Run Dask's NVML tests

[quasiben]

Run tests from dask/distributed#3810

[quasiben]

Current test failures:

PackagesNotFoundError: The following packages are not available from current channels:

  - dask-cudf=0.17
  - ucx-py=0.17
  - cudf=0.17

Looks like a conda issue ?

[quasiben]

rerun tests

[jakirkham]

Yeah OPS is still working on creating those

[quasiben]

Temporarily targeting 0.16

[quasiben]

rerun tests

[jakirkham]

I think this still has history from branch-0.17, which is trying to still install 0.17 packages. Would recommend just rebasing on top of branch-0.16 and removing any commits from branch-0.17. That should hopefully fix the issue.

[quasiben]

I'll just wait until 0.17 things are fixed. Thanks @jakirkham

[quasiben]

rerun tests

[madsbk]

rerun tests

[quasiben]

Still seeing some conda issues. I think they should be resolved soon...

[jakirkham]

Actually I think we need this in branch-0.16 as we consolidated the tests into distributed/diagnostics/tests/test_nvml.py. Though this is only a CI change, so should be safe. Would you be willing to retarget this to branch-0.16 and rebase on top of that branch, Ben?

[jakirkham]

May need to rebase to get rid of commit ( 25327eb ).

[quasiben]

@raydouglass I removed the doc update 0.17 commit. We are targeting this fix for the upcoming release

[jakirkham]

For some reason the commit history shows some commit with explicit comms changes. However those changes are the same ones already in branch-0.16 as commit ( b6cfacc ). I think we can ignore those.

[jakirkham]

For some reason the commit history shows some commit with explicit comms changes. However those changes are the same ones already in branch-0.16 as commit ( b6cfacc ). I think we can ignore those.

Note the resulting commit ( 93458a4 ) only contains the relevant CI change as expected.