Create a ParallelCluster via AWS CLI

Create the cluster

pcluster create-cluster -c <pcluster-create.yml> -n <cluster-name> -r <region>

Checklist for `pcluster-create.yml`

Minimum required sections
- Region, Image, HeadNode, and Scheduling (with SlurmQueues).
SSH access
- Head node security group allows inbound TCP 22 from your IP (or VPN / bastion).
- HeadNode.Ssh.KeyName matches an existing EC2 key pair in the region.
Networking
- Head node subnet, compute subnets, and any FSx file system are in the same VPC.
- Security groups referenced in the config belong to the same VPC.
Slurm queues
- Each queue in SlurmQueues has a unique Name.
- Each queue includes CapacityType, Networking.SubnetIds, and ComputeResources.
- Each compute resource specifies InstanceType and integer MinCount / MaxCount.
IAM note (common failure point)
- The AWS identity used to run pcluster create-cluster must have permission to create and manage required resources (EC2, CloudFormation, IAM) and to pass IAM roles.
- You can verify the active identity with:
```
aws sts get-caller-identity
```

Example `pcluster-create.yml`

An example pcluster-create.yml:

Region: us-east-1

Image:
  Os: ubuntu2404
  CustomAmi: ami-08ea421ec9fad0a09

HeadNode:
  InstanceType: c5.large
  Ssh:
    KeyName: dzhang  # <-- change to your keypair name
  Networking:
    SubnetId: subnet-08895ae58a2f9167d
    AdditionalSecurityGroups:
      - sg-0f504cd36a5dc0b34

Scheduling:
  Scheduler: slurm
  SlurmSettings:
    QueueUpdateStrategy: DRAIN
  SlurmQueues:
    - Name: ondemand
      CapacityType: ONDEMAND
      Networking:
        SubnetIds:
          - subnet-08895ae58a2f9167d
        AdditionalSecurityGroups:
          - sg-0f504cd36a5dc0b34
      ComputeResources:
        - Name: c8a12xl
          InstanceType: c8a.12xlarge
          MinCount: 0
          MaxCount: 20

    - Name: spot
      CapacityType: SPOT
      Networking:
        SubnetIds:
          - subnet-08895ae58a2f9167d
        AdditionalSecurityGroups:
          - sg-0f504cd36a5dc0b34
      ComputeResources:
        - Name: c8a12xl
          InstanceType: c8a.12xlarge
          MinCount: 0
          MaxCount: 50

SharedStorage:
  - Name: fsx_input
    StorageType: FsxLustre
    MountDir: /fsx_input
    FsxLustreSettings:
      FileSystemId: fs-009ecfd7e44882657

Monitor the creation process

pcluster describe-cluster -n <cluster-name> -r <region>

clusterStatus:

CREATE_IN_PROGRESS → normal
CREATE_COMPLETE → success
CREATE_FAILED → stop and debug

Delete ParallelCluster

Delete by cluster name

pcluster delete-cluster -n <cluster-name> -r <region>

Find the cluster name (if forgotten)

If the head node still exists, you can infer the cluster name from the head node IAM role in the EC2 console: the string before -RoleHeadNode is typically the cluster name.

If the head node has already been deleted, list clusters in the region:

pcluster list-clusters -r <region>

Then inspect a specific cluster:

pcluster describe-cluster -n <cluster-name> -r <region>

Recover the configuration YAML used to create a cluster

From the output of pcluster describe-cluster, look for:

"clusterConfiguration": {
  "url": "https://parallelcluster-.../clusters/<cluster-name>/configs/cluster-config.yaml?versionId=..."
}

Download the configuration file:

curl -o cluster-config.yaml "<that-URL>"