NVIDIA / NVIDIA/cudf

[FEA] Add Velox-cuDF CI jobs

Open
#22,648 0 comments 0 reactions 0 assignees View on GitHub
ci feature request libcudf Velox
Dominant language
C++
Stars
9.8k
Forks
1.1k
Avg merge
3d 6m
Merged PRs (30d)
278

Description

### Motivation

libcudf currently has no CI coverage for Velox-cuDF integration. As a result, changes in libcudf can inadvertently break Velox builds without being detected until a subsequent cuDF commit bump is attempted in Velox. A recent example was the libcudf migration to CMake 4.0, which introduced build failures in Velox, as Velox is still pinned to CMake 3.28.

Given our ongoing commitment to maintaining Velox-cuDF, it would be beneficial to establish CI coverage for Velox integration against cuDF PRs, similar to the existing setup for [spark-rapids-jni](https://github.com/rapidsai/cudf/blob/main/.github/workflows/spark-rapids-jni.yaml). This would help surface integration issues earlier in the development cycle and follows an already established integration-testing pattern within the repository.

### Proposal

- **Phase 1**: Build verification (non-blocking)
https://github.com/rapidsai/cudf/pull/22647
Introduce a build-only CI job that compiles Velox against the cuDF PR branch. The job would be triggered for relevant C++ changes and initially configured as non-blocking for merges, while still providing useful integration signals and early visibility into potential breakages.

- **Phase 2**: Build + cuDF tests in Velox
Extend the CI coverage to execute the Velox-cuDF GPU test suite.
The number of GPU tests in Velox-cuDF is relatively small, making the additional CI cost manageable. This would be comparable in spirit to the existing upstream compatibility testing we already perform for projects such as Pandas and Polars.

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.