XArray in Dask-CUDA CI

[quasiben]

@jacobtomlinson has been exploring adding more GPU functionality with xarray and in PR pydata/xarray#4214 he has added initial CuPy tests. I'd like to ask if dask-cuda can support this effort by running xarray GPU tests in the CI similar to how we test distributed -- I do not expect this tests to be overly onerous and lengthy.

[pentschev]

@jacobtomlinson and I have briefly chatted earlier today and we'll work on having tests run nightly on our DGXs, just as we do for multi-GPU tests for dask-cuda and UCX-Py. Jacob suggested as well that we report failures automatically to GH in the form of issues, and to prevent many false positives we might come up with a secondary repo just for the automated reporting, which will give us a chance to parse them and file appropriate issues for the appropriate project.

[pentschev]

For broader visibility, NEP-35 should be part of NumPy 1.20, which will allow us to unblock parts of this.

[github-actions]

This issue has been marked stale due to no recent activity in the past 30d. Please close this issue if no further response or action is needed. Otherwise, please respond with a comment indicating any updates or changes to the original issue and/or confirm this issue still needs to be addressed. This issue will be marked rotten if there is no activity in the next 60d.

[jakirkham]

NumPy 1.20 is out. Though guessing we still need fixes to Dask

[pentschev]

We have some xarray tests running nightly and being reported in pentschev/ucx-py-ci#5 . We could extend those eventually though.

[jakirkham]

Gotcha. Thanks Peter! So in that case, maybe we should close and open new issues for any additional needed tests. WDYT?

[pentschev]

I'm fine either way, but I don't know now what are the additional needed tests. 🙂

[jakirkham]

In that case, let's close. We have some testing. If we need more, we can add them to the repo you've already setup