SINGLE_WAIT,
+        internal=True,
+    )
     async def _hardware_events(self, device: Device) -> None:
         """Process Hardware events for adjust device access."""
         if not any(

supervisor/docker/audio.py

@@ -6,7 +6,10 @@ from docker.types import Mount
 
 from ..const import DOCKER_CPU_RUNTIME_ALLOCATION, MACHINE_ID
 from ..coresys import CoreSysAttributes
+from ..exceptions import DockerJobError
 from ..hardware.const import PolicyGroup
+from ..jobs.const import JobExecutionLimit
+from ..jobs.decorator import Job
 from .const import (
     ENV_TIME,
     MOUNT_DBUS,
@@ -42,7 +45,7 @@ class DockerAudio(DockerInterface, CoreSysAttributes):
         mounts = [
             MOUNT_DEV,
             Mount(
-                type=MountType.BIND.value,
+                type=MountType.BIND,
                 source=self.sys_config.path_extern_audio.as_posix(),
                 target="/data",
                 read_only=False,
@@ -65,9 +68,9 @@ class DockerAudio(DockerInterface, CoreSysAttributes):
         ) + self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.BLUETOOTH)
 
     @property
-    def capabilities(self) -> list[str]:
+    def capabilities(self) -> list[Capabilities]:
         """Generate needed capabilities."""
-        return [cap.value for cap in (Capabilities.SYS_NICE, Capabilities.SYS_RESOURCE)]
+        return [Capabilities.SYS_NICE, Capabilities.SYS_RESOURCE]
 
     @property
     def ulimits(self) -> list[docker.types.Ulimit]:
@@ -82,20 +85,14 @@ class DockerAudio(DockerInterface, CoreSysAttributes):
             return None
         return DOCKER_CPU_RUNTIME_ALLOCATION
 
-    def _run(self) -> None:
-        """Run Docker image.
-
-        Need run inside executor.
-        """
-        if self._is_running():
-            return
-
-        # Cleanup
-        self._stop()
-
-        # Create & Run container
-        docker_container = self.sys_docker.run(
-            self.image,
+    @Job(
+        name="docker_audio_run",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def run(self) -> None:
+        """Run Docker image."""
+        await self._run(
             tag=str(self.sys_plugins.audio.version),
             init=False,
             ipv4=self.sys_docker.network.audio,
@@ -112,8 +109,6 @@ class DockerAudio(DockerInterface, CoreSysAttributes):
             },
             mounts=self.mounts,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Audio %s with version %s - %s",
             self.image,

supervisor/docker/cli.py

@@ -2,6 +2,9 @@
 import logging
 
 from ..coresys import CoreSysAttributes
+from ..exceptions import DockerJobError
+from ..jobs.const import JobExecutionLimit
+from ..jobs.decorator import Job
 from .const import ENV_TIME, ENV_TOKEN
 from .interface import DockerInterface
 
@@ -23,20 +26,14 @@ class DockerCli(DockerInterface, CoreSysAttributes):
         """Return name of Docker container."""
         return CLI_DOCKER_NAME
 
-    def _run(self) -> None:
-        """Run Docker image.
-
-        Need run inside executor.
-        """
-        if self._is_running():
-            return
-
-        # Cleanup
-        self._stop()
-
-        # Create & Run container
-        docker_container = self.sys_docker.run(
-            self.image,
+    @Job(
+        name="docker_cli_run",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def run(self) -> None:
+        """Run Docker image."""
+        await self._run(
             entrypoint=["/init"],
             tag=str(self.sys_plugins.cli.version),
             init=False,
@@ -54,8 +51,6 @@ class DockerCli(DockerInterface, CoreSysAttributes):
                 ENV_TOKEN: self.sys_plugins.cli.supervisor_token,
             },
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting CLI %s with version %s - %s",
             self.image,

supervisor/docker/const.py

@@ -1,12 +1,12 @@
 """Docker constants."""
-from enum import Enum
+from enum import StrEnum
 
 from docker.types import Mount
 
 from ..const import MACHINE_ID
 
 
-class Capabilities(str, Enum):
+class Capabilities(StrEnum):
     """Linux Capabilities."""
 
     BPF = "BPF"
@@ -24,7 +24,7 @@ class Capabilities(str, Enum):
     SYS_TIME = "SYS_TIME"
 
 
-class ContainerState(str, Enum):
+class ContainerState(StrEnum):
     """State of supervisor managed docker container."""
 
     FAILED = "failed"
@@ -35,7 +35,7 @@ class ContainerState(str, Enum):
     UNKNOWN = "unknown"
 
 
-class RestartPolicy(str, Enum):
+class RestartPolicy(StrEnum):
     """Restart policy of container."""
 
     NO = "no"
@@ -44,7 +44,7 @@ class RestartPolicy(str, Enum):
     ALWAYS = "always"
 
 
-class MountType(str, Enum):
+class MountType(StrEnum):
     """Mount type."""
 
     BIND = "bind"
@@ -53,7 +53,7 @@ class MountType(str, Enum):
     NPIPE = "npipe"
 
 
-class PropagationMode(str, Enum):
+class PropagationMode(StrEnum):
     """Propagataion mode, only for bind type mounts."""
 
     PRIVATE = "private"
@@ -71,23 +71,21 @@ ENV_TOKEN_OLD = "HASSIO_TOKEN"
 LABEL_MANAGED = "supervisor_managed"
 
 MOUNT_DBUS = Mount(
-    type=MountType.BIND.value, source="/run/dbus", target="/run/dbus", read_only=True
-)
-MOUNT_DEV = Mount(
-    type=MountType.BIND.value, source="/dev", target="/dev", read_only=True
+    type=MountType.BIND, source="/run/dbus", target="/run/dbus", read_only=True
 )
+MOUNT_DEV = Mount(type=MountType.BIND, source="/dev", target="/dev", read_only=True)
 MOUNT_DOCKER = Mount(
-    type=MountType.BIND.value,
+    type=MountType.BIND,
     source="/run/docker.sock",
     target="/run/docker.sock",
     read_only=True,
 )
 MOUNT_MACHINE_ID = Mount(
-    type=MountType.BIND.value,
+    type=MountType.BIND,
     source=MACHINE_ID.as_posix(),
     target=MACHINE_ID.as_posix(),
     read_only=True,
 )
 MOUNT_UDEV = Mount(
-    type=MountType.BIND.value, source="/run/udev", target="/run/udev", read_only=True
+    type=MountType.BIND, source="/run/udev", target="/run/udev", read_only=True
 )

supervisor/docker/dns.py

@@ -4,6 +4,9 @@ import logging
 from docker.types import Mount
 
 from ..coresys import CoreSysAttributes
+from ..exceptions import DockerJobError
+from ..jobs.const import JobExecutionLimit
+from ..jobs.decorator import Job
 from .const import ENV_TIME, MOUNT_DBUS, MountType
 from .interface import DockerInterface
 
@@ -25,20 +28,14 @@ class DockerDNS(DockerInterface, CoreSysAttributes):
         """Return name of Docker container."""
         return DNS_DOCKER_NAME
 
-    def _run(self) -> None:
-        """Run Docker image.
-
-        Need run inside executor.
-        """
-        if self._is_running():
-            return
-
-        # Cleanup
-        self._stop()
-
-        # Create & Run container
-        docker_container = self.sys_docker.run(
-            self.image,
+    @Job(
+        name="docker_dns_run",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def run(self) -> None:
+        """Run Docker image."""
+        await self._run(
             tag=str(self.sys_plugins.dns.version),
             init=False,
             dns=False,
@@ -50,7 +47,7 @@ class DockerDNS(DockerInterface, CoreSysAttributes):
             environment={ENV_TIME: self.sys_timezone},
             mounts=[
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_dns.as_posix(),
                     target="/config",
                     read_only=False,
@@ -59,8 +56,6 @@ class DockerDNS(DockerInterface, CoreSysAttributes):
             ],
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting DNS %s with version %s - %s",
             self.image,

supervisor/docker/homeassistant.py

@@ -4,14 +4,14 @@ from ipaddress import IPv4Address
 import logging
 
 from awesomeversion import AwesomeVersion, AwesomeVersionCompareException
-import docker
 from docker.types import Mount
-import requests
 
 from ..const import LABEL_MACHINE, MACHINE_ID
-from ..exceptions import DockerError
+from ..exceptions import DockerJobError
 from ..hardware.const import PolicyGroup
 from ..homeassistant.const import LANDINGPAGE
+from ..jobs.const import JobExecutionLimit
+from ..jobs.decorator import Job
 from .const import (
     ENV_TIME,
     ENV_TOKEN,
@@ -53,9 +53,10 @@ class DockerHomeAssistant(DockerInterface):
     @property
     def timeout(self) -> int:
         """Return timeout for Docker actions."""
-        # Synchronized homeassistant's S6_SERVICES_GRACETIME
-        # to avoid killing Home Assistant Core
-        return 220 + 20
+        # Synchronized with the homeassistant core container's S6_SERVICES_GRACETIME
+        # to avoid killing Home Assistant Core, see
+        # https://github.com/home-assistant/core/tree/dev/Dockerfile
+        return 240 + 20
 
