View a markdown version of this page

Worker states and lifecycle - Deadline Cloud

Worker states and lifecycle

A worker is a host in a fleet that runs the tasks that make up your jobs. Each worker runs the Deadline Cloud worker agent, which connects the host to your farm. Every worker has a status, on both service-managed and customer-managed fleets. You can see the status in the worker status column of the Deadline Cloud monitor and in the status field that the GetWorker and ListWorkers operations return. The status shows whether the worker is available for work, shutting down, or no longer connected. It also determines whether Deadline Cloud charges for the worker.

Worker statuses

The following table lists the worker statuses in lifecycle order.

Status Description

CREATED

The worker agent registered the worker, and Deadline Cloud acknowledged the registration. The worker isn't ready for work yet.

STARTED

The worker is connected and ready to receive tasks.

IDLE

The worker is connected, and Deadline Cloud has no tasks assigned to it.

RUNNING

The worker is connected, and Deadline Cloud has assigned tasks to it.

STOPPING

The worker is shutting down and doesn't accept new tasks.

STOPPED

The worker isn't running tasks, won't receive new tasks, and no longer sends heartbeats. Deadline Cloud still knows about the worker. The host can be shut down.

NOT_RESPONDING

Deadline Cloud stopped receiving heartbeats from the worker, for example after a lost connection or a host failure.

NOT_COMPATIBLE

The worker's capabilities don't meet the worker requirements of the fleet.

The status determines whether Deadline Cloud charges for the worker. For the statuses that incur a charge, see Charges for workers in the Deadline Cloud User Guide.

What changes a worker's status

The following diagram shows how a worker that runs the Deadline Cloud worker agent moves between statuses. The changes are the same on service-managed and customer-managed fleets. If you write your own worker agent, follow the same changes.

A new worker is registered with the CREATED status (1) and changes to STARTED when it's ready for work (2). While connected, it moves between IDLE and RUNNING as tasks are assigned (3). A connected worker changes to STOPPING when it starts to shut down (4), to STOPPED when it finishes shutting down (5), to NOT_RESPONDING when it stops sending heartbeats (6), and to NOT_COMPATIBLE when it no longer meets the fleet's requirements (7). A STOPPING or NOT_COMPATIBLE worker also changes to STOPPED (5), and a CREATED worker that stops sending heartbeats changes to NOT_RESPONDING (6). A STOPPED or NOT_RESPONDING worker returns to STARTED when the worker agent connects again (8, a dashed arrow). The diagram shows a STOPPED worker being deleted (9), which is the usual path. You can also delete a CREATED worker. Each number matches a row in the table that follows the diagram, which describes the cause of each change.

State diagram of worker statuses with numbered arrows. Each number matches a row in the following table.

In the diagram, the connected area holds STARTED, IDLE, and RUNNING. The not-connected area holds STOPPED, NOT_RESPONDING, and NOT_COMPATIBLE. NOT_RESPONDING and NOT_COMPATIBLE indicate a problem. CREATED, STOPPING, and the deleted state sit outside both areas.

The following table lists each change and its cause.

Arrow From To What causes the change

1

(none)

CREATED

The worker agent registers the worker with the CreateWorker operation.

2

CREATED

STARTED

The worker agent obtains its credentials and reports that it's ready for work.

3

STARTED, IDLE, or RUNNING

IDLE or RUNNING

Deadline Cloud sets RUNNING when it assigns tasks to the worker and IDLE when no tasks are assigned.

4

STARTED, IDLE, or RUNNING

STOPPING

The worker starts to shut down. Deadline Cloud asks the worker to stop, for example when the fleet scales in, or the worker agent starts a drain after an operating system stop signal or an Amazon EC2 Spot interruption. The worker stops accepting new tasks.

5

STARTED, IDLE, RUNNING, STOPPING, or NOT_COMPATIBLE

STOPPED

The worker agent finishes shutting down and reports the new status with the UpdateWorker operation. The worker agent also sets STOPPED when a host configuration script fails, and before it restarts a NOT_COMPATIBLE worker. You can also set the status to STOPPED yourself with UpdateWorker, for example to stop a stuck worker.

6

CREATED, STARTED, IDLE, or RUNNING

NOT_RESPONDING

Deadline Cloud stops receiving heartbeats from the worker agent.

7

STARTED, IDLE, or RUNNING

NOT_COMPATIBLE

You update the fleet, and the worker's capabilities no longer meet the fleet's worker requirements.

8

STOPPED or NOT_RESPONDING

STARTED

The worker agent connects again, for example after you restart the agent or the host. A NOT_COMPATIBLE worker must change to STOPPED before it can change to STARTED.

9

CREATED or STOPPED

Deleted

You or the worker agent delete the worker with the DeleteWorker operation. Deadline Cloud no longer tracks a deleted worker.

For information about shutting down Amazon Elastic Compute Cloud (Amazon EC2) instances when their workers stop, see Configure the Deadline Cloud worker agent.