Skip to content

Latest commit

 

History

History
154 lines (115 loc) · 6.17 KB

TROUBLESHOOTING.md

File metadata and controls

154 lines (115 loc) · 6.17 KB
Raw

Troubleshooting

Permission denied

When troubleshooting "permission denied" errors from auth for Workload Identity, the first step is to ask the auth plugin to generate an OAuth access token. Do this by adding token_format: 'access_token' to your YAML:

- uses: 'google-github-actions/auth@v0'
  with:
    # ...
    token_format: 'access_token'

If your workflow succeeds after adding the step to generate an access token, it means Workload Identity Federation is configured correctly and the issue is in subsequent actions. You can remove the token_format from your YAML. To further debug:

  1. Look at the debug logs to see exactly which step is failing. Ensure you are using the latest version of that GitHub Action.

  2. Make sure you use actions/checkout@v2 before the auth action in your workflow.

  3. If the failing action is from google-github-action/*, please file an issue in the corresponding repository.

  4. If the failing action is from an external action, please file an issue against that repository. The auth action exports Google Application Default Credentials (ADC). Ask the action author to ensure they are processing ADC correctly and using the latest versions of the Google client libraries. Please note that we do not have control over actions outside of google-github-actions.

If your workflow fails after adding the the step to generate an access token, it likely means there is a misconfiguration with Workload Identity. Here are some common sources of errors:

  1. Look at the debug logs to see exactly which step is failing. Ensure you are using the latest version of that GitHub Action.

  2. Ensure the value for workload_identity_provider is the full Provider name, not the Pool name:

    - projects/NUMBER/locations/global/workloadIdentityPools/POOL
+ projects/NUMBER/locations/global/workloadIdentityPools/POOL/providers/PROVIDER

  3. Ensure you have created an Attribute Mapping for any Attribute Conditions or Service Account Impersonation principals. You cannot create an Attribute Condition unless you map that value from the incoming GitHub OIDC token. You cannot grant permissions to impersonate a Service Account on an attribute unless you map that value from the incoming GitHub OIDC token.

  4. Check the specific error message that is returned.

    • If the error message includes "failed to generate Google Cloud federated token", it means admission into the Workload Identity Pool failed. Check your Attribute Conditions.

    • If the error message inclues "failed to generate Google Cloud access token", it means Service Account Impersonation failed. Check your Service Account Impersonation settings and ensure the principalSet is correct.

  5. Enable Admin Read, Data Read, and Data Write Audit Logging for Identity and Access Management (IAM) in your Google Cloud project.

    Warning! This will increase log volume which may increase costs. To keep costs low, you can disable this audit logging after you have debugged the issue.

    Try to authenticate again, and then explore the logs for your Workload Identity Provider and Workload Identity Pool. Sometimes these error messages are helpful in hinting at the root problem.

  6. Ensure you have waited at least 5 minutes between making changes to the Workload Identity Pool and Workload Identity Provider. Changes to these resources are eventually consistent.

Subject exceeds the 127 byte limit

If you get an error like:

The size of mapped attribute exceeds the 127 bytes limit.

it means that the GitHub OIDC token had a claim that exceeded the maximum allowed value of 127 bytes. In general, 1 byte = 1 character. This most common reason this occurs is due to long repo names or long branch names.

This is a limit imposed by Google Cloud IAM. We have no control over this value. It is documented here. Please file feedback with the Google Cloud IAM team. The only mitigation is to use shorter repo names or shorter branch names.

Dirty git or bundled credentials

By default, the auth action exports credentials to the current workspace so that the credentials are automatically available to future steps and Docker-based actions. The credentials file is automatically removed when the job finishes.

This means, after the auth action runs, the workspace is dirty and contains a credentials file. This means creating a pull request, compiling a binary, or building a Docker container, will include said credential file. There are a few ways to fix this issue:

  • Add and commit the following lines to your .gitignore:

    # Ignore generated credentials from google-github-actions/auth
gha-creds-*.json

    This requires the auth action be v0.6.0 or later.

  • Re-order your steps. In most cases, you can re-order your steps such that auth comes after the "compilation" step:

    1. Checkout
2. Compile (e.g. "docker build", "go build", "git add")
3. Auth
4. Push

    This ensures that no authentication data is present during artifact creation.

  • In situations where auth must occur before compilation, you can use the output to exclude the credential:

    1. Checkout
2. Auth
3. Inject "${{ steps.auth.outputs.credentials_file_path }}" into ignore file (e.g. .gitignore, .dockerignore)
4. Compile (e.g. "docker build", "go build", "git add")
5. Push