     @property
     def ip_address(self) -> IPv4Address:
@@ -66,10 +67,14 @@ class DockerHomeAssistant(DockerInterface):
     def cgroups_rules(self) -> list[str]:
         """Return a list of needed cgroups permission."""
         return (
-            self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.UART)
-            + self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.VIDEO)
-            + self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.GPIO)
-            + self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.USB)
+            []
+            if self.sys_homeassistant.version == LANDINGPAGE
+            else (
+                self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.UART)
+                + self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.VIDEO)
+                + self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.GPIO)
+                + self.sys_hardware.policy.get_cgroups_rules(PolicyGroup.USB)
+            )
         )
 
     @property
@@ -79,79 +84,81 @@ class DockerHomeAssistant(DockerInterface):
             MOUNT_DEV,
             MOUNT_DBUS,
             MOUNT_UDEV,
-            # Add folders
+            # HA config folder
             Mount(
-                type=MountType.BIND.value,
+                type=MountType.BIND,
                 source=self.sys_config.path_extern_homeassistant.as_posix(),
                 target="/config",
                 read_only=False,
             ),
-            Mount(
-                type=MountType.BIND.value,
-                source=self.sys_config.path_extern_ssl.as_posix(),
-                target="/ssl",
-                read_only=True,
-            ),
-            Mount(
-                type=MountType.BIND.value,
-                source=self.sys_config.path_extern_share.as_posix(),
-                target="/share",
-                read_only=False,
-                propagation=PropagationMode.RSLAVE.value,
-            ),
-            Mount(
-                type=MountType.BIND.value,
-                source=self.sys_config.path_extern_media.as_posix(),
-                target="/media",
-                read_only=False,
-                propagation=PropagationMode.RSLAVE.value,
-            ),
-            # Configuration audio
-            Mount(
-                type=MountType.BIND.value,
-                source=self.sys_homeassistant.path_extern_pulse.as_posix(),
-                target="/etc/pulse/client.conf",
-                read_only=True,
-            ),
-            Mount(
-                type=MountType.BIND.value,
-                source=self.sys_plugins.audio.path_extern_pulse.as_posix(),
-                target="/run/audio",
-                read_only=True,
-            ),
-            Mount(
-                type=MountType.BIND.value,
-                source=self.sys_plugins.audio.path_extern_asound.as_posix(),
-                target="/etc/asound.conf",
-                read_only=True,
-            ),
         ]
 
+        # Landingpage does not need all this access
+        if self.sys_homeassistant.version != LANDINGPAGE:
+            mounts.extend(
+                [
+                    # All other folders
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.sys_config.path_extern_ssl.as_posix(),
+                        target="/ssl",
+                        read_only=True,
+                    ),
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.sys_config.path_extern_share.as_posix(),
+                        target="/share",
+                        read_only=False,
+                        propagation=PropagationMode.RSLAVE.value,
+                    ),
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.sys_config.path_extern_media.as_posix(),
+                        target="/media",
+                        read_only=False,
+                        propagation=PropagationMode.RSLAVE.value,
+                    ),
+                    # Configuration audio
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.sys_homeassistant.path_extern_pulse.as_posix(),
+                        target="/etc/pulse/client.conf",
+                        read_only=True,
+                    ),
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.sys_plugins.audio.path_extern_pulse.as_posix(),
+                        target="/run/audio",
+                        read_only=True,
+                    ),
+                    Mount(
+                        type=MountType.BIND,
+                        source=self.sys_plugins.audio.path_extern_asound.as_posix(),
+                        target="/etc/asound.conf",
+                        read_only=True,
+                    ),
+                ]
+            )
+
         # Machine ID
         if MACHINE_ID.exists():
             mounts.append(MOUNT_MACHINE_ID)
 
         return mounts
 
-    def _run(self) -> None:
-        """Run Docker image.
-
-        Need run inside executor.
-        """
-        if self._is_running():
-            return
-
-        # Cleanup
-        self._stop()
-
-        # Create & Run container
-        docker_container = self.sys_docker.run(
-            self.image,
+    @Job(
+        name="docker_home_assistant_run",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def run(self) -> None:
+        """Run Docker image."""
+        await self._run(
             tag=(self.sys_homeassistant.version),
             name=self.name,
             hostname=self.name,
             detach=True,
-            privileged=True,
+            privileged=self.sys_homeassistant.version != LANDINGPAGE,
             init=False,
             security_opt=self.security_opt,
             network_mode="host",
@@ -171,18 +178,19 @@ class DockerHomeAssistant(DockerInterface):
             tmpfs={"/tmp": ""},
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Home Assistant %s with version %s", self.image, self.version
         )
 
-    def _execute_command(self, command: str) -> CommandReturn:
-        """Create a temporary container and run command.
-
-        Need run inside executor.
-        """
-        return self.sys_docker.run_command(
+    @Job(
+        name="docker_home_assistant_execute_command",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def execute_command(self, command: str) -> CommandReturn:
+        """Create a temporary container and run command."""
+        return await self.sys_run_in_executor(
+            self.sys_docker.run_command,
             self.image,
             version=self.sys_homeassistant.version,
             command=command,
@@ -194,19 +202,19 @@ class DockerHomeAssistant(DockerInterface):
             stderr=True,
             mounts=[
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_homeassistant.as_posix(),
                     target="/config",
                     read_only=False,
                 ),
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_ssl.as_posix(),
                     target="/ssl",
                     read_only=True,
                 ),
                 Mount(
-                    type=MountType.BIND.value,
+                    type=MountType.BIND,
                     source=self.sys_config.path_extern_share.as_posix(),
                     target="/share",
                     read_only=False,
@@ -217,34 +225,14 @@ class DockerHomeAssistant(DockerInterface):
 
     def is_initialize(self) -> Awaitable[bool]:
         """Return True if Docker container exists."""
-        return self.sys_run_in_executor(self._is_initialize)
+        return self.sys_run_in_executor(
+            self.sys_docker.container_is_initialized,
+            self.name,
+            self.image,
+            self.sys_homeassistant.version,
+        )
 
-    def _is_initialize(self) -> bool:
-        """Return True if docker container exists.
-
-        Need run inside executor.
-        """
-        try:
-            docker_container = self.sys_docker.containers.get(self.name)
-            docker_image = self.sys_docker.images.get(
-                f"{self.image}:{self.sys_homeassistant.version}"
-            )
-        except docker.errors.NotFound:
-            return False
-        except (docker.errors.DockerException, requests.RequestException):
-            return DockerError()
-
-        # we run on an old image, stop and start it
-        if docker_container.image.id != docker_image.id:
-            return False
-
-        # Check of correct state
-        if docker_container.status not in ("exited", "running", "created"):
-            return False
-
-        return True
-
-    def _validate_trust(
+    async def _validate_trust(
         self, image_id: str, image: str, version: AwesomeVersion
     ) -> None:
         """Validate trust of content."""
@@ -254,4 +242,4 @@ class DockerHomeAssistant(DockerInterface):
         except AwesomeVersionCompareException:
             return
 
-        super()._validate_trust(image_id, image, version)
+        await super()._validate_trust(image_id, image, version)

supervisor/docker/interface.py

@@ -1,13 +1,14 @@
 """Interface class for Supervisor Docker object."""
 from __future__ import annotations
 
-import asyncio
+from collections import defaultdict
 from collections.abc import Awaitable
 from contextlib import suppress
 import logging
 import re
 from time import time
 from typing import Any
+from uuid import uuid4
 
 from awesomeversion import AwesomeVersion
 from awesomeversion.strategy import AwesomeVersionStrategy
@@ -24,18 +25,21 @@ from ..const import (
     BusEvent,
     CpuArch,
 )
-from ..coresys import CoreSys, CoreSysAttributes
+from ..coresys import CoreSys
 from ..exceptions import (
     CodeNotaryError,
     CodeNotaryUntrusted,
     DockerAPIError,
     DockerError,
+    DockerJobError,
     DockerNotFound,
     DockerRequestError,
     DockerTrustError,
 )
+from ..jobs.const import JOB_GROUP_DOCKER_INTERFACE, JobExecutionLimit
+from ..jobs.decorator import Job
+from ..jobs.job_group import JobGroup
 from ..resolution.const import ContextType, IssueType, SuggestionType
-from ..utils import process_lock
 from ..utils.sentry import capture_exception
 from .const import ContainerState, RestartPolicy
 from .manager import CommandReturn
@@ -73,14 +77,20 @@ def _container_state_from_model(docker_container: Container) -> ContainerState:
     return ContainerState.STOPPED
 
 
-class DockerInterface(CoreSysAttributes):
+class DockerInterface(JobGroup):
     """Docker Supervisor interface."""
 
     def __init__(self, coresys: CoreSys):
         """Initialize Docker base wrapper."""
+        super().__init__(
+            coresys,
+            JOB_GROUP_DOCKER_INTERFACE.format_map(
+                defaultdict(str, name=self.name or uuid4().hex)
+            ),
+            self.name,
+        )
         self.coresys: CoreSys = coresys
         self._meta: dict[str, Any] | None = None
-        self.lock: asyncio.Lock = asyncio.Lock()
 
     @property
     def timeout(self) -> int:
@@ -141,7 +151,7 @@ class DockerInterface(CoreSysAttributes):
     @property
     def in_progress(self) -> bool:
         """Return True if a task is in progress."""
-        return self.lock.locked()
+        return self.active_job
 
     @property
     def restart_policy(self) -> RestartPolicy | None:
@@ -193,7 +203,7 @@ class DockerInterface(CoreSysAttributes):
 
         return credentials
 
-    def _docker_login(self, image: str) -> None:
+    async def _docker_login(self, image: str) -> None:
         """Try to log in to the registry if there are credentials available."""
         if not self.sys_docker.config.registries:
             return
@@ -202,30 +212,21 @@ class DockerInterface(CoreSysAttributes):
         if not credentials:
             return
 
-        self.sys_docker.docker.login(**credentials)
+        await self.sys_run_in_executor(self.sys_docker.docker.login, **credentials)
 
-    @process_lock
-    def install(
-        self,
-        version: AwesomeVersion,
-        image: str | None = None,
-        latest: bool = False,
-        arch: CpuArch | None = None,
-    ):
-        """Pull docker image."""
-        return self.sys_run_in_executor(self._install, version, image, latest, arch)
-
-    def _install(
+    @Job(
+        name="docker_interface_install",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def install(
         self,
         version: AwesomeVersion,
         image: str | None = None,
         latest: bool = False,
         arch: CpuArch | None = None,
     ) -> None:
-        """Pull Docker image.
-
-        Need run inside executor.
-        """
+        """Pull docker image."""
         image = image or self.image
         arch = arch or self.sys_arch.supervisor
 
@@ -233,21 +234,24 @@ class DockerInterface(CoreSysAttributes):
         try:
             if self.sys_docker.config.registries:
                 # Try login if we have defined credentials
-                self._docker_login(image)
+                await self._docker_login(image)
 
             # Pull new image
-            docker_image = self.sys_docker.images.pull(
+            docker_image = await self.sys_run_in_executor(
+                self.sys_docker.images.pull,
                 f"{image}:{version!s}",
                 platform=MAP_ARCH[arch],
             )
 
             # Validate content
             try:
-                self._validate_trust(docker_image.id, image, version)
+                await self._validate_trust(docker_image.id, image, version)
             except CodeNotaryError:
                 with suppress(docker.errors.DockerException):
-                    self.sys_docker.images.remove(
-                        image=f"{image}:{version!s}", force=True
+                    await self.sys_run_in_executor(
+                        self.sys_docker.images.remove,
+                        image=f"{image}:{version!s}",
+                        force=True,
                     )
                 raise
 
@@ -256,7 +260,7 @@ class DockerInterface(CoreSysAttributes):
                 _LOGGER.info(
                     "Tagging image %s with version %s as latest", image, version
                 )
-                docker_image.tag(image, tag="latest")
+                await self.sys_run_in_executor(docker_image.tag, image, tag="latest")
         except docker.errors.APIError as err:
             if err.status_code == 429:
                 self.sys_resolution.create_issue(
@@ -289,34 +293,21 @@ class DockerInterface(CoreSysAttributes):
 
         self._meta = docker_image.attrs
 
-    def exists(self) -> Awaitable[bool]:
+    async def exists(self) -> bool:
         """Return True if Docker image exists in local repository."""
-        return self.sys_run_in_executor(self._exists)
-
-    def _exists(self) -> bool:
-        """Return True if Docker image exists in local repository.
-
-        Need run inside executor.
-        """
         with suppress(docker.errors.DockerException, requests.RequestException):
-            self.sys_docker.images.get(f"{self.image}:{self.version!s}")
+            await self.sys_run_in_executor(
+                self.sys_docker.images.get, f"{self.image}:{self.version!s}"
+            )
             return True
         return False
 
-    def is_running(self) -> Awaitable[bool]:
-        """Return True if Docker is running.
-
-        Return a Future.
-        """
-        return self.sys_run_in_executor(self._is_running)
-
-    def _is_running(self) -> bool:
-        """Return True if Docker is running.
-
-        Need run inside executor.
-        """
+    async def is_running(self) -> bool:
+        """Return True if Docker is running."""
         try:
-            docker_container = self.sys_docker.containers.get(self.name)
+            docker_container = await self.sys_run_in_executor(
+                self.sys_docker.containers.get, self.name
+            )
         except docker.errors.NotFound:
             return False
         except docker.errors.DockerException as err:
@@ -326,20 +317,12 @@ class DockerInterface(CoreSysAttributes):
 
         return docker_container.status == "running"
 
-    def current_state(self) -> Awaitable[ContainerState]:
-        """Return current state of container.
-
-        Return a Future.
-        """
-        return self.sys_run_in_executor(self._current_state)
-
-    def _current_state(self) -> ContainerState:
-        """Return current state of container.
-
-        Need run inside executor.
-        """
+    async def current_state(self) -> ContainerState:
+        """Return current state of container."""
         try:
-            docker_container = self.sys_docker.containers.get(self.name)
+            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:
@@ -349,22 +332,15 @@ class DockerInterface(CoreSysAttributes):
 
         return _container_state_from_model(docker_container)
 
-    @process_lock
-    def attach(
+    @Job(name="docker_interface_attach", limit=JobExecutionLimit.GROUP_WAIT)
+    async def attach(
         self, version: AwesomeVersion, *, skip_state_event_if_down: bool = False
-    ) -> Awaitable[None]:
-        """Attach to running Docker container."""
-        return self.sys_run_in_executor(self._attach, version, skip_state_event_if_down)
-
-    def _attach(
-        self, version: AwesomeVersion, skip_state_event_if_down: bool = False
     ) -> None:
-        """Attach to running docker container.
-
-        Need run inside executor.
-        """
+        """Attach to running Docker container."""
         with suppress(docker.errors.DockerException, requests.RequestException):
-            docker_container = self.sys_docker.containers.get(self.name)
+            docker_container = await self.sys_run_in_executor(
+                self.sys_docker.containers.get, self.name
+            )
             self._meta = docker_container.attrs
             self.sys_docker.monitor.watch_container(docker_container)
 
@@ -374,8 +350,7 @@ class DockerInterface(CoreSysAttributes):
                 and state in [ContainerState.STOPPED, ContainerState.FAILED]
             ):
                 # Fire event with current state of container
-                self.sys_loop.call_soon_threadsafe(
-                    self.sys_bus.fire_event,
+                self.sys_bus.fire_event(
                     BusEvent.DOCKER_CONTAINER_STATE_CHANGE,
                     DockerContainerStateEvent(
                         self.name, state, docker_container.id, int(time())
@@ -393,114 +368,85 @@ class DockerInterface(CoreSysAttributes):
             raise DockerError()
         _LOGGER.info("Attaching to %s with version %s", self.image, self.version)
 
-    @process_lock
-    def run(self) -> Awaitable[None]:
+    @Job(
+        name="docker_interface_run",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def run(self) -> None:
         """Run Docker image."""
-        return self.sys_run_in_executor(self._run)
-
-    def _run(self) -> None:
-        """Run Docker image.
-
-        Need run inside executor.
-        """
         raise NotImplementedError()
 
-    @process_lock
-    def stop(self, remove_container=True) -> Awaitable[None]:
-        """Stop/remove Docker container."""
-        return self.sys_run_in_executor(self._stop, remove_container)
-
-    def _stop(self, remove_container=True) -> None:
-        """Stop/remove Docker container.
-
-        Need run inside executor.
-        """
-        try:
-            docker_container = self.sys_docker.containers.get(self.name)
-        except docker.errors.NotFound:
+    async def _run(self, **kwargs) -> None:
+        """Run Docker image with retry inf necessary."""
+        if await self.is_running():
             return
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError() from err
 
-        if docker_container.status == "running":
-            _LOGGER.info("Stopping %s application", self.name)
-            with suppress(docker.errors.DockerException, requests.RequestException):
-                docker_container.stop(timeout=self.timeout)
+        # Cleanup
+        await self.stop()
 
-        if remove_container:
-            with suppress(docker.errors.DockerException, requests.RequestException):
-                _LOGGER.info("Cleaning %s application", self.name)
-                docker_container.remove(force=True)
+        # Create & Run container
+        try:
+            docker_container = await self.sys_run_in_executor(
+                self.sys_docker.run, self.image, **kwargs
+            )
+        except DockerNotFound as err:
+            # If image is missing, capture the exception as this shouldn't happen
+            capture_exception(err)
+            raise
 
-    @process_lock
+        # Store metadata
+        self._meta = docker_container.attrs
+
+    @Job(
+        name="docker_interface_stop",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def stop(self, remove_container: bool = True) -> None:
+        """Stop/remove Docker container."""
+        with suppress(DockerNotFound):
+            await self.sys_run_in_executor(
+                self.sys_docker.stop_container,
+                self.name,
+                self.timeout,
+                remove_container,
+            )
+
+    @Job(
+        name="docker_interface_start",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
     def start(self) -> Awaitable[None]:
         """Start Docker container."""
-        return self.sys_run_in_executor(self._start)
+        return self.sys_run_in_executor(self.sys_docker.start_container, self.name)
 
-    def _start(self) -> None:
-        """Start docker container.
-
-        Need run inside executor.
-        """
-        try:
-            docker_container = self.sys_docker.containers.get(self.name)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError(
-                f"{self.name} not found for starting up", _LOGGER.error
-            ) from err
-
-        _LOGGER.info("Starting %s", self.name)
-        try:
-            docker_container.start()
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError(f"Can't start {self.name}: {err}", _LOGGER.error) from err
-
-    @process_lock
-    def remove(self) -> Awaitable[None]:
+    @Job(
+        name="docker_interface_remove",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def remove(self) -> None:
         """Remove Docker images."""
-        return self.sys_run_in_executor(self._remove)
-
-    def _remove(self) -> None:
-        """Remove docker images.
-
-        Needs run inside executor.
-        """
         # Cleanup container
         with suppress(DockerError):
-            self._stop()
-
-        _LOGGER.info("Removing image %s with latest and %s", self.image, self.version)
-
-        try:
-            with suppress(docker.errors.ImageNotFound):
-                self.sys_docker.images.remove(image=f"{self.image}:latest", force=True)
-
-            with suppress(docker.errors.ImageNotFound):
-                self.sys_docker.images.remove(
-                    image=f"{self.image}:{self.version!s}", force=True
-                )
-
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError(
-                f"Can't remove image {self.image}: {err}", _LOGGER.warning
-            ) from err
+            await self.stop()
 
+        await self.sys_run_in_executor(
+            self.sys_docker.remove_image, self.image, self.version
+        )
         self._meta = None
 
-    @process_lock
-    def update(
-        self, version: AwesomeVersion, image: str | None = None, latest: bool = False
-    ) -> Awaitable[None]:
-        """Update a Docker image."""
-        return self.sys_run_in_executor(self._update, version, image, latest)
-
-    def _update(
+    @Job(
+        name="docker_interface_update",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def update(
         self, version: AwesomeVersion, image: str | None = None, latest: bool = False
     ) -> None:
-        """Update a docker image.
-
-        Need run inside executor.
-        """
+        """Update a Docker image."""
         image = image or self.image
 
         _LOGGER.info(
@@ -508,163 +454,69 @@ class DockerInterface(CoreSysAttributes):
         )
 
         # Update docker image
-        self._install(version, image=image, latest=latest)
+        await self.install(version, image=image, latest=latest)
 
         # Stop container & cleanup
         with suppress(DockerError):
-            self._stop()
+            await self.stop()
 
-    def logs(self) -> Awaitable[bytes]:
-        """Return Docker logs of container.
-
-        Return a Future.
-        """
-        return self.sys_run_in_executor(self._logs)
-
-    def _logs(self) -> bytes:
-        """Return Docker logs of container.
-
-        Need run inside executor.
-        """
-        try:
-            docker_container = self.sys_docker.containers.get(self.name)
-        except (docker.errors.DockerException, requests.RequestException):
-            return b""
-
-        try:
-            return docker_container.logs(tail=100, stdout=True, stderr=True)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            _LOGGER.warning("Can't grep logs from %s: %s", self.image, err)
+    async def logs(self) -> bytes:
+        """Return Docker logs of container."""
+        with suppress(DockerError):
+            return await self.sys_run_in_executor(
+                self.sys_docker.container_logs, self.name
+            )
 
         return b""
 
-    @process_lock
-    def cleanup(self, old_image: str | None = None) -> Awaitable[None]:
+    @Job(name="docker_interface_cleanup", limit=JobExecutionLimit.GROUP_WAIT)
+    def cleanup(
+        self,
+        old_image: str | None = None,
+        image: str | None = None,
+        version: AwesomeVersion | None = None,
+    ) -> Awaitable[None]:
         """Check if old version exists and cleanup."""
-        return self.sys_run_in_executor(self._cleanup, old_image)
+        return 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,
+        )
 
-    def _cleanup(self, old_image: str | None = None) -> None:
-        """Check if old version exists and cleanup.
-
-        Need run inside executor.
-        """
-        try:
-            origin = self.sys_docker.images.get(f"{self.image}:{self.version!s}")
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError(
-                f"Can't find {self.image} for cleanup", _LOGGER.warning
-            ) from err
-
-        # Cleanup Current
-        try:
-            images_list = self.sys_docker.images.list(name=self.image)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError(
-                f"Corrupt docker overlayfs found: {err}", _LOGGER.warning
-            ) from err
-
-        for image in images_list:
-            if origin.id == image.id:
-                continue
-
-            with suppress(docker.errors.DockerException, requests.RequestException):
-                _LOGGER.info("Cleanup images: %s", image.tags)
-                self.sys_docker.images.remove(image.id, force=True)
-
-        # Cleanup Old
-        if not old_image or self.image == old_image:
-            return
-
-        try:
-            images_list = self.sys_docker.images.list(name=old_image)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError(
-                f"Corrupt docker overlayfs found: {err}", _LOGGER.warning
-            ) from err
-
-        for image in images_list:
-            if origin.id == image.id:
-                continue
-
-            with suppress(docker.errors.DockerException, requests.RequestException):
-                _LOGGER.info("Cleanup images: %s", image.tags)
-                self.sys_docker.images.remove(image.id, force=True)
-
-    @process_lock
+    @Job(
+        name="docker_interface_restart",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
     def restart(self) -> Awaitable[None]:
         """Restart docker container."""
-        return self.sys_loop.run_in_executor(None, self._restart)
+        return self.sys_run_in_executor(
+            self.sys_docker.restart_container, self.name, self.timeout
+        )
 
-    def _restart(self) -> None:
-        """Restart docker container.
-
-        Need run inside executor.
-        """
-        try:
-            container = self.sys_docker.containers.get(self.name)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError() from err
-
-        _LOGGER.info("Restarting %s", self.image)
-        try:
-            container.restart(timeout=self.timeout)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError(
-                f"Can't restart {self.image}: {err}", _LOGGER.warning
-            ) from err
-
-    @process_lock
-    def execute_command(self, command: str) -> Awaitable[CommandReturn]:
+    @Job(
+        name="docker_interface_execute_command",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def execute_command(self, command: str) -> CommandReturn:
         """Create a temporary container and run command."""
-        return self.sys_run_in_executor(self._execute_command, command)
-
-    def _execute_command(self, command: str) -> CommandReturn:
-        """Create a temporary container and run command.
-
-        Need run inside executor.
-        """
         raise NotImplementedError()
 
-    def stats(self) -> Awaitable[DockerStats]:
+    async def stats(self) -> DockerStats:
         """Read and return stats from container."""
-        return self.sys_run_in_executor(self._stats)
+        stats = await self.sys_run_in_executor(
+            self.sys_docker.container_stats, self.name
+        )
+        return DockerStats(stats)
 
-    def _stats(self) -> DockerStats:
-        """Create a temporary container and run command.
-
-        Need run inside executor.
-        """
+    async def is_failed(self) -> bool:
+        """Return True if Docker is failing state."""
         try:
-            docker_container = self.sys_docker.containers.get(self.name)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError() from err
-
-        # container is not running
-        if docker_container.status != "running":
-            raise DockerError(f"Container {self.name} is not running", _LOGGER.error)
-
-        try:
-            stats = docker_container.stats(stream=False)
-            return DockerStats(stats)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError(
-                f"Can't read stats from {self.name}: {err}", _LOGGER.error
-            ) from err
-
-    def is_failed(self) -> Awaitable[bool]:
-        """Return True if Docker is failing state.
-
-        Return a Future.
-        """
-        return self.sys_run_in_executor(self._is_failed)
-
-    def _is_failed(self) -> bool:
-        """Return True if Docker is failing state.
-
-        Need run inside executor.
-        """
-        try:
-            docker_container = self.sys_docker.containers.get(self.name)
+            docker_container = await self.sys_run_in_executor(
+                self.sys_docker.containers.get, self.name
+            )
         except docker.errors.NotFound:
             return False
         except (docker.errors.DockerException, requests.RequestException) as err:
@@ -677,18 +529,13 @@ class DockerInterface(CoreSysAttributes):
         # Check return value
         return int(docker_container.attrs["State"]["ExitCode"]) != 0
 
-    def get_latest_version(self) -> Awaitable[AwesomeVersion]:
+    async def get_latest_version(self) -> AwesomeVersion:
         """Return latest version of local image."""
-        return self.sys_run_in_executor(self._get_latest_version)
-
-    def _get_latest_version(self) -> AwesomeVersion:
-        """Return latest version of local image.
-
-        Need run inside executor.
-        """
         available_version: list[AwesomeVersion] = []
         try:
-            for image in self.sys_docker.images.list(self.image):
+            for image in await self.sys_run_in_executor(
+                self.sys_docker.images.list, self.image
+            ):
                 for tag in image.tags:
                     version = AwesomeVersion(tag.partition(":")[2])
                     if version.strategy == AwesomeVersionStrategy.UNKNOWN:
@@ -713,51 +560,36 @@ class DockerInterface(CoreSysAttributes):
         available_version.sort(reverse=True)
         return available_version[0]
 
-    @process_lock
+    @Job(
+        name="docker_interface_run_inside",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
     def run_inside(self, command: str) -> Awaitable[CommandReturn]:
         """Execute a command inside Docker container."""
-        return self.sys_run_in_executor(self._run_inside, command)
+        return self.sys_run_in_executor(
+            self.sys_docker.container_run_inside, self.name, command
+        )
 
-    def _run_inside(self, command: str) -> CommandReturn:
-        """Execute a command inside Docker container.
-
-        Need run inside executor.
-        """
-        try:
-            docker_container = self.sys_docker.containers.get(self.name)
-        except docker.errors.NotFound:
-            raise DockerNotFound() from None
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError() from err
-
-        # Execute
-        try:
-            code, output = docker_container.exec_run(command)
-        except (docker.errors.DockerException, requests.RequestException) as err:
-            raise DockerError() from err
-
-        return CommandReturn(code, output)
-
-    def _validate_trust(
+    async def _validate_trust(
         self, image_id: str, image: str, version: AwesomeVersion
     ) -> None:
         """Validate trust of content."""
         checksum = image_id.partition(":")[2]
-        job = asyncio.run_coroutine_threadsafe(
-            self.sys_security.verify_own_content(checksum), self.sys_loop
-        )
-        job.result()
+        return await self.sys_security.verify_own_content(checksum)
 
-    @process_lock
-    def check_trust(self) -> Awaitable[None]:
+    @Job(
+        name="docker_interface_check_trust",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def check_trust(self) -> None:
         """Check trust of exists Docker image."""
-        return self.sys_run_in_executor(self._check_trust)
-
-    def _check_trust(self) -> None:
-        """Check trust of current image."""
         try:
-            image = self.sys_docker.images.get(f"{self.image}:{self.version!s}")
+            image = await self.sys_run_in_executor(
+                self.sys_docker.images.get, f"{self.image}:{self.version!s}"
+            )
         except (docker.errors.DockerException, requests.RequestException):
             return
 
-        self._validate_trust(image.id, self.image, self.version)
+        await self._validate_trust(image.id, self.image, self.version)

supervisor/docker/manager.py

@@ -11,8 +11,9 @@ 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 ImageCollection
+from docker.models.images import Image, ImageCollection
 from docker.models.networks import Network
 from docker.types.daemon import CancellableStream
 import requests
@@ -351,3 +352,224 @@ class DockerAPI:
 
             with suppress(docker_errors.DockerException, requests.RequestException):
                 network.disconnect(data.get("Name", cid), force=True)
+
+    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:
+            return False
+        except (DockerException, requests.RequestException) as err:
+            raise DockerError() from err
+
+        # Check the image is correct and state is good
+        return (
+            docker_container.image.id == docker_image.id
+            and docker_container.status in ("exited", "running", "created")
+        )
+
+    def stop_container(
+        self, name: str, timeout: int, remove_container: bool = True
+    ) -> None:
+        """Stop/remove 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
+
+        if docker_container.status == "running":
+            _LOGGER.info("Stopping %s application", name)
+            with suppress(DockerException, requests.RequestException):
+                docker_container.stop(timeout=timeout)
+
+        if remove_container:
+            with suppress(DockerException, requests.RequestException):
+                _LOGGER.info("Cleaning %s application", name)
+                docker_container.remove(force=True)
+
+    def start_container(self, name: str) -> None:
+        """Start Docker container."""
+        try:
+            docker_container: Container = self.containers.get(name)
+        except NotFound:
+            raise DockerNotFound(
+                f"{name} not found for starting up", _LOGGER.error
+            ) from None
+        except (DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Could not get {name} for starting up", _LOGGER.error
+            ) from err
+
+        _LOGGER.info("Starting %s", name)
+        try:
+            docker_container.start()
+        except (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
+
+        _LOGGER.info("Restarting %s", name)
+        try:
+            container.restart(timeout=timeout)
+        except (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
+
+        try:
+            return docker_container.logs(tail=tail, stdout=True, stderr=True)
+        except (DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Can't grep logs from {name}: {err}", _LOGGER.warning
+            ) from err
+
+    def container_stats(self, name: str) -> dict[str, Any]:
+        """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
+
+        # container is not running
+        if docker_container.status != "running":
+            raise DockerError(f"Container {name} is not running", _LOGGER.error)
+
+        try:
+            return docker_container.stats(stream=False)
+        except (DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Can't read stats from {name}: {err}", _LOGGER.error
+            ) from err
+
+    def container_run_inside(self, name: str, command: str) -> CommandReturn:
+        """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
+
+        # Execute
+        try:
+            code, output = docker_container.exec_run(command)
+        except (DockerException, requests.RequestException) as err:
+            raise DockerError() from err
+
+        return CommandReturn(code, output)
+
+    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)
+
+            _LOGGER.info("Removing image %s with %s", image, version)
+            with suppress(ImageNotFound):
+                self.images.remove(image=f"{image}:{version!s}", force=True)
+
+        except (DockerException, 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:
+        """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)
+
+            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:
+            raise DockerError(
+                f"Can't import image from tar: {err}", _LOGGER.error
+            ) from err
+
+    def export_image(self, image: str, version: AwesomeVersion, tar_file: Path) -> None:
+        """Export current images into a tar file."""
+        try:
+            image = self.api.get_image(f"{image}:{version}")
+        except (DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Can't fetch image {image}: {err}", _LOGGER.error
+            ) from err
+
+        _LOGGER.info("Export image %s to %s", image, tar_file)
+        try:
+            with tar_file.open("wb") as write_tar:
+                for chunk in image:
+                    write_tar.write(chunk)
+        except (OSError, requests.RequestException) as err:
+            raise DockerError(
+                f"Can't write tar file {tar_file}: {err}", _LOGGER.error
+            ) from err
+
+        _LOGGER.info("Export image %s done", image)
+
+    def cleanup_old_images(
+        self,
+        current_image: str,
+        current_version: AwesomeVersion,
+        old_images: set[str] | None = None,
+    ) -> None:
+        """Clean up old versions of an image."""
+        try:
+            current: Image = self.images.get(f"{current_image}:{current_version!s}")
+        except ImageNotFound:
+            raise DockerNotFound(
+                f"{current_image} not found for cleanup", _LOGGER.warning
+            ) from None
+        except (DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Can't get {current_image} for cleanup", _LOGGER.warning
+            ) from err
+
+        # Cleanup old and current
+        image_names = list(
+            old_images | {current_image} if old_images else {current_image}
+        )
+        try:
+            images_list = self.images.list(name=image_names)
+        except (DockerException, requests.RequestException) as err:
+            raise DockerError(
+                f"Corrupt docker overlayfs found: {err}", _LOGGER.warning
+            ) from err
+
+        for image in images_list:
+            if current.id == image.id:
+                continue
+
+            with suppress(DockerException, requests.RequestException):
+                _LOGGER.info("Cleanup images: %s", image.tags)
+                self.images.remove(image.id, force=True)

supervisor/docker/multicast.py

@@ -2,6 +2,9 @@
 import logging
 
 from ..coresys import CoreSysAttributes
+from ..exceptions import DockerJobError
+from ..jobs.const import JobExecutionLimit
+from ..jobs.decorator import Job
 from .const import ENV_TIME, Capabilities
 from .interface import DockerInterface
 
@@ -24,24 +27,18 @@ class DockerMulticast(DockerInterface, CoreSysAttributes):
         return MULTICAST_DOCKER_NAME
 
     @property
-    def capabilities(self) -> list[str]:
+    def capabilities(self) -> list[Capabilities]:
         """Generate needed capabilities."""
-        return [Capabilities.NET_ADMIN.value]
+        return [Capabilities.NET_ADMIN]
 
-    def _run(self) -> None:
-        """Run Docker image.
-
-        Need run inside executor.
-        """
-        if self._is_running():
-            return
-
-        # Cleanup
-        self._stop()
-
-        # Create & Run container
-        docker_container = self.sys_docker.run(
-            self.image,
+    @Job(
+        name="docker_multicast_run",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def run(self) -> None:
+        """Run Docker image."""
+        await self._run(
             tag=str(self.sys_plugins.multicast.version),
             init=False,
             name=self.name,
@@ -53,8 +50,6 @@ class DockerMulticast(DockerInterface, CoreSysAttributes):
             extra_hosts={"supervisor": self.sys_docker.network.supervisor},
             environment={ENV_TIME: self.sys_timezone},
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Multicast %s with version %s - Host", self.image, self.version
         )

supervisor/docker/network.py

@@ -21,7 +21,7 @@ class DockerNetwork:
     def __init__(self, docker_client: docker.DockerClient):
         """Initialize internal Supervisor network."""
         self.docker: docker.DockerClient = docker_client
-        self.network: docker.models.networks.Network = self._get_network()
+        self._network: docker.models.networks.Network = self._get_network()
 
     @property
     def name(self) -> str:
@@ -29,18 +29,14 @@ class DockerNetwork:
         return DOCKER_NETWORK
 
     @property
-    def containers(self) -> list[docker.models.containers.Container]:
-        """Return of connected containers from network."""
-        containers: list[docker.models.containers.Container] = []
-        for cid, _ in self.network.attrs.get("Containers", {}).items():
-            try:
-                containers.append(self.docker.containers.get(cid))
-            except docker.errors.NotFound:
-                _LOGGER.warning("Docker network is corrupt! %s", cid)
-            except (docker.errors.DockerException, requests.RequestException) as err:
-                _LOGGER.error("Unknown error with container lookup %s", err)
+    def network(self) -> docker.models.networks.Network:
+        """Return docker network."""
+        return self._network
 
-        return containers
+    @property
+    def containers(self) -> list[str]:
+        """Return of connected containers from network."""
+        return list(self.network.attrs.get("Containers", {}).keys())
 
