Merge pull request #10 from Novaloop-AG/fix/ci-runners-and-permissions

Move releases to the persistent self-hosted runners and fix CI checkout

.github/workflows/build.yml

@@ -7,18 +7,32 @@ on:
   push:
     branches: [ develop ]
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 # Enrich gradle.properties for CI/CD
 env:
-  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx8g -XX:MaxMetaspaceSize=512m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx4g
+  # Reduced from upstream's -Xmx8g / 4 GB Kotlin daemon for the self-hosted runners.
+  # Each incus container reports 48 GiB and 8 cores, so a single build has ample room —
+  # but the containers are oversubscribed against their host, and two concurrent heavy
+  # Gradle jobs get their daemons SIGKILLed ("the daemon has disappeared", no
+  # OutOfMemoryError, no heap dump). Running the same job alone succeeds. Lower heaps
+  # make two builds likelier to coexist; the real fix is sizing the containers to the
+  # host. See the "Report runner capacity" step.
+  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx6g -XX:MaxMetaspaceSize=768m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx2g
   CI_GRADLE_ARG_PROPERTIES: --stacktrace -Dsonar.gradle.skipCompile=true --no-configuration-cache
 
 jobs:
   build:
     name: Build APKs
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     permissions:
+      # A job-level block replaces the workflow-level one, so contents:read has to be
+      # repeated here or checkout fails on this private fork.
+      contents: read
       # For NejcZdovc/comment-pr
       pull-requests: write
     strategy:
@@ -31,6 +45,9 @@ jobs:
       cancel-in-progress: true
     steps:
       - name: Free Disk Space (Ubuntu)
+        # Only meaningful on GitHub's throwaway VMs. On a persistent self-hosted runner this
+        # would delete dotnet/haskell/docker images from the host on every run.
+        if: runner.environment == 'github-hosted'
         uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be
         with:
           # This might remove tools that are actually needed, if set to "true" but frees about 6 GB

.github/workflows/build_enterprise.yml

@@ -7,17 +7,28 @@ on:
   push:
     branches: [ develop ]
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 # Enrich gradle.properties for CI/CD
 env:
-  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx8g -XX:MaxMetaspaceSize=512m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx4g
+  # Reduced from upstream's -Xmx8g / 4 GB Kotlin daemon for the self-hosted runners.
+  # Each incus container reports 48 GiB and 8 cores, so a single build has ample room —
+  # but the containers are oversubscribed against their host, and two concurrent heavy
+  # Gradle jobs get their daemons SIGKILLed ("the daemon has disappeared", no
+  # OutOfMemoryError, no heap dump). Running the same job alone succeeds. Lower heaps
+  # make two builds likelier to coexist; the real fix is sizing the containers to the
+  # host. See the "Report runner capacity" step.
+  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx6g -XX:MaxMetaspaceSize=768m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx2g
   CI_GRADLE_ARG_PROPERTIES: --stacktrace -Dsonar.gradle.skipCompile=true --no-configuration-cache
 
 jobs:
   build:
     name: Build Enterprise APKs
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     # Skip in forks
     if: ${{ github.event_name != 'pull_request' || github.event.pull_request.head.repo.full_name == 'element-hq/element-x-android' }}
     strategy:
@@ -30,6 +41,9 @@ jobs:
       cancel-in-progress: true
     steps:
       - name: Free Disk Space (Ubuntu)
+        # Only meaningful on GitHub's throwaway VMs. On a persistent self-hosted runner this
+        # would delete dotnet/haskell/docker images from the host on every run.
+        if: runner.environment == 'github-hosted'
         uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be
         with:
           # This might remove tools that are actually needed, if set to "true" but frees about 6 GB

.github/workflows/danger.yml

@@ -2,7 +2,11 @@ name: Danger CI
 
 on: [pull_request, merge_group]
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 jobs:
   build:

