hpc-unibe-ch / puppet-module-slurm

Basic puppet module to manage packges, directories and files of the different slurm aspects. However, does NOT manage the slurm config in /etc/slurm/ - by intention!

slurm

Table of Contents

• Description
• Setup
  • What slurm affects
  • Setup Requirements
  • Beginning with slurm
• Usage
• Reference
• Limitations
• Development

Description

This slurm module is a downsized version of treydock/slurm. Its complexity is reduced by removing all aspects of configuring slum. In short, every aspect that is done in the "slurm dir" is removed, i.e. managing partitions, QOS, users, GRES. We at our site are manging the contents of /etc/slurm in its own git repository independent from managing packges, files and daemons.

Setup

What slurm affects

In short this module only installs relevant packages and dependencies, deploys files and directories (state, log, logroration...) and manages daemons and firewall rules.

Setup Requirements

The module's design allows changing the most important things through the slurm class. The daemons getting installed are chosen by the setting the respective attributes to true. More than one role can be deployed on a host, e.g. slurmdbd and slurmctld. See usage examples below.

In order to use SLURM the munge daemon must be configured and a munge key has to be deployed. This module does not manage neither of them. Instead make use of an additional module like treydock/munge.

Instead of declaring the attributes within the a manifest it's perfectly possible to provide the configuration data in hiera. The module itself is using hiera for its default data.

Beginning with slurm

By just including the slurm class as follow, the node is setup as a SLURM client and the defaults set in module's common.yaml is applied:

include slurm

Other use cases are outlined in the right below.

Usage

By just including the class slurm the given host acts as a SLURM client. This is the role primarily used on thing like login nodes.

On a worker node the daemon slurmd is needed. To designate a node as a worker node, the following snippet will do the job.

class { 'slurm':
  slurmd => true,
}

Likewise for the designation of a slurm control daemon, the parameter slurmctld is used. In the following example, not only the control daemon gets installed but also the slurmdbd daemon to connect to a MySQSL database server. Setting up the database server itself is out of the scope of this module and has to be installed elsewhere.

class { 'slurm':
  slurmdbd  => true,
  slurmctld => true,
}

All of the parameters can also be set by putting the configuration data to hiera and make use automatic parameter binding feature of Puppet.

For all possible configuration parameters the reference below should be consulted.

Reference

See REFRENCE.md in this repository.

Limitations

This module is currently only tested on CentOS-7.

Development

This project is using PDK and feature unit tests. In order to participate, please take the time to get into it to if not already familiar with.

If you want to contribute to this project:

• File an issue over at github.com telling us about the changes you have mind
• Fork the repository and make your changes on a feature branch
• Genereate a pull request against the main branch. Don't forget to rebase before creating a PR.