Pause, resume, restart and delete

Four commands change whether a deployed service runs without changing its descriptor. None of them ever removes the service's data: entities, workflows, sessions, view rows and timers live in the service's database, and the database outlives all four.

Command	Instances	Generation	Hostname	Data
ankka services pause <name>	stopped	unchanged	kept	kept
ankka services resume <name>	started again	unchanged	kept	kept
ankka services restart <name>	replaced one at a time	incremented	kept	kept
ankka services delete <name>	removed	unchanged	removed	kept

Each is recorded in the service's history with who asked; see Status and history.

Pause and resume

ankka services pause cart
ankka services resume cart

Pausing stops every instance of a service and keeps everything else: its descriptor, its database, and whether it is exposed. The service reports Paused with zero desired instances. While it is paused nothing answers its in-cluster address or its hostname, and its timers wait; a timer that fell due while the service was paused fires once it is running again.

Resuming starts the instances the descriptor asks for. The service reports UpdateInProgress until they have joined their cluster and are ready, then Ready. Entities rebuild from the journal as they are first used.

Pausing is the way to take a service down. Scaling its Kubernetes Deployment to zero by hand does not work: the operator restores the instance count from the descriptor on its next pass.

Restart

ankka services restart cart

Restarting replaces every instance without any change to the descriptor, one instance at a time: a new instance starts, joins the cluster and takes over entities before an old one is stopped. The service keeps answering throughout, including on its hostname. The generation increments, so the status tracks the new rollout and ignores reports about the old one.

Restart is for picking up something outside the descriptor that the instances read only at start, such as a changed secret, or for clearing a problem in a running process. An apply that changes the image or the environment rolls the instances by itself; an apply that changes only the instance count adds or removes instances without replacing the others. See Scale and roll out.

Delete

ankka services delete cart

Deleting a service removes its instances, its in-cluster address and its route, and the service reports NotDeployed. Its database is not deleted. Nothing the platform does can delete a service's database; the operator is not permitted to by the cluster itself.

Applying a descriptor with the same name in the same project brings the service back with its data:

ankka services apply -f service.json
ankka services get cart
# ...
# database    recovered existing data

The generation continues from where it was, so the history is continuous. A re-created service is private until it is exposed again, because its route was removed with it.

A service name is a deployment target, so reusing one is intended. Organization and project ids are not reusable in the same way; see Tenancy and access.

Paused and suspended

A service can also be stopped by a platform administrator disabling its organization. It then reports Suspended, not Paused, so it is clear that the members did not stop it, and its members cannot resume it: every change in a disabled organization is refused. When the organization is enabled again, each service returns to what its members had chosen. A service they had paused stays Paused; everything else starts again.