Slurm Debian Package Builder

This repository automates the process of building Debian packages for Slurm and NVIDIA NCCL. It also builds binary files for running NCCL tests.

Slurm is built using default options, with support for some additional libraries, such as OpenPMIx (v5).

The Build slurm packages workflow builds the maintained upstream 25.11.5 release and Nebius Slurm 26.05.4-nebius-1 from one build matrix. The Nebius source comes from the Nebius Slurm fork and uses the immutable slurm-26-05-4-nebius-1 release.

Every source uses distinct artifact names, Debian package versions, and GitHub release tags. The Nebius source archive already records Debian version 26.05.4-nebius-1; the packaging workflow does not rewrite source metadata.

The Pyxis workflow builds the SPANK plugin against every supported Slurm development package. Add each Nebius Slurm release to the Pyxis matrix as well, using its exact Debian version (currently 26.05.4-nebius-1). Other ancillary packages built in this repository do not compile or link against Slurm. Run the Slurm package workflow and wait for those packages to reach the public APT repository before starting Pyxis, which installs the exact listed version.

Slurm is a highly scalable cluster management and job scheduling system for Linux clusters. To facilitate its installation and integration into Debian-based systems, this repository provides automated scripts to build .deb packages for Slurm and some of its key dependencies, including OpenPMIx and NCCL.

Additionally, binary files for running NCCL tests are built to verify and benchmark the performance of the NCCL library in various configurations.

1. Add the public key and repository

   sudo curl -fsSL https://dr.nebius.cloud/public.gpg -o /usr/share/keyrings/nebius.gpg.pub
   
   echo "deb [signed-by=/usr/share/keyrings/nebius.gpg.pub] https://dr.nebius.cloud/ stable main" | \
     sudo tee /etc/apt/sources.list.d/nebius.list > /dev/null

2. Install the package

   sudo apt update
   sudo apt install slurm-smd

The build process includes the following components:

• Slurm: A workload manager that facilitates resource management and scheduling in HPC environments.
• OpenPMIx: The Open Process Management Interface for Exascale, which provides a set of interfaces for managing the execution of applications at large scales.
• NVIDIA NCCL: NVIDIA's collective communication library optimized for multi-GPU and multi-node systems.
• NCCL tests: A suite of tests provided by NVIDIA to validate and benchmark the performance of the NCCL library across different hardware configurations.

To explore and download available packages, go to the Releases page in this GitHub repository.