Compute environments on Amazon ECS Managed Instances

AWS Batch compute environments on Amazon ECS Managed Instances use a managedInstancesProvider configuration block inside computeResources. This structure aligns with the Amazon ECS capacity provider model and is self-contained. All networking, instance profile, and instance selection configuration lives within this block.

The following list describes the compute environment parameters for Amazon ECS Managed Instances compute environments.

Updating Amazon ECS Managed Instances compute environments

All Amazon ECS Managed Instances compute environment attributes can be updated except the compute environment type, capacityOptionType, and fipsEnabled. Updates flow through to the underlying Amazon ECS capacity provider. New instances use the updated configuration. Existing instances continue running until their jobs complete and then drain naturally.

The following cannot be changed after creation:

Examples

Minimal configuration

The following example shows the minimum configuration required to create an Amazon ECS Managed Instances compute environment. Amazon ECS selects instance types automatically from all available types.

{
  "computeEnvironmentName": "my-managed-instances-ce",
  "type": "MANAGED",
  "state": "ENABLED",
  "computeResources": {
    "type": "ECS_MANAGED_INSTANCES",
    "maxvCpus": 256,
    "managedInstancesProvider": {
      "infrastructureRoleArn": "arn:aws:iam::123456789012:role/ecsInfrastructureRole",
      "instanceLaunchTemplate": {
        "ec2InstanceProfileArn": "arn:aws:iam::123456789012:instance-profile/ecsInstanceProfile",
        "networkConfiguration": {
          "subnets": ["subnet-abcde012", "subnet-bcde012a"],
          "securityGroups": ["sg-abcde012"]
        }
      }
    }
  }
}

GPU workload with specific instance types

The following example creates a compute environment constrained to NVIDIA GPU instance types for machine learning workloads.

{
  "computeEnvironmentName": "my-gpu-managed-instances-ce",
  "type": "MANAGED",
  "state": "ENABLED",
  "computeResources": {
    "type": "ECS_MANAGED_INSTANCES",
    "maxvCpus": 1000,
    "managedInstancesProvider": {
      "infrastructureRoleArn": "arn:aws:iam::123456789012:role/ecsInfrastructureRole",
      "instanceLaunchTemplate": {
        "ec2InstanceProfileArn": "arn:aws:iam::123456789012:instance-profile/ecsInstanceProfile",
        "networkConfiguration": {
          "subnets": ["subnet-abcde012", "subnet-bcde012a"],
          "securityGroups": ["sg-abcde012"]
        },
        "instanceRequirements": {
          "allowedInstanceTypes": ["g5.xlarge", "g5.2xlarge", "g5.4xlarge"]
        },
        "capacityOptionType": "ON_DEMAND"
      }
    }
  }
}

Spot capacity for cost-sensitive workloads

The following example creates a Spot-backed compute environment for fault-tolerant batch workloads.

{
  "computeEnvironmentName": "my-spot-managed-instances-ce",
  "type": "MANAGED",
  "state": "ENABLED",
  "computeResources": {
    "type": "ECS_MANAGED_INSTANCES",
    "maxvCpus": 5000,
    "managedInstancesProvider": {
      "infrastructureRoleArn": "arn:aws:iam::123456789012:role/ecsInfrastructureRole",
      "instanceLaunchTemplate": {
        "ec2InstanceProfileArn": "arn:aws:iam::123456789012:instance-profile/ecsInstanceProfile",
        "networkConfiguration": {
          "subnets": ["subnet-abcde012", "subnet-bcde012a", "subnet-cde012ab"],
          "securityGroups": ["sg-abcde012"]
        },
        "instanceRequirements": {
          "allowedInstanceTypes": ["m5.large", "m5.xlarge", "m6i.large", "m6i.xlarge"]
        },
        "capacityOptionType": "SPOT",
        "storageConfiguration": {
          "storageSizeGiB": 100
        }
      }
    }
  }
}