Skip to content

Remove epel-release package from wheel Dockerfiles#2987

Open
ksivaman wants to merge 1 commit into
NVIDIA:mainfrom
ksivaman:rm_epel_release_pkg_from_wheel_dockerfile
Open

Remove epel-release package from wheel Dockerfiles#2987
ksivaman wants to merge 1 commit into
NVIDIA:mainfrom
ksivaman:rm_epel_release_pkg_from_wheel_dockerfile

Conversation

@ksivaman
Copy link
Copy Markdown
Member

Description

We not longer need this package and the installation of it is causing a significant bloat/time when constructing the wheels.

Type of change

  • Documentation change (change only to the documentation, either a fix or a new content)
  • Bug fix (non-breaking change which fixes an issue)
  • New feature (non-breaking change which adds functionality)
  • Breaking change (fix or feature that would cause existing functionality to not work as expected)
  • Infra/Build change
  • Code refactoring

Changes

  • Remove epel-release package from wheel Dockerfiles.

Checklist:

  • I have read and followed the contributing guidelines
  • The functionality is complete
  • I have commented my code, particularly in hard-to-understand areas
  • I have made corresponding changes to the documentation
  • My changes generate no new warnings
  • I have added tests that prove my fix is effective or that my feature works
  • New and existing unit tests pass locally with my changes

Signed-off-by: Kirthi Shankar Sivamani <ksivamani@nvidia.com>
@ksivaman ksivaman requested a review from timmoon10 May 13, 2026 17:30
@greptile-apps
Copy link
Copy Markdown
Contributor

greptile-apps Bot commented May 13, 2026

Greptile Summary

This PR removes the epel-release package installation from both wheel-building Dockerfiles (Dockerfile.aarch and Dockerfile.x86) to reduce build time and image bloat. It also fixes a missing trailing newline at the end of Dockerfile.x86.

  • epel-release is removed from both Dockerfile.aarch and Dockerfile.x86; all CUDA/NCCL/cuDNN packages are already sourced from the NVIDIA CUDA repo added via dnf config-manager.
  • Dockerfile.x86 gains a trailing newline as a minor formatting fix.

Confidence Score: 4/5

Safe to merge if the author has verified the Docker build completes successfully after removing epel-release, particularly that glog resolves correctly.

The change is minimal and well-scoped. The only uncertainty is whether glog and glog-devel — packages commonly sourced from EPEL on RHEL 8 / AlmaLinux 8 — remain resolvable via the NVIDIA CUDA repo or the manylinux_2_28 base image alone. If the build was tested end-to-end after this removal, the risk is negligible.

Both Dockerfiles warrant a quick sanity check to confirm glog/glog-devel installation still succeeds without EPEL.

Important Files Changed

Filename Overview
build_tools/wheel_utils/Dockerfile.aarch Removes the epel-release installation step; potential risk if glog/glog-devel relies on EPEL on this AlmaLinux 8-based image
build_tools/wheel_utils/Dockerfile.x86 Removes the epel-release step; also fixes a missing newline at end of file; same glog/glog-devel EPEL-dependency concern as the aarch Dockerfile

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[FROM manylinux_2_28] --> B[dnf config-manager: add NVIDIA CUDA repo]
    B --> C{epel-release removed}
    C -->|Before PR| D[dnf install epel-release]
    C -->|After PR| E[skipped]
    D --> F[dnf install CUDA packages]
    E --> F
    F --> G[dnf install cuDNN]
    G --> H[dnf install cuda-toolkit]
    H --> I[dnf install glog / glog-devel]
    I -->|Needs repo with glog| J{Available?}
    J -->|EPEL removed| K[⚠️ May fail if glog not in base/NVIDIA repo]
    J -->|Already in image| L[dnf install libnccl]
    L --> M[Set ENV vars]
    M --> N[CMD: build_wheels.sh]
Loading

Comments Outside Diff (1)

  1. build_tools/wheel_utils/Dockerfile.aarch, line 35 (link)

    P1 glog may no longer resolve without EPEL

    On RHEL 8 / AlmaLinux 8 systems, glog and glog-devel are commonly provided only by EPEL; they are absent from the BaseOS and AppStream repos. If the manylinux_2_28 base image does not ship its own glog package or bundle a suitable repo, both dnf -y install glog.aarch64 glog-devel.aarch64 (this file) and the equivalent glog.x86_64 line in Dockerfile.x86 will fail with "No match for argument: glog". Can you confirm that glog is available via the NVIDIA CUDA repo or that the base image already provides it?

Reviews (1): Last reviewed commit: "Remove epel-release package from wheel D..." | Re-trigger Greptile

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