ci: apply WSL hanging workaround to MSI builder

This change refactors the WSL hanging workaround into a reusable GitHub composite action and applies it to the MSI builder workflow. Signed-off-by: Austin Vazquez <[email protected]>
runfinch · Jul 3, 2024 · dbd3b7c · dbd3b7c
1 parent 12b6934
commit dbd3b7c
Show file tree

Hide file tree

Showing 3 changed files with 62 additions and 86 deletions.
diff --git a/.github/actions/cleanup-windows-env/action.yml b/.github/actions/cleanup-windows-env/action.yml
@@ -0,0 +1,46 @@
+name: "Clean Up Windows Environment"
+description: "Reusable action to clean up the virtual machine for Finch on Windows"
+inputs:
+  deep-clean:
+    default: true
+    description: "Clean up Finch virtual machine and configuration files only"
+
+runs:
+  using: composite
+  steps:
+    - name: Remove Finch VM
+      timeout: 3 minutes
+      run: |
+          wsl --list --verbose
+          wsl --shutdown
+          wsl --unregister lima-finch
+          wsl --list --verbose
+
+    # This is a workaround for https://github.com/microsoft/WSL/issues/8529
+    #
+    # If WSL is suspected of hanging for longer than 180 seconds, then
+    # kill the WSL service and retry the shutdown command.
+    - name: Force remove Finch VM
+      needs: remove-vm
+      if: failure()
+      timeout: 3 minutes
+      run: |
+        $ErrorActionPreference = 'Ignore' 
+        taskkill /f /im wslservice.exe 2> nul || cmd /c "exit /b 0"
+        wsl --list --verbose
+        wsl --shutdown
+        Start-Sleep -s 10
+        wsl --unregister lima-finch
+        wsl --list --verbose
+
+    - name: Clean up Finch application data
+      run: |
+        Remove-Item C:\Users\Administrator\AppData\Local\.finch -Recurse -ErrorAction Ignore
+        exit 0 # Cleanup may set the exit code e.g. if a file doesn't exist; just ignore
+
+    - name: Clean up Finch configuration and artifacts
+      if: inputs.deep-clean
+      run: |
+        Remove-Item C:\Users\Administrator\.finch -Recurse -ErrorAction Ignore
+        make clean
+        exit 0 # Cleanup may set the exit code e.g. if a file doesn't exist; just ignore
diff --git a/.github/workflows/build-and-test-msi.yaml b/.github/workflows/build-and-test-msi.yaml
@@ -73,27 +73,14 @@ jobs:
           fetch-depth: 0
           persist-credentials: false
           submodules: recursive
+      - name: Ensure runner environment is in clean state
+        uses: ${{ github.workspace }}/.github/actions/cleanup-windows-env
       - name: configure aws credentials
         uses: aws-actions/configure-aws-credentials@e3dd6a429d7300a6a4c196c26e071d42e0343502 # v4.0.2
         with:
           role-to-assume: ${{ secrets.WINDOWS_ROLE }}
           role-session-name: windows-msi
           aws-region: ${{ secrets.WINDOWS_REGION }}
-      - name: Remove Finch VM
-        run: |
-          $ErrorActionPreference = 'Ignore'
-          taskkill /f /im wslservice.exe 2> nul || cmd /c "exit /b 0"
-          wsl --list --verbose
-          wsl --shutdown
-          wsl --unregister lima-finch
-          wsl --list --verbose
-      - name: Clean up previous files
-        run: |
-          takeown /F C:\actions-runner\_work\finch /R
-          Remove-Item C:\Users\Administrator\.finch -Recurse -ErrorAction Ignore
-          Remove-Item C:\Users\Administrator\AppData\Local\.finch -Recurse -ErrorAction Ignore
-          make clean
-          cd deps/finch-core && make clean
       - name: Build project
         run: |
           make FINCH_OS_IMAGE_LOCATION_ROOT=__INSTALLFOLDER__
@@ -143,20 +130,9 @@ jobs:
         run: |
           $tag="${{ needs.get-tag-name.outputs.tag }}"
           aws s3 cp "./msi-builder/build/signed/Finch-$tag.msi" "s3://${{ secrets.INSTALLER_PRIVATE_BUCKET_NAME }}/Finch-$tag.msi" --no-progress
-      - name: Remove Finch VM and Clean Up Previous Environment
+      - name: Clean up runner
         if: ${{ always() }}
-        run: |
-          # We want these cleanup commands to always run, ignore errors so the step completes.
-          $ErrorActionPreference = 'Ignore'
-          taskkill /f /im wslservice.exe 2> nul || cmd /c "exit /b 0"
-          wsl --list --verbose
-          wsl --shutdown
-          wsl --unregister lima-finch
-          wsl --list --verbose
-          Remove-Item C:\Users\Administrator\AppData\Local\.finch -Recurse
-          make clean
-          cd deps/finch-core && make clean
-          exit 0 # Cleanup may set the exit code e.g. if a file doesn't exist; just ignore
+        uses: ${{ github.workspace }}/.github/actions/cleanup-windows-env
 
