Skip to content

Fix Spark RAPIDS Ubuntu 22 driver install#1394

Open
nvliyuan wants to merge 2 commits into
GoogleCloudDataproc:mainfrom
nvliyuan:fix-spark-rapids-ubuntu22-dkms
Open

Fix Spark RAPIDS Ubuntu 22 driver install#1394
nvliyuan wants to merge 2 commits into
GoogleCloudDataproc:mainfrom
nvliyuan:fix-spark-rapids-ubuntu22-dkms

Conversation

@nvliyuan
Copy link
Copy Markdown
Contributor

Summary

  • Updates spark-rapids.sh so Ubuntu 22 uses the online NVIDIA CUDA repo for driver packages instead of the older local CUDA repo driver embedded in the installer.
  • Selects gcc-12 for Ubuntu 22 CUDA 12 driver builds, matching the newer Dataproc 2.2 GCP kernel toolchain requirements.
  • Keeps CUDA toolkit installation pinned to the selected CUDA major/minor package when using the online repo.

Test plan

  • bash -n spark-rapids/spark-rapids.sh
  • IDE lint check for spark-rapids/spark-rapids.sh
  • On yuanli@spark-yuanli:~/work/csps/dataproc/spark-rapids-ubuntu22-fix, ran ./create-cluster.sh against 2.2-ubuntu22 (2.2.82-ubuntu22) with the patched init action; cluster sr-u22-fix-160203 created successfully and nvidia-smi --query-gpu=driver_version --format=csv,noheader returned 550.163.01.

Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist Bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request updates the spark-rapids.sh script to enable online repository installation on Ubuntu 22, ensuring compatibility with newer GCP kernels on Dataproc 2.2. It also introduces a compiler configuration helper to upgrade to gcc-12 on Ubuntu 22 when CUDA 12 is targeted, and dynamically resolves the CUDA toolkit package name. The reviewer suggested also installing and configuring g++-12 alongside gcc-12 to prevent potential compilation and compatibility issues with host compilers like nvcc.

Comment thread spark-rapids/spark-rapids.sh
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant