Skip to content

Deploy SLURM Workload Manager on Ubuntu to Microsoft Azure

Notifications You must be signed in to change notification settings

tamhinsf/Slurm4Azure

Repository files navigation

Slurm Workload Manager for Microsoft Azure

It's easy to deploy a Slurm cluster to Microsoft Azure! The Azure template on this page will help you quickly create and configure a Slurm cluster running Ubuntu 16.04 (LTS).

When you're done, you'll have an operational Slurm cluster consisting of:

  • One master node - containing a shared directory exported to worker nodes
  • One or more worker nodes - mounting the shared directory
  • A local installation of Slurm and related binaries on both master and worker nodes
  • Full configuration of Slurm and related binaries. This includes the creation of a cluster wide admin user, automatic startup of Slurm processes at boot, and configuration of Slurm compute parameters to match the processor and memory configuration of the underlying virtual machines

NOTE: The Slurm workers created by this template are individual virtual machines and not part of an Azure Virtual Machine Scale Set. We'll enhance functionality in the future to use VM Scale Sets. For now, onwards!

Our Slurm4Azure repository is based on the "slurm" repository found here:

Let's Get Started!

Begin deploying your Slurm cluster by clicking the Deploy to Azure button below. You'll be able to adjust:

  • Azure region you wish to target.
  • Azure Resource Group. We strongly recommend you create a new Resource Group.
  • Username and Password of the cluster-wide Slurm administrator.
  • Azure VM Series used for both the master and worker nodes
    • Make sure the VM Series you enter is available in the Azure region you target. Need help? The Azure VM Comparision website will show you the VMs available in a given region https://azureprice.net/
    • Optionally, you can enable Accelerated Networking if the VM series you select supports it. Not sure if it does? Check the list of VMs that support Acclerated Networking
    • WARNING: If you enable Accelerated Networking on a VM series that doesn't support it, your deployment will fail.
    • WARNING: Optionally, if you make your master node also a worker node, then the VM series you pick must be the same for both. In the future, we'll add validation to ensure that you don't pick VMs from different series.
  • Size of the cluster-wide shared disk, which we'll mount on the master and share to worker nodes.
    • Azure supports data disks of up to 4TB (4095 GB). If you need a shared disk larger than this size, you have the option of specifying multiple data disks. We'll merge these drives into a single, software-based RAID 0.
    • The number of data disks that you can attach to a VM varies based upon the VM series and size. Don't know what your VM supports? Check the list of Linux VMs on Azure
  • Number of worker nodes

Before you deploy, you'll need to Agree to the terms and click Purchase to begin deployment. As a reminder, you're not actually paying to use this free template. However, the resources that you deploy and make use of will be billed to your subscription.

Deploy to Azure

Using your Slurm Cluster

Once you've begun your deployment, you can remain in the Azure Portal or navigate away and come back. Either way, you'll receive a notification in the Azure Portal upon completion. Once this has occured:

  • Navigate to the Azure Resource Group you targeted
  • Look for a virtual machine called "master". This machine is the head node in your Slurm cluster.
    • This VM will be assigned a public IP address that you can use to gain access using SSH.
    • On the "Overview" Pane click the Connect button. You'll be shown the username@ipaddress value you can supply to your SSH client.
  • Connect to "master" using your SSH credentials
  • After you login, look for a file called "done" in your home directory. This is an indication that the scripts used for configuration and deployment have completed. However, this does not necessarily mean that these operations have suceeded. We suggest you do the following:
    • Review the file called /tmp/azuredeploy.log.xxxx where xxxx is a random four digit number. Check for errors. The operations performed by our scripts may have failed due to unexpected network timeouts or other reasons.

    • Use the Slurm commands "sinfo" and "scontrol show nodes" to validate the number of nodes in your cluster and their configuration. Here's the output of "sinfo" showing a cluster containing one master and two workers.

      slurmadmin@master:/tmp$ sinfo
      PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
      debug*       up   infinite      3   idle master,worker[0-1]
      
    • Go to the /data directory: the cluster-wide shared folder. You'll want to run any Slurm workloads requiring cluster-wide shared files in this directory.

      • Take a look at the file called shared-bashrc. The .bashrc file for the Slurm admin account on the master and all worker nodes references this file - making it a great place to store cluster-wide environmnent settings.
      • There should also be a folder called "canu", which contains an open source genomics application of the same name. You can use canu to validate the functionality of your Slurm cluster.

Releases

No releases published

Packages

No packages published

Languages