   msi-e2e-tests:
     needs:
@@ -185,6 +161,8 @@ jobs:
           fetch-depth: 0
           persist-credentials: false
           submodules: recursive
+      - name: Ensure runner environment is in clean state
+        uses: ${{ github.workspace }}/.github/actions/cleanup-windows-env
       - name: Set output variables
         id: vars
         run: |
@@ -197,18 +175,6 @@ jobs:
           role-to-assume: ${{ secrets.ROLE }}
           role-session-name: msi-test
           aws-region: ${{ secrets.REGION }}
-      - name: Remove Finch VM
-        run: |
-          wsl --list --verbose
-          wsl --shutdown
-          wsl --unregister lima-finch
-          wsl --list --verbose
-      - name: Clean up previous files
-        run: |
-          Remove-Item C:\Users\Administrator\.finch -Recurse -ErrorAction Ignore
-          Remove-Item C:\Users\Administrator\AppData\Local\.finch -Recurse -ErrorAction Ignore
-          make clean
-          cd deps/finch-core && make clean
       - name: Uninstall Finch silently
         run: |
           $productCode = (Get-WmiObject -Class Win32_Product | Where-Object { $_.Name -like "*Finch*" } | Select-Object -ExpandProperty IdentifyingNumber)
@@ -239,14 +205,10 @@ jobs:
             git clean -f -d
             $env:INSTALLED="true"
             make test-e2e-vm
-      - name: Remove Finch VM
-        run: |
-          wsl --list --verbose
-          wsl --shutdown
-          Start-Sleep -s 10
-          wsl --unregister lima-finch
-          Remove-Item C:\Users\Administrator\AppData\Local\.finch -Recurse
-          wsl --list --verbose    
+      - name: Remove Finch VM and application data
+        uses: ${{ github.workspace }}/.github/actions/cleanup-windows-env
+        with:
+          deep-clean: false
       - name: Run container e2e tests
         uses: nick-fields/retry@7152eba30c6575329ac0576536151aca5a72780e # v3.0.0
         with:
@@ -270,17 +232,6 @@ jobs:
           } else {
               Write-Output "Finch not found or it wasn't installed using MSI."
           }
-      - name: Remove Finch VM and Clean Up Previous Environment
+      - name: Clean up runner
         if: ${{ always() }}
-        run: |
-          # We want these cleanup commands to always run, ignore errors so the step completes.
-          $ErrorActionPreference = 'Ignore'
-          taskkill /f /im wslservice.exe 2> nul || cmd /c "exit /b 0"
-          wsl --list --verbose
-          wsl --shutdown
-          wsl --unregister lima-finch
-          wsl --list --verbose
-          Remove-Item C:\Users\Administrator\AppData\Local\.finch -Recurse
-          make clean
-          cd deps/finch-core && make clean
-          exit 0 # Cleanup may set the exit code e.g. if a file doesn't exist; just ignore
+        uses: ${{ github.workspace }}/.github/actions/cleanup-windows-env
diff --git a/.github/workflows/ci.yaml b/.github/workflows/ci.yaml
@@ -225,6 +225,8 @@ jobs:
           fetch-depth: 0
           persist-credentials: false
           submodules: recursive
+      - name: Ensure runner environment is in clean state
+        uses: ${{ github.workspace }}/.github/actions/cleanup-windows-env
       - name: Set output variables
         id: vars
         run: |
@@ -238,18 +240,6 @@ jobs:
           role-to-assume: ${{ secrets.ROLE }}
           role-session-name: credhelper-test
           aws-region: ${{ secrets.REGION }}
-      - name: Remove Finch VM
-        run: |
-          wsl --list --verbose
-          wsl --shutdown
-          wsl --unregister lima-finch
-          wsl --list --verbose
-      - name: Clean up previous files
-        run: |
-          Remove-Item C:\Users\Administrator\.finch -Recurse -ErrorAction Ignore
-          Remove-Item C:\Users\Administrator\AppData\Local\.finch -Recurse -ErrorAction Ignore
-          make clean
-          cd deps/finch-core && make clean
       - name: Build project
         run: |
           git status
@@ -266,20 +256,9 @@ jobs:
           git status
           git clean -f -d
           make ${{ matrix.test-command }}
-      - name: Remove Finch VM and Clean Up Previous Environment
+      - name: Clean up runner
         if: ${{ always() }}
-        run: |
-          # We want these cleanup commands to always run, ignore errors so the step completes.
-          $ErrorActionPreference = 'Ignore'
-          taskkill /f /im wslservice.exe 2> nul || cmd /c "exit /b 0"
-          wsl --list --verbose
-          wsl --shutdown
-          wsl --unregister lima-finch
-          wsl --list --verbose
-          Remove-Item C:\Users\Administrator\AppData\Local\.finch -Recurse
-          make clean
-          cd deps/finch-core && make clean
-          exit 0 # Cleanup may set the exit code e.g. if a file doesn't exist; just ignore
+        uses: ${{ github.workspace }}/.github/actions/cleanup-windows-env
   mdlint:
     runs-on: ubuntu-latest
     steps: