It looks like you’re running into a classic AADSTS7000215 invalid client secret issue, which often comes up with Airflow operators that connect via Azure AD. Based on your steps, you’ve already confirmed the basics, but here are a few things worth trying:
1. Double-check secret validity and expiration
In Azure AD, ensure the client secret hasn’t expired. Even if it looks fine, try regenerating a new secret and updating your FabricUser connection. Sometimes a refresh clears hidden mismatches.
2. Verify redirect URIs and API permissions
Make sure your app registration includes the correct redirect URI for Airflow and has API permissions for Power BI (e.g., Dataset.ReadWrite.All). Without these, the operator may fail authentication.
3. Use a Service Principal instead of FabricUser (if possible)
While the FabricUser connection works well for orchestrating notebooks, PowerBIDatasetRefreshOperator is often more stable with a Service Principal–based connection. You can configure this in Azure AD and update your Airflow connection accordingly.
4. Cross-check Airflow logs for token caching issues
Sometimes Airflow workers cache tokens incorrectly, especially if multiple DAGs are using the same connection. Clearing the token cache or restarting the Airflow worker can resolve intermittent auth errors.
For more details, Microsoft has a helpful guide on using Service Principals for Power BI API authentication, https://learn.microsoft.com/en-us/power-bi/developer/embedded/embed-service-principal?tabs=azure-por... which may give you additional clarity.
Since you mentioned needing consistent results, treating this as part of a guaranteed work process is crucial ensure secrets are rotated securely and connections are validated before each DAG run. You may also want to review the community thread on Airflow Power BI operator configuration, https://community.fabric.microsoft.com/t5/Power-Query/Authentication-issues-with-Power-BI-service-pr... where users shared fixes that align closely with your situation.
