Working around the gotchas, logging inside a pandas_udf, etc