View a markdown version of this page

Job definitions on Amazon ECS Managed Instances - AWS Batch

Job definitions on Amazon ECS Managed Instances

AWS Batch jobs on Amazon ECS Managed Instances use the ecsProperties job definition format with a platform capability of MANAGED_INSTANCES. Amazon ECS Managed Instances jobs do not support legacy containerProperties.

The following list describes job definition parameters that are specific to or restricted for Amazon ECS Managed Instances jobs.

platformCapabilities

Must be specified as MANAGED_INSTANCES.

"platformCapabilities": [ "MANAGED_INSTANCES" ]
type

Must be specified as container.

"type": "container"
Parameters in ecsProperties
networkMode

Controls how the task's network is configured. This parameter only applies to Amazon ECS Managed Instances job definitions. This parameter does not apply to Fargate or Amazon EC2 platform job definitions.

Valid values: host

When not specified, the default is host. With host mode, the container shares the host instance's network stack directly, providing maximum network bandwidth.

"networkMode": "host"
runtimePlatform

Specifies the operating system family and CPU architecture for the task. Use this to run jobs on ARM64 (Graviton) instances. The valid value for operatingSystemFamily is LINUX (default). The valid values for cpuArchitecture are X86_64 and ARM64.

"runtimePlatform": {
  "operatingSystemFamily": "LINUX",
  "cpuArchitecture": "ARM64"
}
executionRoleArn

The IAM role that grants the Amazon ECS container agent permission to make AWS API calls on your behalf, such as pulling container images from Amazon ECR. For more information, see IAM Roles for Tasks in the Amazon Elastic Container Service Developer Guide.

resourceRequirements

Both memory and vCPU requirements must be specified in resourceRequirements. Unlike Fargate, Amazon ECS Managed Instances does not restrict which vCPU and memory combinations are valid — you can specify any values that fit within the instance types available to your compute environment. GPU resources are supported.

"resourceRequirements": [
  {"type": "MEMORY", "value": "8192"},
  {"type": "VCPU", "value": "4"},
  {"type": "GPU", "value": "1"}
]
Parameters supported on Amazon ECS Managed Instances but not Fargate

The following parameters are available for Amazon ECS Managed Instances jobs but are blocked for Fargate:

  • privileged — run the container with elevated privileges on the host instance.

  • resourceRequirements type GPU — request NVIDIA GPU devices.

  • linuxParameters.devices — expose host devices to the container.

  • linuxParameters.tmpfs — mount RAM-backed tmpfs filesystems.

  • ulimits — set resource limits (nofile, nproc, etc.) on the container.

  • volumes with host.sourcePath — bind mount host paths into the container.

Parameters not supported on Amazon ECS Managed Instances
fargatePlatformConfiguration

Not applicable for Amazon ECS Managed Instances jobs.

Multi-node parallel (MNP) jobs

Amazon ECS Managed Instances does not support multi-node parallel jobs. Use Amazon EC2 managed compute environments for MNP workloads.

Parameters in logConfiguration
logDriver

Only awslogs, splunk, and awsfirelens are supported for Amazon ECS Managed Instances jobs.