I think the issue comes from installing Office365-REST-Python-Client using dbutils.library.installPyPI, which seems to create a conflicting Python environment for Spark executors. Since notebook specific installs modify the environment dynamically, the executors and driver end up out of sync, leading to errors. A better approach is to install the library at the cluster level using the Databricks UI or an init script, so everything runs in a stable, shared environment.