Seems like R2 does not work well with pyarrow.
See pyarrow JIRA ticket: https://issues.apache.org/jira/browse/ARROW-18076
For context:
- pyarrow is a library for reading/writing parquet files, but it also handles providing a S3 file system API for Python
This pyarrow file system API works fine for files stored in S3 but chokes on files stored in R2.