Ray Dataset: ArrowInvalid: Unrecognized filesystem type in URI: gs://

Pyarrow (as of 6.0.1, the latest at the moment) does not have support for a direct GCS filesystem and is not able to recognize gs:// URIs. You can create a GCS filesystem using an fsspec adapter. I think you can use this in Ray with:

import gcsfs
fs = gcsfs.GCSFileSystem(project='my-google-project')

ray.data.read_parquet("path", filesystem=fs)

Pyarrow 7.0.0 (which will likely release in the next month) should include some support for a native GCS filesystem (I'm not entirely clear on how much will be included in 7.0.0 but I know it is actively being worked on so check the release notes). Support for gs:// URIs does not yet appear to have been implemented.