Skip to content

Latest commit

 

History

History
180 lines (123 loc) · 5.69 KB

README.md

File metadata and controls

180 lines (123 loc) · 5.69 KB

Learning Slurm

Builds a slurm cluster in AWS for learning.

The built system looks like the following diagram. Each server is running Ubuntu 22.04 LTS.

System Architeture Diagram

This configuration is very basic at the moment - definitely just for learning, not production.

It's a work in progress with more to come at some point:

  • more sophisticated cluster configuration
  • users with home directories on the nfs server
  • adding support for slurmrestd
  • some tools to use the rest API

And I'm sure there are a lot of things I don't know about yet that will get added as well.

Prerequisites

You need an AWS account for this to work. If you can get the AWS CLI working, then you will have everything in place that you need to get this to run.

Terraform is used to build the AWS infrastructure and also create the ansible scripts.

Ansible is used to configure the servers.

Quickstart

The build and configuration is fully automated with the terraform and ansible files in the build-aws directory.

First, within the build-aws directory, copy the terraform.tfvars.example file to terraform.tfvars and update the variables to:

  • set the aws profile and region to use
  • set a custom internal domain
  • set the number of workers to create

Build the infrastructure and ansible configurations with terraform:

terraform init
terraform apply

This create a directory called local and in there you can see the ansible configs as well as an ssh configuration file so you can log into all the machines that have been created. For example, to log into the slurm controller:

ssh -F local/ssh.cfg slurmctl

Before moving on to the next step, wait for the servers to be fully up and running. If you can log into them and see that their hostname is different from the default AWS naming, then everything is ready.

Configure the machines:

./local/ansible/run-ansible.sh

At this point you should be able to do some basic testing as in the next section.

To shut it all down, run:

terraform destroy

Testing

You should be able to follow along with this tutorial video that's referenced from the SchedMD website. The instructions below are basically the same as in the video.

Log into the controller:

ssh -F local/ssh.cfg slurmctl

Get some information about the cluster:

sinfo

PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
studio*      up   infinite      2   idle slurm-[00-01]

Take a look at the details for one of the nodes:

scontrol show node slurm-00

NodeName=slurm-00 Arch=x86_64 CoresPerSocket=1 
   CPUAlloc=0 CPUTot=1 CPULoad=0.06
   AvailableFeatures=(null)
   ActiveFeatures=(null)
   Gres=(null)
   NodeAddr=slurm-00 NodeHostName=slurm-00 Version=21.08.5
   OS=Linux 5.19.0-1024-aws #25~22.04.1-Ubuntu SMP Tue Apr 18 23:41:58 UTC 2023 
   RealMemory=1 AllocMem=0 FreeMem=201 Sockets=1 Boards=1
   State=IDLE ThreadsPerCore=1 TmpDisk=0 Weight=1 Owner=N/A MCS_label=N/A
   Partitions=studio 
   BootTime=2023-05-10T21:33:59 SlurmdStartTime=2023-05-10T22:11:46
   LastBusyTime=2023-05-10T22:16:35
   CfgTRES=cpu=1,mem=1M,billing=1
   AllocTRES=
   CapWatts=n/a
   CurrentWatts=0 AveWatts=0
   ExtSensorsJoules=n/s ExtSensorsWatts=0 ExtSensorsTemp=n/s

Create a simple test script called hello-slurm.sh:

#!/usr/bin/env bash
echo hello $HOSTNAME
sleep 10
exit $?

Submit the job:

sbatch -N1 -n1 hello-slurm.sh

Submitted batch job 1

Check the status using the job id returned:

scontrol show job 1

JobId=1 JobName=hello-slurm.sh
   UserId=ubuntu(1000) GroupId=ubuntu(1000) MCS_label=N/A
   Priority=4294901759 Nice=0 Account=(null) QOS=normal
   JobState=RUNNING Reason=None Dependency=(null)
   Requeue=1 Restarts=0 BatchFlag=1 Reboot=0 ExitCode=0:0
   RunTime=00:00:06 TimeLimit=UNLIMITED TimeMin=N/A
   SubmitTime=2023-05-10T22:16:25 EligibleTime=2023-05-10T22:16:25
   AccrueTime=2023-05-10T22:16:25
   StartTime=2023-05-10T22:16:25 EndTime=Unknown Deadline=N/A
   SuspendTime=None SecsPreSuspend=0 LastSchedEval=2023-05-10T22:16:25 Scheduler=Main
   Partition=studio AllocNode:Sid=slurmctl:4722
   ReqNodeList=(null) ExcNodeList=(null)
   NodeList=slurm-00
   BatchHost=slurm-00
   NumNodes=1 NumCPUs=1 NumTasks=1 CPUs/Task=1 ReqB:S:C:T=0:0:*:*
   TRES=cpu=1,node=1,billing=1
   Socks/Node=* NtasksPerN:B:S:C=0:0:*:* CoreSpec=*
   MinCPUsNode=1 MinMemoryNode=0 MinTmpDiskNode=0
   Features=(null) DelayBoot=00:00:00
   OverSubscribe=NO Contiguous=0 Licenses=(null) Network=(null)
   Command=/home/ubuntu/hello-slurm.sh
   WorkDir=/home/ubuntu
   StdErr=/home/ubuntu/slurm-1.out
   StdIn=/dev/null
   StdOut=/home/ubuntu/slurm-1.out
   Power=

Submit a bunch of jobs in held state so they don't disappear too quickly:

sbatch -H -N1 -n1 hello-slurm.sh
sbatch -H -N1 -n1 hello-slurm.sh
sbatch -H -N1 -n1 hello-slurm.sh

Check the status of them all:

squeue

JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
    4    studio hello-sl   ubuntu PD       0:00      1 (JobHeldUser)
    3    studio hello-sl   ubuntu PD       0:00      1 (JobHeldUser)
    2    studio hello-sl   ubuntu PD       0:00      1 (JobHeldUser)

And cancel them:

scancel 2
squeue

JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
    4    studio hello-sl   ubuntu PD       0:00      1 (JobHeldUser)
    3    studio hello-sl   ubuntu PD       0:00      1 (JobHeldUser)

scancel 3 4
squeue

JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)