     @property
     def gateway(self) -> IPv4Address:

supervisor/docker/observer.py

@@ -3,6 +3,9 @@ import logging
 
 from ..const import DOCKER_NETWORK_MASK
 from ..coresys import CoreSysAttributes
+from ..exceptions import DockerJobError
+from ..jobs.const import JobExecutionLimit
+from ..jobs.decorator import Job
 from .const import ENV_TIME, ENV_TOKEN, MOUNT_DOCKER, RestartPolicy
 from .interface import DockerInterface
 
@@ -25,20 +28,14 @@ class DockerObserver(DockerInterface, CoreSysAttributes):
         """Return name of Docker container."""
         return OBSERVER_DOCKER_NAME
 
-    def _run(self) -> None:
-        """Run Docker image.
-
-        Need run inside executor.
-        """
-        if self._is_running():
-            return
-
-        # Cleanup
-        self._stop()
-
-        # Create & Run container
-        docker_container = self.sys_docker.run(
-            self.image,
+    @Job(
+        name="docker_observer_run",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=DockerJobError,
+    )
+    async def run(self) -> None:
+        """Run Docker image."""
+        await self._run(
             tag=str(self.sys_plugins.observer.version),
             init=False,
             ipv4=self.sys_docker.network.observer,
@@ -46,7 +43,7 @@ class DockerObserver(DockerInterface, CoreSysAttributes):
             hostname=self.name.replace("_", "-"),
             detach=True,
             security_opt=self.security_opt,
-            restart_policy={"Name": RestartPolicy.ALWAYS.value},
+            restart_policy={"Name": RestartPolicy.ALWAYS},
             extra_hosts={"supervisor": self.sys_docker.network.supervisor},
             environment={
                 ENV_TIME: self.sys_timezone,
@@ -57,8 +54,6 @@ class DockerObserver(DockerInterface, CoreSysAttributes):
             ports={"80/tcp": 4357},
             oom_score_adj=-300,
         )
-
-        self._meta = docker_container.attrs
         _LOGGER.info(
             "Starting Observer %s with version %s - %s",
             self.image,

supervisor/docker/supervisor.py

@@ -8,15 +8,16 @@ from awesomeversion.awesomeversion import AwesomeVersion
 import docker
 import requests
 
-from ..coresys import CoreSysAttributes
 from ..exceptions import DockerError
+from ..jobs.const import JobExecutionLimit
+from ..jobs.decorator import Job
 from .const import PropagationMode
 from .interface import DockerInterface
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
 
-class DockerSupervisor(DockerInterface, CoreSysAttributes):
+class DockerSupervisor(DockerInterface):
     """Docker Supervisor wrapper for Supervisor."""
 
     @property
@@ -38,20 +39,20 @@ class DockerSupervisor(DockerInterface, CoreSysAttributes):
     def host_mounts_available(self) -> bool:
         """Return True if container can see mounts on host within its data directory."""
         return self._meta and any(
-            mount.get("Propagation") == PropagationMode.SLAVE.value
+            mount.get("Propagation") == PropagationMode.SLAVE
             for mount in self.meta_mounts
             if mount.get("Destination") == "/data"
         )
 
-    def _attach(
-        self, version: AwesomeVersion, skip_state_event_if_down: bool = False
+    @Job(name="docker_supervisor_attach", limit=JobExecutionLimit.GROUP_WAIT)
+    async def attach(
+        self, version: AwesomeVersion, *, skip_state_event_if_down: bool = False
     ) -> None:
-        """Attach to running docker container.
-
-        Need run inside executor.
-        """
+        """Attach to running docker container."""
         try:
-            docker_container = self.sys_docker.containers.get(self.name)
+            docker_container = await self.sys_run_in_executor(
+                self.sys_docker.containers.get, self.name
+            )
         except (docker.errors.DockerException, requests.RequestException) as err:
             raise DockerError() from err
 
@@ -63,17 +64,19 @@ class DockerSupervisor(DockerInterface, CoreSysAttributes):
         )
 
         # If already attach
-        if docker_container in self.sys_docker.network.containers:
+        if docker_container.id in self.sys_docker.network.containers:
             return
 
         # Attach to network
         _LOGGER.info("Connecting Supervisor to hassio-network")
-        self.sys_docker.network.attach_container(
+        await self.sys_run_in_executor(
+            self.sys_docker.network.attach_container,
             docker_container,
             alias=["supervisor"],
             ipv4=self.sys_docker.network.supervisor,
         )
 
+    @Job(name="docker_supervisor_retag", limit=JobExecutionLimit.GROUP_WAIT)
     def retag(self) -> Awaitable[None]:
         """Retag latest image to version."""
         return self.sys_run_in_executor(self._retag)
@@ -93,6 +96,7 @@ class DockerSupervisor(DockerInterface, CoreSysAttributes):
                 f"Can't retag Supervisor version: {err}", _LOGGER.error
             ) from err
 
+    @Job(name="docker_supervisor_update_start_tag", limit=JobExecutionLimit.GROUP_WAIT)
     def update_start_tag(self, image: str, version: AwesomeVersion) -> Awaitable[None]:
         """Update start tag to new version."""
         return self.sys_run_in_executor(self._update_start_tag, image, version)

supervisor/exceptions.py

@@ -36,6 +36,22 @@ class JobConditionException(JobException):
     """Exception happening for job conditions."""
 
 
+class JobStartException(JobException):
+    """Exception occurred starting a job on in current asyncio task."""
+
+
+class JobNotFound(JobException):
+    """Exception for job not found."""
+
+
+class JobInvalidUpdate(JobException):
+    """Exception for invalid update to a job."""
+
+
+class JobGroupExecutionLimitExceeded(JobException):
+    """Exception when job group execution limit exceeded."""
+
+
 # HomeAssistant
 
 
@@ -51,6 +67,10 @@ class HomeAssistantCrashError(HomeAssistantError):
     """Error on crash of a Home Assistant startup."""
 
 
+class HomeAssistantStartupTimeout(HomeAssistantCrashError):
+    """Timeout waiting for Home Assistant successful startup."""
+
+
 class HomeAssistantAPIError(HomeAssistantError):
     """Home Assistant API exception."""
 
@@ -315,6 +335,10 @@ class DBusNotConnectedError(HostNotSupportedError):
     """D-Bus is not connected and call a method."""
 
 
+class DBusServiceUnkownError(HassioNotSupportedError):
+    """D-Bus service was not available."""
+
+
 class DBusInterfaceError(HassioNotSupportedError):
     """D-Bus interface not connected."""
 
@@ -343,6 +367,10 @@ class DBusTimeoutError(DBusError):
     """D-Bus call timed out."""
 
 
+class DBusNoReplyError(DBusError):
+    """D-Bus remote didn't reply/disconnected."""
+
+
 class DBusFatalError(DBusError):
     """D-Bus call going wrong.
 
@@ -478,6 +506,10 @@ class DockerNotFound(DockerError):
     """Docker object don't Exists."""
 
 
+class DockerJobError(DockerError, JobException):
+    """Error executing docker job."""
+
+
 # Hardware
 
 
@@ -561,6 +593,14 @@ class HomeAssistantBackupError(BackupError, HomeAssistantError):
     """Raise if an error during Home Assistant Core backup is happening."""
 
 
+class BackupInvalidError(BackupError):
+    """Raise if backup or password provided is invalid."""
+
+
+class BackupJobError(BackupError, JobException):
+    """Raise on Backup job error."""
+
+
 # Security
 
 
@@ -593,3 +633,10 @@ class MountNotFound(MountError):
 
 class MountJobError(MountError, JobException):
     """Raise on Mount job error."""
+
+
+# Network
+
+
+class NetworkInterfaceNotFound(HassioError):
+    """Raise on network interface not found."""

supervisor/hardware/const.py

@@ -1,8 +1,8 @@
 """Constants for hardware."""
-from enum import Enum
+from enum import StrEnum
 
 
-class UdevSubsystem(str, Enum):
+class UdevSubsystem(StrEnum):
     """Udev subsystem class."""
 
     SERIAL = "tty"
@@ -24,7 +24,7 @@ class UdevSubsystem(str, Enum):
     RPI_H264MEM = "rpivid-h264mem"
 
 
-class PolicyGroup(str, Enum):
+class PolicyGroup(StrEnum):
     """Policy groups backend."""
 
     UART = "uart"
@@ -35,14 +35,14 @@ class PolicyGroup(str, Enum):
     BLUETOOTH = "bluetooth"
 
 
-class HardwareAction(str, Enum):
+class HardwareAction(StrEnum):
     """Hardware device action."""
 
     ADD = "add"
     REMOVE = "remove"
 
 
-class UdevKernelAction(str, Enum):
+class UdevKernelAction(StrEnum):
     """Udev kernel device action."""
 
