View a markdown version of this page

Create a job definition on Amazon ECS Managed Instances - AWS Batch
Creating a job definition on Amazon ECS Managed Instances (console)Creating a job definition on Amazon ECS Managed Instances (AWS CLI)

Create a job definition on Amazon ECS Managed Instances

Amazon ECS Managed Instances job definitions use ecsProperties with the MANAGED_INSTANCES platform capability. Unlike Fargate, there are no restrictions on vCPU and memory size combinations, and GPU resources are supported. For more information about supported parameters, see Job definitions on Amazon ECS Managed Instances.

Creating a job definition on Amazon ECS Managed Instances (console)

To create a new job definition on Amazon ECS Managed Instances:

  1. Open the AWS Batch console at https://console.aws.amazon.com/batch/.

  2. From the top navigation bar, choose the AWS Region to use.

  3. In the left navigation pane, choose Job definitions.

  4. Choose Create.

  5. For Orchestration type, choose ECS Managed Instances.

  6. For Name, enter a unique name for your job definition. The name can be up to 128 characters in length. It can contain uppercase and lowercase letters, numbers, hyphens (-), and underscores (_).

  7. (Optional) For Execution timeout, enter the timeout value (in seconds). The execution timeout is the length of time before an unfinished job is terminated. If an attempt exceeds the timeout duration, the attempt is stopped and moves to a FAILED status. For more information, see Job timeouts. The minimum value is 60 seconds.

  8. For Execution role, choose an IAM role that grants the Amazon ECS container agent permission to make AWS API calls on your behalf. This role is required for Amazon ECS Managed Instances jobs. For example, the role is used to pull container images from Amazon ECR.

  9. In the Container configuration section:

    1. For Image, enter the container image to use for the job. Images in Amazon ECR registries can be specified with the repository-url/image:tag naming convention. Images in public registries use the full URI (for example, public.ecr.aws/amazonlinux/amazonlinux:2023).

    2. For Command, enter the command to pass to the container.

    3. For vCPUs, specify the number of vCPUs to reserve for the container.

    4. For Memory, specify the memory limit (in MiB) available to the container.

    5. (Optional) For GPUs, specify the number of GPU devices to make available to the container. The compute environment must have GPU instances available.

  10. (Optional) Expand Tags, and then choose Add tag to add tags to the resource. Turn on Propagate tags to propagate tags from the job and job definition.

  11. Choose Create job definition.

Creating a job definition on Amazon ECS Managed Instances (AWS CLI)

Use the register-job-definition command to create a job definition for Amazon ECS Managed Instances.

Basic example

The following example creates a minimal job definition that runs a simple command.

$ aws batch register-job-definition \
    --job-definition-name my-managed-instances-job-def \
    --type container \
    --platform-capabilities MANAGED_INSTANCES \
    --ecs-properties '{
      "taskProperties": [{
        "containers": [{
          "image": "public.ecr.aws/amazonlinux/amazonlinux:2023",
          "name": "main",
          "command": ["echo", "hello managed instances"],
          "resourceRequirements": [
            {"type": "VCPU", "value": "1"},
            {"type": "MEMORY", "value": "1024"}
          ]
        }],
        "executionRoleArn": "arn:aws:iam::123456789012:role/ecsTaskExecutionRole"
      }]
    }'

GPU job definition

The following example creates a job definition that requests GPU resources.

$ aws batch register-job-definition \
    --job-definition-name my-gpu-managed-instances-job-def \
    --type container \
    --platform-capabilities MANAGED_INSTANCES \
    --ecs-properties '{
      "taskProperties": [{
        "containers": [{
          "image": "123456789012.dkr.ecr.us-east-1.amazonaws.com/my-gpu-image:latest",
          "name": "main",
          "command": ["nvidia-smi"],
          "resourceRequirements": [
            {"type": "VCPU", "value": "4"},
            {"type": "MEMORY", "value": "16384"},
            {"type": "GPU", "value": "1"}
          ]
        }],
        "executionRoleArn": "arn:aws:iam::123456789012:role/ecsTaskExecutionRole"
      }]
    }'

Multi-container (sidecar) job definition

The following example creates a job definition with a main container and a sidecar logging container.

$ aws batch register-job-definition \
    --job-definition-name my-sidecar-managed-instances-job-def \
    --type container \
    --platform-capabilities MANAGED_INSTANCES \
    --ecs-properties '{
      "taskProperties": [{
        "containers": [{
          "image": "public.ecr.aws/amazonlinux/amazonlinux:2023",
          "name": "main",
          "command": ["echo", "processing data"],
          "essential": true,
          "resourceRequirements": [
            {"type": "VCPU", "value": "2"},
            {"type": "MEMORY", "value": "4096"}
          ]
        },
        {
          "image": "public.ecr.aws/amazonlinux/amazonlinux:2023",
          "name": "sidecar",
          "command": ["echo", "logging sidecar"],
          "essential": false,
          "resourceRequirements": [
            {"type": "VCPU", "value": "1"},
            {"type": "MEMORY", "value": "512"}
          ]
        }],
        "executionRoleArn": "arn:aws:iam::123456789012:role/ecsTaskExecutionRole"
      }]
    }'