Registering the job definition

As a last step to configuring AWS Batch, we will register a job definition that will act as a template when we submit jobs.

Run the following to generate the configuration file that will be used to create the job definition:

export JOB_DEFINITION_NAME=RenderingJobDefinition

cat <<EoF > job-definition-config.json
{
    "jobDefinitionName": "${JOB_DEFINITION_NAME}",
    "type": "container",
    "containerProperties": {
        "image": "${IMAGE}",
        "vcpus": 1,
        "memory": 8000,
        "command": ["Ref::action", "-i", "Ref::inputUri", "-o", "Ref::outputUri", "-f", "Ref::framesPerJob"]
    },
    "retryStrategy": {
        "attempts": 3
    },
    "platformCapabilities": [
        "EC2"
    ]
}
EoF

Let’s explore the configuration parameters in the structure:

  • type: container is the default type and allows to run loosely coupled HPC workloads at scale. The other available type is multi-node. With AWS Batch multi-node you can run large-scale, tightly coupled, high performance computing applications. Note multi-node jobs are not supported with Spot instances. To learn more about multi-node jobs, visit multi-node parallel jobs.
  • image: the image used to start a container, this value is passed directly to the Docker daemon.
  • vcpus: The number of vCPUs reserved for the job. Each vCPU is equivalent to 1,024 CPU shares.
  • memory: hard limit (in MiB) for a container. If your container attempts to exceed the specified number, it’s terminated.
  • command: this is the command that will be executed in the container when the job is started. It has placeholders for some parameters that will be substituted when submitting the job using AWS Batch.
  • retryStrategy: this sections applies a retry strategy to your job definitions that allows failed jobs to be automatically retried. to learn more about these strategies, visit automated job retries.
  • platformCapabilities: the platform capabilities required by the job definition. Either EC2 or FARGATE.

The values of vcpus and memory have been defined based on the resources needed to render a specific file. Each Blender file can be different in this sense and those values should be adapted accordingly to prevent the container from running out of memory when executing Blender.

Execute this command to create the job definition. To learn more about this API, see register-job-definition CLI command reference.

export JOB_DEFINITION_ARN=$(aws batch register-job-definition --cli-input-json file://job-definition-config.json | jq -r '.jobDefinitionArn')
echo "Job definition Arn: ${JOB_DEFINITION_ARN}"

Finally, you are going to submit a job request.

Site Info