     ADD = "add"

supervisor/hardware/helper.py

@@ -1,5 +1,5 @@
 """Read hardware info from system."""
-from datetime import datetime
+from datetime import UTC, datetime
 import logging
 from pathlib import Path
 import re
@@ -55,7 +55,7 @@ class HwHelper(CoreSysAttributes):
             _LOGGER.error("Can't found last boot time!")
             return None
 
-        return datetime.utcfromtimestamp(int(found.group(1)))
+        return datetime.fromtimestamp(int(found.group(1)), UTC)
 
     def hide_virtual_device(self, udev_device: pyudev.Device) -> bool:
         """Small helper to hide not needed Devices."""

supervisor/hardware/manager.py

@@ -94,7 +94,7 @@ class HardwareManager(CoreSysAttributes):
         udev_device: pyudev.Device = pyudev.Devices.from_sys_path(
             self._udev, str(device.sysfs)
         )
-        return udev_device.find_parent(subsystem.value) is not None
+        return udev_device.find_parent(subsystem) is not None
 
     def _import_devices(self) -> None:
         """Import fresh from udev database."""

supervisor/homeassistant/api.py

@@ -7,16 +7,19 @@ from typing import Any, AsyncContextManager
 
 import aiohttp
 from aiohttp import hdrs
+from awesomeversion import AwesomeVersion
 
 from ..coresys import CoreSys, CoreSysAttributes
 from ..exceptions import HomeAssistantAPIError, HomeAssistantAuthError
 from ..jobs.const import JobExecutionLimit
 from ..jobs.decorator import Job
-from ..utils import check_port
+from ..utils import check_port, version_is_new_enough
 from .const import LANDINGPAGE
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
+GET_CORE_STATE_MIN_VERSION: AwesomeVersion = AwesomeVersion("2023.8.0.dev20230720")
+
 
 class HomeAssistantAPI(CoreSysAttributes):
     """Home Assistant core object for handle it."""
@@ -29,7 +32,11 @@ class HomeAssistantAPI(CoreSysAttributes):
         self.access_token: str | None = None
         self._access_token_expires: datetime | None = None
 
-    @Job(limit=JobExecutionLimit.SINGLE_WAIT)
+    @Job(
+        name="home_assistant_api_ensure_access_token",
+        limit=JobExecutionLimit.SINGLE_WAIT,
+        internal=True,
+    )
     async def ensure_access_token(self) -> None:
         """Ensure there is an access token."""
         if (
@@ -100,42 +107,64 @@ class HomeAssistantAPI(CoreSysAttributes):
                         continue
                     yield resp
                     return
-            except (asyncio.TimeoutError, aiohttp.ClientError) as err:
+            except (TimeoutError, aiohttp.ClientError) as err:
                 _LOGGER.error("Error on call %s: %s", url, err)
                 break
 
         raise HomeAssistantAPIError()
 
-    async def get_config(self) -> dict[str, Any]:
-        """Return Home Assistant config."""
-        async with self.make_request("get", "api/config") as resp:
+    async def _get_json(self, path: str) -> dict[str, Any]:
+        """Return Home Assistant get API."""
+        async with self.make_request("get", path) as resp:
             if resp.status in (200, 201):
                 return await resp.json()
             else:
                 _LOGGER.debug("Home Assistant API return: %d", resp.status)
         raise HomeAssistantAPIError()
 
-    async def check_api_state(self) -> bool:
-        """Return True if Home Assistant up and running."""
+    async def get_config(self) -> dict[str, Any]:
+        """Return Home Assistant config."""
+        return await self._get_json("api/config")
+
+    async def get_core_state(self) -> dict[str, Any]:
+        """Return Home Assistant core state."""
+        return await self._get_json("api/core/state")
+
+    async def get_api_state(self) -> str | None:
+        """Return state of Home Assistant Core or None."""
         # Skip check on landingpage
         if (
             self.sys_homeassistant.version is None
             or self.sys_homeassistant.version == LANDINGPAGE
         ):
-            return False
+            return None
 
         # Check if port is up
-        if not await self.sys_run_in_executor(
-            check_port,
+        if not await check_port(
             self.sys_homeassistant.ip_address,
             self.sys_homeassistant.api_port,
         ):
-            return False
+            return None
 
         # Check if API is up
         with suppress(HomeAssistantAPIError):
-            data = await self.get_config()
+            # get_core_state is available since 2023.8.0 and preferred
+            # since it is significantly faster than get_config because
+            # it does not require serializing the entire config
+            if version_is_new_enough(
+                self.sys_homeassistant.version, GET_CORE_STATE_MIN_VERSION
+            ):
+                data = await self.get_core_state()
+            else:
+                data = await self.get_config()
             # Older versions of home assistant does not expose the state
-            if data and data.get("state", "RUNNING") == "RUNNING":
-                return True
+            if data:
+                return data.get("state", "RUNNING")
+
+        return None
+
+    async def check_api_state(self) -> bool:
+        """Return Home Assistant Core state if up."""
+        if state := await self.get_api_state():
+            return state == "RUNNING"
         return False

supervisor/homeassistant/const.py

@@ -1,6 +1,6 @@
 """Constants for homeassistant."""
 from datetime import timedelta
-from enum import Enum
+from enum import StrEnum
 
 from awesomeversion import AwesomeVersion
 
@@ -19,7 +19,7 @@ CLOSING_STATES = [
 ]
 
 
-class WSType(str, Enum):
+class WSType(StrEnum):
     """Websocket types."""
 
     AUTH = "auth"
@@ -28,12 +28,13 @@ class WSType(str, Enum):
     BACKUP_END = "backup/end"
 
 
-class WSEvent(str, Enum):
+class WSEvent(StrEnum):
     """Websocket events."""
 
     ADDON = "addon"
     HEALTH_CHANGED = "health_changed"
     ISSUE_CHANGED = "issue_changed"
     ISSUE_REMOVED = "issue_removed"
+    JOB = "job"
     SUPERVISOR_UPDATE = "supervisor_update"
     SUPPORTED_CHANGED = "supported_changed"

supervisor/homeassistant/core.py

@@ -2,16 +2,18 @@
 import asyncio
 from collections.abc import Awaitable
 from contextlib import suppress
+from dataclasses import dataclass
+from datetime import datetime, timedelta
 import logging
 import re
 import secrets
 import shutil
+from typing import Final
 
-import attr
 from awesomeversion import AwesomeVersion
 
 from ..const import ATTR_HOMEASSISTANT, BusEvent
-from ..coresys import CoreSys, CoreSysAttributes
+from ..coresys import CoreSys
 from ..docker.const import ContainerState
 from ..docker.homeassistant import DockerHomeAssistant
 from ..docker.monitor import DockerContainerStateEvent
@@ -21,12 +23,15 @@ from ..exceptions import (
     HomeAssistantCrashError,
     HomeAssistantError,
     HomeAssistantJobError,
+    HomeAssistantStartupTimeout,
     HomeAssistantUpdateError,
+    JobException,
 )
-from ..jobs.const import JobExecutionLimit
+from ..jobs.const import JOB_GROUP_HOME_ASSISTANT_CORE, JobExecutionLimit
 from ..jobs.decorator import Job, JobCondition
+from ..jobs.job_group import JobGroup
 from ..resolution.const import ContextType, IssueType
-from ..utils import convert_to_ascii, process_lock
+from ..utils import convert_to_ascii
 from ..utils.sentry import capture_exception
 from .const import (
     LANDINGPAGE,
@@ -38,25 +43,29 @@ from .const import (
 
 _LOGGER: logging.Logger = logging.getLogger(__name__)
 
+SECONDS_BETWEEN_API_CHECKS: Final[int] = 5
+# Core Stage 1 and some wiggle room
+STARTUP_API_RESPONSE_TIMEOUT: Final[timedelta] = timedelta(minutes=3)
+# All stages plus event start timeout and some wiggle rooom
+STARTUP_API_CHECK_RUNNING_TIMEOUT: Final[timedelta] = timedelta(minutes=15)
 RE_YAML_ERROR = re.compile(r"homeassistant\.util\.yaml")
 
 
-@attr.s(frozen=True)
+@dataclass
 class ConfigResult:
     """Return object from config check."""
 
-    valid = attr.ib()
-    log = attr.ib()
+    valid: bool
+    log: str
 
 
-class HomeAssistantCore(CoreSysAttributes):
+class HomeAssistantCore(JobGroup):
     """Home Assistant core object for handle it."""
 
     def __init__(self, coresys: CoreSys):
         """Initialize Home Assistant object."""
-        self.coresys: CoreSys = coresys
+        super().__init__(coresys, JOB_GROUP_HOME_ASSISTANT_CORE)
         self.instance: DockerHomeAssistant = DockerHomeAssistant(coresys)
-        self.lock: asyncio.Lock = asyncio.Lock()
         self._error_state: bool = False
 
     @property
@@ -95,9 +104,13 @@ class HomeAssistantCore(CoreSysAttributes):
         _LOGGER.info("Starting HomeAssistant landingpage")
         if not await self.instance.is_running():
             with suppress(HomeAssistantError):
-                await self._start()
+                await self.start()
 
-    @process_lock
+    @Job(
+        name="home_assistant_core_install_landing_page",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=HomeAssistantJobError,
+    )
     async def install_landingpage(self) -> None:
         """Install a landing page."""
         # Try to use a preinstalled landingpage
@@ -116,7 +129,7 @@ class HomeAssistantCore(CoreSysAttributes):
         while True:
             if not self.sys_updater.image_homeassistant:
                 _LOGGER.warning(
-                    "Found no information about Home Assistant. Retry in 30sec"
+                    "Found no information about Home Assistant. Retrying in 30sec"
                 )
                 await asyncio.sleep(30)
                 await self.sys_updater.reload()
@@ -127,19 +140,23 @@ class HomeAssistantCore(CoreSysAttributes):
                     LANDINGPAGE, image=self.sys_updater.image_homeassistant
                 )
                 break
-            except DockerError:
+            except (DockerError, JobException):
                 pass
             except Exception as err:  # pylint: disable=broad-except
                 capture_exception(err)
 
-            _LOGGER.warning("Fails install landingpage, retry after 30sec")
+            _LOGGER.warning("Failed to install landingpage, retrying after 30sec")
             await asyncio.sleep(30)
 
         self.sys_homeassistant.version = LANDINGPAGE
         self.sys_homeassistant.image = self.sys_updater.image_homeassistant
         self.sys_homeassistant.save_data()
 
-    @process_lock
+    @Job(
+        name="home_assistant_core_install",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=HomeAssistantJobError,
+    )
     async def install(self) -> None:
         """Install a landing page."""
         _LOGGER.info("Home Assistant setup")
@@ -155,12 +172,12 @@ class HomeAssistantCore(CoreSysAttributes):
                         image=self.sys_updater.image_homeassistant,
                     )
                     break
-                except DockerError:
+                except (DockerError, JobException):
                     pass
                 except Exception as err:  # pylint: disable=broad-except
                     capture_exception(err)
 
-            _LOGGER.warning("Error on Home Assistant installation. Retry in 30sec")
+            _LOGGER.warning("Error on Home Assistant installation. Retrying in 30sec")
             await asyncio.sleep(30)
 
         _LOGGER.info("Home Assistant docker now installed")
@@ -171,7 +188,7 @@ class HomeAssistantCore(CoreSysAttributes):
         # finishing
         try:
             _LOGGER.info("Starting Home Assistant")
-            await self._start()
+            await self.start()
         except HomeAssistantError:
             _LOGGER.error("Can't start Home Assistant!")
 
@@ -179,8 +196,8 @@ class HomeAssistantCore(CoreSysAttributes):
         with suppress(DockerError):
             await self.instance.cleanup()
 
-    @process_lock
     @Job(
+        name="home_assistant_core_update",
         conditions=[
             JobCondition.FREE_SPACE,
             JobCondition.HEALTHY,
@@ -188,6 +205,7 @@ class HomeAssistantCore(CoreSysAttributes):
             JobCondition.PLUGINS_UPDATED,
             JobCondition.SUPERVISOR_UPDATED,
         ],
+        limit=JobExecutionLimit.GROUP_ONCE,
         on_condition=HomeAssistantJobError,
     )
     async def update(
@@ -231,7 +249,7 @@ class HomeAssistantCore(CoreSysAttributes):
             self.sys_homeassistant.image = self.sys_updater.image_homeassistant
 
             if running:
-                await self._start()
+                await self.start()
             _LOGGER.info("Successfully started Home Assistant %s", to_version)
 
             # Successfull - last step
@@ -281,23 +299,11 @@ class HomeAssistantCore(CoreSysAttributes):
             self.sys_resolution.create_issue(IssueType.UPDATE_FAILED, ContextType.CORE)
             raise HomeAssistantUpdateError()
 
-    async def _start(self) -> None:
-        """Start Home Assistant Docker & wait."""
-        # Create new API token
-        self.sys_homeassistant.supervisor_token = secrets.token_hex(56)
-        self.sys_homeassistant.save_data()
-
-        # Write audio settings
-        self.sys_homeassistant.write_pulse()
-
-        try:
-            await self.instance.run()
-        except DockerError as err:
-            raise HomeAssistantError() from err
-
-        await self._block_till_run(self.sys_homeassistant.version)
-
-    @process_lock
+    @Job(
+        name="home_assistant_core_start",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=HomeAssistantJobError,
+    )
     async def start(self) -> None:
         """Run Home Assistant docker."""
         if await self.instance.is_running():
@@ -314,20 +320,37 @@ class HomeAssistantCore(CoreSysAttributes):
             await self._block_till_run(self.sys_homeassistant.version)
         # No Instance/Container found, extended start
         else:
-            await self._start()
+            # Create new API token
+            self.sys_homeassistant.supervisor_token = secrets.token_hex(56)
+            self.sys_homeassistant.save_data()
 
-    @process_lock
+            # Write audio settings
+            self.sys_homeassistant.write_pulse()
+
+            try:
+                await self.instance.run()
+            except DockerError as err:
+                raise HomeAssistantError() from err
+
+            await self._block_till_run(self.sys_homeassistant.version)
+
+    @Job(
+        name="home_assistant_core_stop",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=HomeAssistantJobError,
+    )
     async def stop(self) -> None:
-        """Stop Home Assistant Docker.
-
-        Return a coroutine.
-        """
+        """Stop Home Assistant Docker."""
         try:
             return await self.instance.stop(remove_container=False)
         except DockerError as err:
             raise HomeAssistantError() from err
 
-    @process_lock
+    @Job(
+        name="home_assistant_core_restart",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=HomeAssistantJobError,
+    )
     async def restart(self) -> None:
         """Restart Home Assistant Docker."""
         try:
@@ -337,12 +360,16 @@ class HomeAssistantCore(CoreSysAttributes):
 
         await self._block_till_run(self.sys_homeassistant.version)
 
-    @process_lock
+    @Job(
+        name="home_assistant_core_rebuild",
+        limit=JobExecutionLimit.GROUP_ONCE,
+        on_condition=HomeAssistantJobError,
+    )
     async def rebuild(self) -> None:
         """Rebuild Home Assistant Docker container."""
         with suppress(DockerError):
             await self.instance.stop()
-        await self._start()
+        await self.start()
 
     def logs(self) -> Awaitable[bytes]:
         """Get HomeAssistant docker logs.
@@ -359,10 +386,7 @@ class HomeAssistantCore(CoreSysAttributes):
         return self.instance.check_trust()
 
     async def stats(self) -> DockerStats:
-        """Return stats of Home Assistant.
-
-        Return a coroutine.
-        """
+        """Return stats of Home Assistant."""
         try:
             return await self.instance.stats()
         except DockerError as err:
@@ -385,13 +409,16 @@ class HomeAssistantCore(CoreSysAttributes):
     @property
     def in_progress(self) -> bool:
         """Return True if a task is in progress."""
-        return self.instance.in_progress or self.lock.locked()
+        return self.instance.in_progress or self.active_job
 
     async def check_config(self) -> ConfigResult:
         """Run Home Assistant config check."""
-        result = await self.instance.execute_command(
-            "python3 -m homeassistant -c /config --script check_config"
-        )
+        try:
+            result = await self.instance.execute_command(
+                "python3 -m homeassistant -c /config --script check_config"
+            )
+        except DockerError as err:
+            raise HomeAssistantError() from err
 
         # If not valid
         if result.exit_code is None:
@@ -416,28 +443,46 @@ class HomeAssistantCore(CoreSysAttributes):
             return
         _LOGGER.info("Wait until Home Assistant is ready")
 
-        while True:
-            await asyncio.sleep(5)
+        deadline = datetime.now() + STARTUP_API_RESPONSE_TIMEOUT
+        last_state = None
+        while not (timeout := datetime.now() >= deadline):
+            await asyncio.sleep(SECONDS_BETWEEN_API_CHECKS)
 
             # 1: Check if Container is is_running
             if not await self.instance.is_running():
                 _LOGGER.error("Home Assistant has crashed!")
                 break
 
-            # 2: Check if API response
-            if await self.sys_homeassistant.api.check_api_state():
-                _LOGGER.info("Detect a running Home Assistant instance")
-                self._error_state = False
-                return
+            # 2: Check API response
+            if state := await self.sys_homeassistant.api.get_api_state():
+                if last_state is None:
+                    # API initially available, move deadline up and check API
+                    # state to be running now
+                    deadline = datetime.now() + STARTUP_API_CHECK_RUNNING_TIMEOUT
+
+                if last_state != state:
+                    _LOGGER.info("Home Assistant Core state changed to %s", state)
+                    last_state = state
+
+                if state == "RUNNING":
+                    _LOGGER.info("Detect a running Home Assistant instance")
+                    self._error_state = False
+                    return
 
         self._error_state = True
+        if timeout:
+            raise HomeAssistantStartupTimeout(
+                "No Home Assistant Core response, assuming a fatal startup error",
+                _LOGGER.error,
+            )
         raise HomeAssistantCrashError()
 
     @Job(
+        name="home_assistant_core_repair",
         conditions=[
             JobCondition.FREE_SPACE,
             JobCondition.INTERNET_HOST,
-        ]
+        ],
     )
     async def repair(self):
         """Repair local Home Assistant data."""
@@ -459,6 +504,7 @@ class HomeAssistantCore(CoreSysAttributes):
             await self._restart_after_problem(event.state)
 
     @Job(
+        name="home_assistant_core_restart_after_problem",
         limit=JobExecutionLimit.THROTTLE_RATE_LIMIT,
         throttle_period=WATCHDOG_THROTTLE_PERIOD,
         throttle_max_calls=WATCHDOG_THROTTLE_MAX_CALLS,
@@ -470,7 +516,7 @@ class HomeAssistantCore(CoreSysAttributes):
             # Don't interrupt a task in progress or if rollback is handling it
             if not (self.in_progress or self.error_state):
                 _LOGGER.warning(
-                    "Watchdog found Home Assistant %s, restarting...", state.value
+                    "Watchdog found Home Assistant %s, restarting...", state
                 )
                 if state == ContainerState.FAILED and attempts == 0:
                     try: