cache/workarounds.md

Cache segment restore timeout

A cache gets downloaded in multiple segments of fixed sizes (1GB for a 32-bit runner and 2GB for a 64-bit runner). Sometimes, a segment download gets stuck which causes the workflow job to be stuck forever and fail. Version v3.0.8 of actions/cache introduces a segment download timeout. The segment download timeout will allow the segment download to get aborted and hence allow the job to proceed with a cache miss.

Default value of this timeout is 60 minutes and can be customized by specifying an environment variable named SEGMENT_DOWNLOAD_TIMEOUT_MINS with timeout value in minutes.

Update a cache

A cache today is immutable and cannot be updated. But some use cases require the cache to be saved even though there was a "hit" during restore. To do so, use a key which is unique for every run and use restore-keys to restore the nearest cache. For example:

    - name: update cache on every commit
      uses: actions/cache@v3
      with:
        path: prime-numbers
        key: primes-${{ runner.os }}-${{ github.run_id }} # Can use time based key as well
        restore-keys: |
          primes-${{ runner.os }}

Please note that this will create a new cache on every run and hence will consume the cache quota.

Use cache across feature branches

Reusing cache across feature branches is not allowed today to provide cache isolation. However if both feature branches are from the default branch, a good way to achieve this is to ensure that the default branch has a cache. This cache will then be consumable by both feature branches.

Improving cache restore performance on Windows/Using cross-os caching

Currently, cache restore is slow on Windows due to tar being inherently slow and the compression algorithm gzip in use. zstd is the default algorithm in use on linux and macos. It was disabled on Windows due to issues with bsd tar(libarchive), the tar implementation in use on Windows.

To improve cache restore performance, we can re-enable zstd as the compression algorithm using the following workaround. Add the following step to your workflow before the cache step:

    - if: ${{ runner.os == 'Windows' }}
      name: Use GNU tar
      shell: cmd
      run: |
        echo "Adding GNU tar to PATH"
        echo C:\Program Files\Git\usr\bin>>"%GITHUB_PATH%"

The cache action will use GNU tar instead of bsd tar on Windows. This should work on all Github Hosted runners as it is. For self-hosted runners, please ensure you have GNU tar and zstd installed.

The above workaround is also needed if you wish to use cross-os caching since difference of compression algorithms will result in different cache versions for the same cache key. So the above workaround will ensure zstd is used for caching on all platforms thus resulting in the same cache version for the same cache key.

Re-evaluating the cache key

By default, the string resolved from the key parameter for restoring the cache will also be used for saving the cache. In some situations it may be preferred to re-evaluate the key parameter when saving, such as when part of the key's name is not available until after the restore has completed. In such cases, you can add the option reevaluate-key: true.

For example, consider a a .NET solution that doesn't normally use packages.lock.json lock files. You may still want to generate a lock file for purposes of determining the cache key only.

  - name: Cache Packages
    uses: actions/cache@v3
    with:
      path: ~/.nuget/packages
      key: ${{ runner.os }}-nuget-${{ hashFiles('**/packages.lock.json') }}
      restore-keys: ${{ runner.os }}-nuget-
      reevaluate-key: true

  - name: Restore Packages
    run: dotnet restore --use-lock-file --nologo

  - name: Build
    run: dotnet build -c Release --no-restore --nologo

Assuming a Linux runner and no pre-existing packages.lock.json files:

• The first run will attempt to restore a cache named "Linux-nuget-", which won't exist.
• The lock files will be generated by dotnet restore due to the --use-lock-file option.
• Because reevaluate-key: true is set, the cache will be saved using a key based on the hash of the newly-generated lock files, such as "Linux-nuget-c2c3fe3d365e39a1be194115ccf5477bde7e17ef902f85ef5b6ef3ce1c6a08bc".
• The next run will restore the latest cache, due to the restore-keys option matching the requested prefix.
• The lock files will be generated again. If they have changed, then so will the cache key and the cache will be updated.