Fabric UDFs Deployment ISSUEs

This looks like a Fabric User Data Functions (UDF) deployment/runtime issue rather than an issue with your Python code itself.

Since the function works in Test mode but fails in Run/Published mode, and you have isolated the problem to the "google-cloud-bigquery" / "google.oauth2" packages, I would check the production dependency resolution first.

A few things I would try:

1. Pin the package versions instead of relying on the latest versions. For example, explicitly define compatible versions of:

   - "google-cloud-bigquery"

   - "google-auth"

   - "google-api-core"

2. Check the published environment's dependency list and confirm that the Google packages are actually being included after publishing. Test mode can sometimes have a different dependency/runtime context than the deployed function.

3. Try importing the packages separately:

   import google.auth

import google.oauth2.credentials

import google.cloud.bigquery

   This can help identify whether the failure occurs during module initialization.

4. Since the error is only:

   "WorkloadException / NotFound"

   and does not expose the underlying Python exception, check the Fabric monitoring/logging output for the published invocation. The actual import/dependency error may be hidden behind the generic workload error.

5. As a diagnostic, temporarily remove the BigQuery imports and the "get_bigquery_client()" call. If the published function runs successfully, that further confirms the dependency/runtime issue.

One other thing I'd test is changing:

from google.oauth2.credentials import Credentials

to a lazy import inside the "PERSONAL" branch. That prevents the Google OAuth package from being loaded when the function is initialized:

if auth_type_upper == "PERSONAL":

    from google.oauth2.credentials import Credentials

    ...

Given that this worked previously and started failing after a relatively small change, I'd also compare the package versions and Fabric runtime/deployment configuration between the last working publication and the current one.

If all of this checks out, I'd raise it with Microsoft as a Fabric Functions deployment regression, providing the function definition, dependency versions, and the fact that Test mode succeeds while Published/Run mode returns "WorkloadException: NotFound".

Please let me know if any of the above solution works for you .