How-to: stop and start a pod to control spend without losing its disks
This is a worked task for the most common cost-control flow. The reference
pages (rp doc pod stop, rp doc pod start, rp doc pod get) document the
individual flags; this page shows them composed into a pause-and-resume
cycle.
Goal: stop paying for a pod's GPU while keeping its disks and data, then bring it back exactly as it was.
Steps
- Find the pod. The verbs take the pod by id, so read it from the list first:
$ rp pod list
- