Skip to content

Integrate install_gpu_driver.sh into spark-rapids.sh to avoid duplicate CUDA/driver installation logic #1360

@SurajAralihalli

Description

@SurajAralihalli

The recent MR #1359 updates the Rocky instructions to use the .run file, which is not compatible with Rocky 9.

We should investigate whether spark-rapids.sh can reuse the existing install_gpu_driver.sh script, as it’s already robust and handles most cases across all supported operating systems. The goal is to avoid copying the CUDA and driver installation logic from install_gpu_driver.sh into spark-rapids.sh, since maintaining two separate implementations would lead to inconsistencies and higher maintenance overhead.

At the same time, we want to make the user experience simpler by allowing them to use a single init script instead of having to provide one for Spark and another for CUDA/driver installation. Ideally, spark-rapids.sh should invoke the existing install_gpu_driver.sh during initialization.

Some known limitations of install_gpu_driver.sh are discussed in #1269 (comment).

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions