Setup Slurm Accounting with AWS ParallelCluster

slurm-accounting-aws-parallelcluster.md

Slurm Accounting with AWS ParallelCluster

In this tutorial we will work through setting up Slurm Accounting. This enables many features within slurm, including job resource tracking and providing a necessary building block to slurm federation.

Step 1 - Setup External Accounting Database

The first requirement is to setup an external database that Slurm can use to store the accounting data.

Use the following CloudFormation Quick-Create link to create the database in your AWS account. Note that if you would like to create the databas in a different region, change the value of the region parameter in the URL to the region of your choice and reload the page.

When you're creating the stack, be sure to specify the VPC ID and Subnets parameters to correspond to the VPC where you are creating the stack. All other values should be suitable as defaults, however feel free to change the database instance type depending on your workload needs.

Note

Change the region in the URL to create the stack in a region separate from us-east-1.

Step 2- Retrieve the outputs from the CloudFormation stack

Once the stack has reached a Completed state. You will need to go to the Outputs tab of the stack and make note of the properties as they will be used in the creation of your cluster.

Step 3 - Add permissions to your lambda

In order to allow our cluster access to secrets we need to add an additional IAM policy.

1. Go to the Lambda Console (deeplink) and search for PclusterManagerFunction
2. Select the function then Configuration > Permissions > Click on the role under Role name.
3. Select Add permissions > Attach policies > search for SecretsManagerPolicy
4. Click Attach policies

Step 4 - Create Your Cluster

Next, go to Pcluster Manager and choose the Create option to create a new cluster.

1. Select Wizard option and click next

Cluster Properties

Choose a suitable name for your cluster, and then in the Cluster Properties window, be sure to choose the VPC that you used when creating the slurm-accounting CloudFormation stack.

HeadNode Properties

You will need to enable the Virtual Console option as that allows Pcluster Manager to interact with the cluster directly:

Be sure to also enable the Security Group referenced in the CloudFormation outputs so that the HeadNode can access the database.

Next we'll enable a known script that will install slurm accounting on the HeadNode.

• Choose the advanced options
• Under the On Configured option, Choose the Multi-Script Runner which has some pre-programmed scripts in it
• In the search box choose Slurm Accounting
• Fill in the values for the Secret ARN and RDS Endpoint from the CloudFormation output
• Under IAM Policies add the arn from the CloudFormation Stack output SecretsManagerPolicy so that the HeadNode can access the password to the database. Be sure to actually click Add so that it is added to the list.

Review Config

After you've configured the HeadNode, Filesystem and Queues, you'll be asked to review the config. The following parameters must be set:

Parameter	Description
AdditionalSecurityGroups	SlurmDbSecurityGroupId (CloudFormation)
AdditionalIamPolicies	AmazonSSMManagedInstanceCore, SecretsManagerPolicy (CloudFormation)
CustomActions/OnNodeConfigured	multi-runner.py
Arg 0: Accounting Script	slurm-accounting.py
Arg 1: SECRET_ARN	SlurmDbPasswordSecretArn (CloudFormation)
Arg 2: RDS Endpoint	SlurmDbEndpoint (CloudFormation)
Arg 3: Port	Default is 3306

Here's an example config file to reference, take a look a the comments to see what's required:

HeadNode:
  InstanceType: t2.micro
  Networking:
    SubnetId: subnet-12345678910
    AdditionalSecurityGroups:
      - sg-12345678910  # Security Group `SlurmDbSecurityGroup`
  Iam:
    AdditionalIamPolicies:
      - Policy: arn:aws:iam::aws:policy/AmazonSSMManagedInstanceCore
      - Policy: arn:aws:iam::12345678910:policy/accounting-SecretsManagerPolicy-1ULXG84GWOZ05 # Policy `SecretsManagerPolicy`
  CustomActions:
    OnNodeConfigured:
      Script: >-
        https://raw.githubusercontent.com/aws-samples/pcluster-manager/main/resources/scripts/multi-runner.py
      Args:
        - >-
          https://raw.githubusercontent.com/aws-samples/pcluster-manager/main/resources/scripts/slurm-accounting.sh
        - '-arn:aws:secretsmanager:us-east-2:1234567890:secret:DbPasswdSecret' # `SlurmDbPasswordSecretArn`
        - '-slurmdb-rds-instance.c123456.us-east-2.rds.amazonaws.com' # RDS Endpoint `SlurmDbEndpoint`
        - '-3306' # Default Port 3306
Scheduling:
  Scheduler: slurm
  SlurmQueues:
    - Name: queue0
      ComputeResources:
        - Name: queue0-t2-micro
          MinCount: 0
          MaxCount: 4
          InstanceType: t2.micro
      Networking:
        SubnetIds:
          - subnet-12345678910
Region: us-east-2
Image:
  Os: alinux2

Step 5 - Submit a job

Once the cluster has been successfully created, go to the Scheduling tab and select Submit Job

Choose a name for your job, a number of nodes to run under, choose to Run a command and provide a simple sleep 30 command.

Step 5 - View the Accounting Tab

Once you've submitted a job, you can see the job information under the Accounting tab

You can use any of the filters at the top to narrow down the number of jobs in the view to select specific jobs.

If you choose the Job ID in the left column you can see further detials about the job.