Repository navigation
Enable WSL testing in CI #114
Description
Currently RAPIDS claims support for WSL but only tests this support via ad hoc testing on individual machines. The primary purpose of this testing is to verify that the CUDA driver in WSL2 works properly and handles everything that RAPIDS libraries need. We would like to be able to run these tests in a more systematic manner, ideally in CI. However, there are currently a number of blockers for this. This issue is meant for tracking the various tasks that must be accomplished in order to get CI testing on WSL2.
- Windows GPU runners: Currently we have some limited support for Windows runners. We do not have WSL2 support because you cannot do nested virtualization on the EC2 runners. Enabling WSL will require that we transition CI to NVKS. That is currently being handled by @ajschmidt8's team.
- WSL2 vs WSL: It looks like currently the Windows VMs that we have support for on our runners are installing WSL. We would like to get these running WSL2. The nested virtualization requirement discussed above is only for WSL2, not WSL1, so we need to address 1 above before we can switch to WSL2 on GPU runners. However, we could update our setup to use WSL2 as long as we are just testing on CPU runners.
- We need to figure out how to set up the WSL2 instance so that it is ready to run everything that we need to test RAPIDS libraries, including perhaps installing Docker.
- We need scripts to actually run tests. This can be worked on by individual libraries.
Activity
vyasr commentedon Oct 28, 2024
I may work with @wmaxey on 2/3 as time permits.
bdice commentedon Nov 3, 2026
I started an experiment in rapidsai/rmm#2339. I'm going to see what I can do to test this a bit more.
bdice commentedon Nov 3, 2026
WSL CI rollout
We'll add WSL test jobs using the migration-matrix process in MATRIX_GUIDELINES.md, the same way as the CUDA 13.4 migration (#335). WSL only adds test jobs (no new builds), so library PRs can merge in any order.
New jobs (
OS: 'wsl', runnerwsl-amd64-gpu-<GPU>-latest-1):conda-cpp-testswsl-pr: amd64, py3.12, CUDA 13.0.3, ubuntu24.04, rtxpro6000wsl-nightly: amd64, py3.12, CUDA 13.0.3, rockylinux8, rtxpro6000wheels-testwsl-pr: amd64, py3.14, CUDA 13.4.3, ubuntu24.04, h100wsl-nightly: amd64, py3.12, CUDA 13.0.3, ubuntu24.04, h100Infrastructure PRs
OSin the conda C++ and wheel test matrices and addwsl-pr/wsl-nightlymatrices (Add WSL migration matrices for conda C++ and wheel tests shared-workflows#683)wsl-ciexampleLibrary PRs
add
wsl-pr/wsl-nightlytomatrix_typeforconda-cpp-testsandwheels-testjobs, fix any WSL-specific test failures, merge in any order once CI passesOther follow-up (after all library PRs)
pull-request/nightlymatrices, replacing equivalent Linux jobs to keep the matrix size constantwsl-pr/wsl-nightlyfrommatrix_typein all libraries (GitHub search)Cleanup after everything else above
wsl-pr/wsl-nightlymatriceswslbranchAdd WSL migration matrices for conda C++ and wheel tests (#683)