.github/workflows/healthchat-release-playstore.yml

@@ -46,17 +46,24 @@ concurrency:
 
 env:
   # Upstream's workflows hardcode -Xmx8g / Kotlin daemon -Xmx4g, sized for the
-  # 4-core/16GB ubuntu-latest that public repos get. On this container the
+  # 4-core/16GB ubuntu-latest that public repos get. On the previous incus
-  # daemon keeps getting SIGKILLed mid-build (DaemonDisappearedException, no
+  # container the daemon kept getting SIGKILLed mid-build
-  # OutOfMemoryError, no heap dump), so stay well under upstream's ceiling and
+  # (DaemonDisappearedException, no OutOfMemoryError, no heap dump), so these
-  # leave room for the Kotlin daemon and the parallel lint workers.
+  # values stay well under upstream's ceiling and leave room for the Kotlin
+  # daemon and the parallel lint workers. Re-check them against the "Report
+  # runner capacity" output now that jobs land on the persistent org runners,
+  # whose specs may differ.
   GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx6g -XX:MaxMetaspaceSize=768m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx2g
   CI_GRADLE_ARG_PROPERTIES: --stacktrace --no-daemon -Dsonar.gradle.skipCompile=true --no-configuration-cache
 
 jobs:
   release:
     name: Build and upload AAB to Play Store
-    runs-on: novaloop-incus-large
+    # Persistent org runners (github-runner-01, gha-runner-1..3). The previous
+    # novaloop-incus-large label pointed at the GARM ephemeral pool, whose freshly
+    # spawned runners were reaped mid-job — jobs died ~10 minutes in with
+    # "The operation was canceled" and no other diagnostic.
+    runs-on: [self-hosted, Linux, X64]
     steps:
       - uses: actions/checkout@v4
         with:
@@ -176,18 +183,14 @@ jobs:
           distribution: 'temurin'
           java-version: '21'
 
-      # The self-hosted runner image has no Android SDK (unlike GitHub-hosted
+      # The Android SDK now comes from the self-hosted runner image, so there is no
-      # ubuntu-latest), so install it here.
+      # setup-android step here. The image must provide it AND export ANDROID_HOME to
-      # - platform: follows COMPILE_SDK in plugins/src/main/kotlin/Versions.kt.
+      # the job environment — Gradle fails with "SDK location not found" otherwise.
-      #   Platform packages are minor-versioned, so compileSdk 37 is android-37.0.
+      # The image also has to carry the platform matching COMPILE_SDK in
-      # - build-tools: nothing sets buildToolsVersion, so AGP picks its own
+      # plugins/src/main/kotlin/Versions.kt (platform packages are minor-versioned, so
-      #   default (SdkConstants.CURRENT_BUILD_TOOLS_VERSION). Bump it when AGP is
+      # compileSdk 37 means android-37.0) and the build-tools version AGP defaults to
-      #   upgraded, not when Versions.kt changes.
+      # (SdkConstants.CURRENT_BUILD_TOOLS_VERSION), since nothing sets buildToolsVersion
-      - name: Set up Android SDK
+      # here. Both need refreshing in the image when AGP or COMPILE_SDK moves.
-        uses: android-actions/setup-android@v3
-        with:
-          packages: 'platforms;android-37.0 build-tools;36.0.0'
-
       - name: Configure gradle
         uses: gradle/actions/setup-gradle@v4
 

.github/workflows/maestro-local.yml

@@ -5,11 +5,22 @@ on:
   workflow_dispatch:
   pull_request:
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 # Enrich gradle.properties for CI/CD
 env:
-  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx8g -XX:MaxMetaspaceSize=512m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx4g
+  # Reduced from upstream's -Xmx8g / 4 GB Kotlin daemon for the self-hosted runners.
+  # Each incus container reports 48 GiB and 8 cores, so a single build has ample room —
+  # but the containers are oversubscribed against their host, and two concurrent heavy
+  # Gradle jobs get their daemons SIGKILLed ("the daemon has disappeared", no
+  # OutOfMemoryError, no heap dump). Running the same job alone succeeds. Lower heaps
+  # make two builds likelier to coexist; the real fix is sizing the containers to the
+  # host. See the "Report runner capacity" step.
+  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx6g -XX:MaxMetaspaceSize=768m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx2g
   CI_GRADLE_ARG_PROPERTIES: --stacktrace --no-daemon -Dsonar.gradle.skipCompile=true --no-configuration-cache
   ARCH: x86_64
   DEVICE: pixel_7_pro
@@ -19,12 +30,15 @@ env:
 jobs:
   build-apk:
     name: Build APK
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     concurrency:
       group: ${{ format('maestro-build-{0}', github.ref) }}
       cancel-in-progress: true
     steps:
       - name: Free Disk Space (Ubuntu)
+        # Only meaningful on GitHub's throwaway VMs. On a persistent self-hosted runner this
+        # would delete dotnet/haskell/docker images from the host on every run.
+        if: runner.environment == 'github-hosted'
         uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be
         with:
           # This might remove tools that are actually needed, if set to "true" but frees about 6 GB

.github/workflows/pull_request.yml

@@ -23,33 +23,38 @@ jobs:
           script: |
             core.setFailed("PR has been labeled with X-Blocked; it cannot be merged.");
 
-  community-prs:
+  # Disabled on the HealthChat fork. This job asks whether the PR author belongs to the
-    name: Label Community PRs
+  # element-hq "Vector Core" team, using secrets.ELEMENT_BOT_TOKEN_READ_ORG — a secret
-    runs-on: ubuntu-latest
+  # that only exists upstream. Here it fails on every PR that is opened, and there is no
-    if: github.event.action == 'opened'
+  # community to label anyway. Kept commented rather than deleted so upstream changes to
-    permissions:
+  # it still conflict visibly on merge.
-      pull-requests: write
+  # community-prs:
-    steps:
+  #   name: Label Community PRs
-      - name: Check membership
+  #   runs-on: ubuntu-latest
-        if: github.event.pull_request.user.login != 'renovate[bot]'
+  #   if: github.event.action == 'opened'
-        uses: tspascoal/get-user-teams-membership@b2546c5affc730fd8e3d8483ae9ad3621938c2f9 # v4.0.2
+  #   permissions:
-        id: teams
+  #     pull-requests: write
-        with:
+  #   steps:
-          username: ${{ github.event.pull_request.user.login }}
+  #     - name: Check membership
-          organization: element-hq
+  #       if: github.event.pull_request.user.login != 'renovate[bot]'
-          team: Vector Core
+  #       uses: tspascoal/get-user-teams-membership@b2546c5affc730fd8e3d8483ae9ad3621938c2f9 # v4.0.2
-          GITHUB_TOKEN: ${{ secrets.ELEMENT_BOT_TOKEN_READ_ORG }}
+  #       id: teams
-      - name: Add label
+  #       with:
-        if: steps.teams.outputs.isTeamMember == 'false'
+  #         username: ${{ github.event.pull_request.user.login }}
-        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
+  #         organization: element-hq
-        with:
+  #         team: Vector Core
-          script: |
+  #         GITHUB_TOKEN: ${{ secrets.ELEMENT_BOT_TOKEN_READ_ORG }}
-            github.rest.issues.addLabels({
+  #     - name: Add label
-              issue_number: context.issue.number,
+  #       if: steps.teams.outputs.isTeamMember == 'false'
-              owner: context.repo.owner,
+  #       uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
-              repo: context.repo.repo,
+  #       with:
-              labels: ['Z-Community-PR']
+  #         script: |
-            });
+  #           github.rest.issues.addLabels({
+  #             issue_number: context.issue.number,
+  #             owner: context.repo.owner,
+  #             repo: context.repo.repo,
+  #             labels: ['Z-Community-PR']
+  #           });
 
   close-if-fork-develop:
     name: Forbid develop branch fork contributions

.github/workflows/quality.yml

@@ -7,11 +7,22 @@ on:
   push:
     branches: [ main, develop ]
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 # Enrich gradle.properties for CI/CD
 env:
-  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx8g -XX:MaxMetaspaceSize=512m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx4g
+  # Reduced from upstream's -Xmx8g / 4 GB Kotlin daemon for the self-hosted runners.
+  # Each incus container reports 48 GiB and 8 cores, so a single build has ample room —
+  # but the containers are oversubscribed against their host, and two concurrent heavy
+  # Gradle jobs get their daemons SIGKILLed ("the daemon has disappeared", no
+  # OutOfMemoryError, no heap dump). Running the same job alone succeeds. Lower heaps
+  # make two builds likelier to coexist; the real fix is sizing the containers to the
+  # host. See the "Report runner capacity" step.
+  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx6g -XX:MaxMetaspaceSize=768m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx2g
   CI_GRADLE_ARG_PROPERTIES: --stacktrace --no-daemon -Dsonar.gradle.skipCompile=true --no-configuration-cache
 
 jobs:
@@ -20,6 +31,9 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Free Disk Space (Ubuntu)
+        # Only meaningful on GitHub's throwaway VMs. On a persistent self-hosted runner this
+        # would delete dotnet/haskell/docker images from the host on every run.
+        if: runner.environment == 'github-hosted'
         uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be
         with:
           # This might remove tools that are actually needed, if set to "true" but frees about 6 GB
@@ -63,7 +77,7 @@ jobs:
 
   checkDependencies:
     name: Search for invalid dependencies
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     steps:
       - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
         with:
@@ -87,7 +101,7 @@ jobs:
   # Code checks
   konsist:
     name: Konsist tests
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     # Allow all jobs on main and develop. Just one per PR.
     concurrency:
       group: ${{ github.ref == 'refs/heads/main' && format('check-konsist-main-{0}', github.sha) || github.ref == 'refs/heads/develop' && format('check-konsist-develop-{0}', github.sha) || format('check-konsist-{0}', github.ref) }}
@@ -128,12 +142,21 @@ jobs:
 
   compose:
     name: Compose tests
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     # Allow all jobs on main and develop. Just one per PR.
     concurrency:
       group: ${{ github.ref == 'refs/heads/main' && format('check-compose-main-{0}', github.sha) || github.ref == 'refs/heads/develop' && format('check-compose-develop-{0}', github.sha) || format('check-compose-{0}', github.ref) }}
       cancel-in-progress: true
     steps:
+      # Records the container's real limits so heap settings can be read against them
+      # instead of guessed at. Mirrors healthchat-release-playstore.yml.
+      - name: Report runner capacity
+        if: runner.environment == 'self-hosted'
+        run: |
+          free -h; nproc; df -h .
+          cat /sys/fs/cgroup/memory.max 2>/dev/null \
+            || cat /sys/fs/cgroup/memory/memory.limit_in_bytes 2>/dev/null \
+            || echo "no cgroup memory limit visible"
       - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
         with:
           # Ensure we are building the branch and not the branch after being merged on develop
@@ -162,7 +185,7 @@ jobs:
 
   lint:
     name: Android lint check
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     # Allow all jobs on main and develop. Just one per PR.
     concurrency:
       group: ${{ github.ref == 'refs/heads/main' && format('check-lint-main-{0}', github.sha) || github.ref == 'refs/heads/develop' && format('check-lint-develop-{0}', github.sha) || format('check-lint-{0}', github.ref) }}
@@ -207,7 +230,7 @@ jobs:
 
   detekt:
     name: Detekt checks
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     # Allow all jobs on main and develop. Just one per PR.
     concurrency:
       group: ${{ github.ref == 'refs/heads/main' && format('check-detekt-main-{0}', github.sha) || github.ref == 'refs/heads/develop' && format('check-detekt-develop-{0}', github.sha) || format('check-detekt-{0}', github.ref) }}
@@ -248,7 +271,7 @@ jobs:
 
   ktlint:
     name: Ktlint checks
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     # Allow all jobs on main and develop. Just one per PR.
     concurrency:
       group: ${{ github.ref == 'refs/heads/main' && format('check-ktlint-main-{0}', github.sha) || github.ref == 'refs/heads/develop' && format('check-ktlint-develop-{0}', github.sha) || format('check-ktlint-{0}', github.ref) }}
@@ -331,6 +354,10 @@ jobs:
     name: Run zizmor
     runs-on: ubuntu-latest
     permissions:
+      # A job-level block replaces the workflow-level one, so contents:read has to be
+      # repeated here or checkout fails on this private fork.
+      contents: read
+      actions: read # Required for zizmor to read workflow files.
       security-events: write # Required for upload-sarif (used by zizmor-action) to upload SARIF files.
     steps:
       - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0

