Run portable jobs in containers

Write job templates for the application interface that the workload needs. Don't write them for a specific execution technology. An environment can provide that interface in several ways. Examples include software installed on the worker, a conda or Rez environment, a Docker container, an Apptainer container, or another runtime.

Open Job Description (OpenJD) wrap actions separate what a task runs from where and how it runs. You can run the same job template locally without a wrapping environment. You can also attach a wrapping queue environment to run its actions in a container on Deadline Cloud. The job template doesn't contain Docker commands or identify a container image.

To set up container support on Deadline Cloud, you configure three layers:

Layer	Responsibility	Configured in
Fleet	Provides rootless Docker and optional GPU integration on workers.	A Linux service-managed fleet.
Queue environment	Starts a session container and wraps environment and task actions.	The queue that receives container jobs.
Job template	Defines the application command, arguments, parameters, and files.	A portable OpenJD job bundle.

The chapter includes the following topics:

• Set up container support in the console prepares a fleet and queue.

• Building, tagging, and pushing a container image creates a versioned image in Amazon ECR.

• Docker queue environment overview explains how wrap actions run the job inside the image.

• Run a portable job locally and in a container runs one job template locally and on Deadline Cloud.

• Share data between the worker session and container compares session and persistent storage.

• Security and best practices for container jobs describes image, runtime, and command construction practices.