Processing Deadline Cloud jobs

When a job enters a queue, Deadline Cloud schedules it on one or more fleets associated with the queues. The fleet is chosen based on the capabilities configured for the fleet and the host requirements of a specific step. If a job has a requirement that can't be met by a any of the fleets associated with the queue, the job's status is set to "Not compatible" and the rest of the steps in the job are canceled.

Next, Deadline Cloud sends instructions to the workers to set up a session for the step. The software required for the step must be available on the worker instance for the job to run. The service opens sessions on multiple workers if the fleets scaling settings allow.

You can set up the software in an Amazon Machine Image (AMI), or your worker can load the software at runtime from a repository or package manager. You can use queue, job, or step environments to deploy the software that you prefer.

The Deadline Cloud service uses the OpenJD template to identify the steps required for the job, and the tasks required for each step. Some steps have dependencies on other steps, so Deadline Cloud determines the order to complete the steps. Then, Deadline Cloud sends the tasks for each step to workers to process. When a task is finished, the service sends another task in the same session, or the worker can start a new session.

After all tasks in each step are finished, the job is complete and the output is ready to download to your workstation. Even if the job didn't finish, the output from each step and task that finished is available to download.