.github/workflows/recordScreenshots.yml

@@ -5,11 +5,22 @@ on:
   pull_request:
     types: [ labeled ]
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 # Enrich gradle.properties for CI/CD
 env:
-  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx8g -XX:MaxMetaspaceSize=512m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx4g -Dsonar.gradle.skipCompile=true
+  # Reduced from upstream's -Xmx8g / 4 GB Kotlin daemon for the self-hosted runners.
+  # Each incus container reports 48 GiB and 8 cores, so a single build has ample room —
+  # but the containers are oversubscribed against their host, and two concurrent heavy
+  # Gradle jobs get their daemons SIGKILLed ("the daemon has disappeared", no
+  # OutOfMemoryError, no heap dump). Running the same job alone succeeds. Lower heaps
+  # make two builds likelier to coexist; the real fix is sizing the containers to the
+  # host. See the "Report runner capacity" step.
+  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx6g -XX:MaxMetaspaceSize=768m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx2g -Dsonar.gradle.skipCompile=true
   CI_GRADLE_ARG_PROPERTIES: --no-configuration-cache
 
 jobs:
@@ -19,11 +30,14 @@ jobs:
       pull-requests: write
       contents: write
     name: Record screenshots on branch ${{ github.event.pull_request.head.ref || github.ref_name }}
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     if: github.event_name == 'workflow_dispatch' || github.event.label.name == 'Record-Screenshots'
 
     steps:
       - name: Free Disk Space (Ubuntu)
+        # Only meaningful on GitHub's throwaway VMs. On a persistent self-hosted runner this
+        # would delete dotnet/haskell/docker images from the host on every run.
+        if: runner.environment == 'github-hosted'
         uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be
         with:
           # This might remove tools that are actually needed, if set to "true" but frees about 6 GB

.github/workflows/sonar.yml

@@ -7,24 +7,38 @@ on:
   push:
     branches: [ main, develop ]
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 # Enrich gradle.properties for CI/CD
 env:
-  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx8g -XX:MaxMetaspaceSize=512m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx4g
+  # Reduced from upstream's -Xmx8g / 4 GB Kotlin daemon for the self-hosted runners.
+  # Each incus container reports 48 GiB and 8 cores, so a single build has ample room —
+  # but the containers are oversubscribed against their host, and two concurrent heavy
+  # Gradle jobs get their daemons SIGKILLed ("the daemon has disappeared", no
+  # OutOfMemoryError, no heap dump). Running the same job alone succeeds. Lower heaps
+  # make two builds likelier to coexist; the real fix is sizing the containers to the
+  # host. See the "Report runner capacity" step.
+  GRADLE_OPTS: -Dorg.gradle.jvmargs=-Xmx6g -XX:MaxMetaspaceSize=768m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError -XX:+UseG1GC -Dkotlin.daemon.jvm.options=-Xmx2g
   CI_GRADLE_ARG_PROPERTIES: --stacktrace --warn -Dsonar.gradle.skipCompile=true --no-configuration-cache
   GROUP: ${{ format('sonar-{0}', github.ref) }}
 
 jobs:
   sonar:
     name: Sonar Quality Checks
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
     # Allow all jobs on main and develop. Just one per PR.
     concurrency:
       group: ${{ format('sonar-{0}', github.ref) }}
       cancel-in-progress: ${{ github.ref != 'refs/heads/main' && github.ref != 'refs/heads/develop' }}
     steps:
       - name: Free Disk Space (Ubuntu)
+        # Only meaningful on GitHub's throwaway VMs. On a persistent self-hosted runner this
+        # would delete dotnet/haskell/docker images from the host on every run.
+        if: runner.environment == 'github-hosted'
         uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be
         with:
           # This might remove tools that are actually needed, if set to "true" but frees about 6 GB

