[FEA]: Improve CI coverage of cuFile

[mdboom]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

cuda.bindings

Is your feature request related to a problem? Please describe.

The majority of the tests in test_cufile.py are skipped because:

@pytest.mark.skipif(not isSupportedFilesystem(), reason="cuFile handle_register requires ext4 or xfs filesystem")

Maybe we should try to add a worker to the mix that /is/ on one of these filesystems to get some better coverage there. I don't know what filesystem our existing workers have (maybe btrfs?)

Describe the solution you'd like

Find a worker that meets the requirements of cuFile that we can run on.

Describe alternatives you've considered

No response

Additional context

No response

[leofang]

@vyasr @kingcrimsontianyu How do you guys test kvikio with nv-gha-runners? Are you able to get full test coverage over everything cuFILE supports?

[kingcrimsontianyu]

My underwhelming answer 😄 is that we haven't tested actual GDS I/O on cloud/CI. When KvikIO detects at runtime that the system setup does not meet GDS requirements, it will switch to the parallel I/O implementation which is based on the standard Linux system calls.

[leofang]

Thanks, Tianyu! Sounds like this is blocked 🥲

[leofang]

Blocked by https://github.com/nv-gha-runners/roadmap/issues/318