Forge Home

slurm

Manage compute node clients of a SLURM cluster.

8,740 downloads

5,696 latest version

4.4 quality score

We run a couple of automated
scans to help you access a
module's quality. Each module is
given a score based on how well
the author has formatted their
code and documentation and
modules are also checked for
malware using VirusTotal.

Please note, the information below
is for guidance only and neither of
these methods should be considered
an endorsement by Puppet.

Support the Puppet Community by contributing to this module

You are welcome to contribute to this module by suggesting new features, currency updates, or fixes. Every contribution is valuable to help ensure that the module remains compatible with the latest Puppet versions and continues to meet community needs. Complete the following steps:

  1. Review the module’s contribution guidelines and any licenses. Ensure that your planned contribution aligns with the author’s standards and any legal requirements.
  2. Fork the repository on GitHub, make changes on a branch of your fork, and submit a pull request. The pull request must clearly document your proposed change.

For questions about updating the module, contact the module’s author.

Version information

  • 0.1.3 (latest)
  • 0.1.2
  • 0.1.0
released Jul 23rd 2017
This version is compatible with:
  • ,

Start using this module

  • r10k or Code Manager
  • Bolt
  • Manual installation
  • Direct download

Add this module to your Puppetfile:

mod 'chwilk-slurm', '0.1.3'
Learn more about managing modules with a Puppetfile

Add this module to your Bolt project:

bolt module add chwilk-slurm
Learn more about using this module with an existing project

Manually install this module globally with Puppet module tool:

puppet module install chwilk-slurm --version 0.1.3

Direct download is not typically how you would use a Puppet module to manage your infrastructure, but you may want to download the module in order to inspect the code.

Download
Tags: slurm, munge

Documentation

chwilk/slurm — version 0.1.3 Jul 23rd 2017

slurm

Build Status

Table of Contents

  1. Overview
  2. Module Description - What the module does and why it is useful
  3. Setup - The basics of getting started with slurm
  4. Usage - Configuration options and additional functionality
  5. Reference - An under-the-hood peek at what the module is doing and how
  6. Limitations - OS compatibility, etc.
  7. Development - Guide for contributing to the module

Overview

The Slurm Workload Manager is an open source project designed to manage compute resources on computational clusters of various sizes.

Module Description

This module is intended for the automated provisioning and management of compute nodes in a RedHat/CentOS based SLURM cluster.

For simplicity, some key configuration decisions have been assumed, namely:

  • MUNGE will be used for shared key authentication, and the shared key can be provided to puppet via a URI.
  • Standard MUNGE and SLURM packages have been compiled and made available to compute nodes via their installation repository of choice.
  • slurm.conf is present on a shared file system, and that directory may be provided to the module as a parameter. /etc/slurm will be symlinked to it.

Setup

What slurm affects

  • MUNGE packages will be installed (optional)
  • The shared MUNGE key will be installed from a URI passed as a parameter to the module
  • /etc/slurm will be removed and recreated as a symbolic link to the shared file system passed as a parameter to the module.
  • munge and slurmd services will be kept alive
  • The PAM stack will be (optionally) edited to insert SLURM PAM controls to meter access to the node only to users running jobs on said node.
  • Optionally, the Warewulf Node Health Check will be installed and configured.

Setup Requirements

This module requires SLURM packages be made available by repo to the local node, and a shared file system with the SLURM configuration file be mounted already.

Beginning with slurm

The slurm module requires some parameters to be functional.

The most basic example, using munge with a provided munge key:

class slurm: {
    munge_key_filename => '/shared/secret/munge.key',
    slurm_conf_location => '/shared/slurm/etc',
}

Alternatively, if sharing the munge key over NFS is undesireable, you could set it up on the puppet file server as documented here and then pass the puppet URI as the munge_key_filename.

Usage

The slurm module is intended to be modular for use in differently managed clusters. For instance:

We use something other than MUNGE to authenticate

class slurm: {
    disable_munge => true,
    slurm_conf_location => '/shared/slurm/etc',
}

We don't want users logging into compute nodes whether or not they have jobs there.

class slurm: {
    munge_key_filename => '/shared/secret/munge.key',
    slurm_conf_location => '/shared/slurm/etc',
    disable_pam => true,
}

Reference

The style of this module has been borrowed heavily from the puppetlabs-ntp module.

Classes

Public Classes

  • slurm: Main class, includes all other classes

Private Classes

  • slurm::install: Handles Package resources

  • slurm::config: Handles editing configuration files, symbolic link of /etc/slurm, and the munge key.

  • slurm::service: Handles slurmd and munge services

Parameters

Booleans

Some features of the slurm module can be turned on or off through the use of boolean switches:

#####disable_munge

Turns off all handling of munge keys or services. This may be used in case munge is to be handled separately, or if another authentication system is desired altogether.

Defaults to false

#####disable_pam

Turns off all editing of the PAM stack. PAM will no longer meter access by users running jobs.

Defaults to false

#####disable_slurmd

Turns off slurm daemon service (might be useful on login nodes, for instance)

Defaults to false

#####force_munge

Turns on the munged option --force which causes the munge server to attempt to run even if it is unhappy with its environment.

Defaults to false

#####package_manage

Turns off package installation, in case SLURM and/or MUNGE are to be handled in a different way.

Defaults to true

Strings

#####munge_key_filename

File or Puppet file server path to munge-key accessible by compute node.

#####munge_service_name

Which service to manage for munge.

Defaults to munge for most OS's

#####package_ensure

Set to 'present' by default, you could change this to 'latest' to force Puppet to automatically keep SLURM/MUNGE packages updated.

#####slurm_conf_location

Directory on compute node that contains the shared slurm.conf

Set to undef by default.

#####slurm_service_name

Which service to manage for the local slurmd daemon.

Varies based on distribution.

#####sysconfigdir

Where SLURM expects to find daemon config files on this distro.

Varies based on distribution.

Arrays

#####munge_packages

Set of packages to be maintained for munge.

#####pam_packages

Set of packages to be maintained for SLURM PAM integration.

#####slurm_packages

Set of packages to be maintained for SLURM itself.

Limitations

This module is being developed on Red Hat Enterprise Linux (RHEL) version 6. Contributions helping to port to other distributions or operating systems are welcome. I've tried to leave it in a state that will be considerate of porting efforts.

Development

I would be happy to review bug reports and pull requests via GitHub.

Release Notes/Contributors/Etc

  • Updating disable_slurmd parameter to ensure_slurmd
  • Added disable_slurmd parameter
  • Packaged and uploaded 0.1.0 release to Puppet Forge