.github/workflows/tests.yml

@@ -7,24 +7,53 @@ on:
   push:
     branches: [ main, develop ]
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 # Enrich gradle.properties for CI/CD
 env:
-  GRADLE_OPTS: -Dorg.gradle.jvmargs="-Xmx7g -XX:MaxMetaspaceSize=512m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError" -Dkotlin.daemon.jvm.options=-Xmx2g -XX:+UseG1GC
+  # Reduced from upstream's -Xmx8g / 4 GB Kotlin daemon for the self-hosted runners.
-  CI_GRADLE_ARG_PROPERTIES: --stacktrace --no-daemon -Dsonar.gradle.skipCompile=true --no-configuration-cache
+  # Each incus container reports 48 GiB and 8 cores, so a single build has ample room —
+  # but the containers are oversubscribed against their host, and two concurrent heavy
+  # Gradle jobs get their daemons SIGKILLed ("the daemon has disappeared", no
+  # OutOfMemoryError, no heap dump). Running the same job alone succeeds. Lower heaps
+  # make two builds likelier to coexist; the real fix is sizing the containers to the
+  # host. See the "Report runner capacity" step.
+  GRADLE_OPTS: -Dorg.gradle.jvmargs="-Xmx6g -XX:MaxMetaspaceSize=768m -Dfile.encoding=UTF-8 -XX:+HeapDumpOnOutOfMemoryError" -Dkotlin.daemon.jvm.options=-Xmx2g -XX:+UseG1GC
+  # --max-workers caps how many test tasks run at once. build.gradle.kts sets
+  # maxParallelForks = cores/2 and screenshot tests fork at maxHeapSize=2g, so on these
+  # 8-core containers the default (workers = cores = 8) allows 8 x 4 x 2g = 64 GB of test
+  # JVMs against a 48 GiB cgroup limit, with no swap — the daemon gets OOM-killed and
+  # Gradle reports "the daemon has disappeared". GitHub's 4-core hosted runners reach only
+  # 2 x 2 x 2g = 16 GB and additionally have the 8 GB swapfile added above.
+  CI_GRADLE_ARG_PROPERTIES: --stacktrace --no-daemon -Dsonar.gradle.skipCompile=true --no-configuration-cache --max-workers=4
 
 jobs:
   tests:
     name: Runs unit tests
-    runs-on: ubuntu-latest
+    runs-on: [self-hosted, Linux, X64]
 
     # Allow all jobs on main and develop. Just one per PR.
     concurrency:
       group: ${{ github.ref == 'refs/heads/main' && format('unit-tests-main-{0}', github.sha) || github.ref == 'refs/heads/develop' && format('unit-tests-develop-{0}', github.sha) || format('unit-tests-{0}', github.ref) }}
       cancel-in-progress: true
     steps:
+      # Records the container's real limits so heap settings can be read against them
+      # instead of guessed at. Mirrors healthchat-release-playstore.yml.
+      - name: Report runner capacity
+        if: runner.environment == 'self-hosted'
+        run: |
+          free -h; nproc; df -h .
+          cat /sys/fs/cgroup/memory.max 2>/dev/null \
+            || cat /sys/fs/cgroup/memory/memory.limit_in_bytes 2>/dev/null \
+            || echo "no cgroup memory limit visible"
       - name: Free Disk Space (Ubuntu)
+        # Only meaningful on GitHub's throwaway VMs. On a persistent self-hosted runner this
+        # would delete dotnet/haskell/docker images from the host on every run.
+        if: runner.environment == 'github-hosted'
         uses: jlumbroso/free-disk-space@54081f138730dfa15788a46383842cd2f914a1be
         with:
           # This might remove tools that are actually needed, if set to "true" but frees about 6 GB
@@ -41,6 +70,9 @@ jobs:
       # Increase swapfile size to prevent screenshot tests getting terminated
       # https://github.com/actions/runner-images/discussions/7188#discussioncomment-6750749
       - name: 💽 Increase swapfile size
+        # /mnt/swapfile is a GitHub runner-image path and this needs passwordless sudo,
+        # so it only applies to the hosted VMs.
+        if: runner.environment == 'github-hosted'
         run: |
           sudo swapoff -a
           sudo fallocate -l 8G /mnt/swapfile

.github/workflows/validate-lfs.yml

@@ -2,7 +2,11 @@ name: Validate Git LFS
 
 on: [pull_request, merge_group]
 
-permissions: {}
+permissions:
+  # Private fork: actions/checkout needs contents:read to clone. Upstream is public,
+  # where an empty token can still read the repo, so an empty permissions block works there and
+  # fails here with `fatal: repository not found`.
+  contents: read
 
 jobs:
   build: