add devcontainers

[jameslamb]

Follow-up to these PRs:

• move cugraph projects around for 24.12 devcontainers#417
• fix devcontainer builds #68

Proposes adding devcontainers and a devcontainers CI job to the repo.

Notes for Reviewers

Benefits of these changes

• faster and easier local development
• reduced risk of changes here breaking the RAPIDS unified devcontainers maintained in https://github.com/rapidsai/devcontainers

Similar to rapidsai/nx-cugraph#25

How I made these changes

Copied the .devcontainer/ directory from https://github.com/rapidsai/cugraph, then just changed cugraph references to cugraph-gnn.

How I tested this

Tested the update-version.sh changes like this:

./ci/release/update-version.sh '25.04.00'
git grep -E '25\.[0-9]+'

[jameslamb]

Think this will need rapidsai/devcontainers#423 before it can pass.

[jameslamb]

Copied this compiler block directly from cugraph: https://github.com/rapidsai/cugraph/blob/3d681cc9e6d85336822ccb6faf0d1cba5f759ead/dependencies.yaml#L304-L330

To try to fix this error in the conda devcontainer build:

  CMake Error at /home/coder/.conda/envs/rapids/share/cmake-3.31/Modules/CMakeDetermineCXXCompiler.cmake:48 (message):
    Could not find compiler set in environment variable CXX:
  
    /usr/bin/g++.
  
  Call Stack (most recent call first):
    CMakeLists.txt:39 (project)

(build link)

Most other RAPIDS repos I checked have something like this, and I found offline conversations where it was suggested that the conda devcontainer builds expect the conda environment to provide compilers.

[bdice]

Yes, this is a good fix.

[bdice]

One question about ucx versions. Otherwise approved.

[bdice]

This uses ucx 1.15 but the CUDA 12 container uses ucx 1.17. Is that right?

[jameslamb]

That's a great catch! I think it should be 1.17 here too.

I copied these directly from cugraph, looks like that was an oversight in rapidsai/cugraph#4562 (comment).

Let's try 1.17 here.

[bdice]

We need to fix this in cugraph and nx-cugraph. https://github.com/search?q=org%3Arapidsai%20ucx1.15&type=code

[jameslamb]

I agree. I'll do that alongside this other UCX-related devcontainers change I'm about to go make: rapidsai/devcontainers#421 (comment)

[jameslamb]

I just pushed a commit adding those changes (setting RAPIDS_LIBUCX_PREFER_SYSTEM_LIBRARY=true in the pip devcontainers) here.

And manually cancelled CI because it'll fail until rapidsai/raft#2513 is done.

[bdice]

Yes, this is a good fix.

[jameslamb]

wholegraph wheel builds are failing like this:

  /__w/cugraph-gnn/cugraph-gnn/python/pylibwholegraph/build/cp311-cp311-linux_aarch64/_deps/raft-src/cpp/include/raft/core/detail/callback_sink.hpp:22:10: fatal error: spdlog/common.h: No such file or directory
     22 | #include <spdlog/common.h>
        |          ^~~~~~~~~~~~~~~~~

(build link)

I think that's a result of the ongoing logging changes, and should be fixed by rapidsai/raft#2513.

[jameslamb]

/merge