Unexpected behaviors in recent notebook versions

[bernatjvg]

Issue summary:

When trying to use notebooks with M87 version in order to use LTS Tensorflow 2.6, and configuring them to use jgscm, this behaves unexpectedly:

• First, when opening the Jupyterlab UI, we can observe a “Directory not found” toast that can be dismissed.
• Upon trying to create a new folder or performing similar actions we encounter “Unhandled error” errors, or similar 403 Forbidden errors like [1].
• Going to the /var/log directory of the instance and executing cat syslog we’re also able to find the permission error [2] related to this.
• Furthermore, in the directory structure on the left, the list of buckets is displayed, while instead the contents of the bucket linked to the notebook should be displayed as the top directory.

This error [2] can be circumvented by granting additional permissions to the service account used in the notebook (the minimum set of roles we have found is granting Storage Object Creator + Storage Object Viewer to the bucket linked to the notebook), but there are still unexpected behaviors after this too, like not being able to access directories that contain ipynb files.

Please, address this issue so that this module can be used in more recent versions of the notebooks such as M87 or higher.

---

[1]:
Error: Unexpected error while saving file: Untitled.ipynb HTTP 403: Forbidden (You may only create directories (buckets) at the root level.)

[2]:
google.api_core.exceptions.Forbidden: 403 GET https://storage.googleapis.com/storage/v1/b/untitled-folder?projection=noAcl&prettyPrint=false: SERVICE_ACCOUNT@PROJECT_ID.iam.gserviceaccount.com does not have storage.buckets.get access to the Google Cloud Storage bucket.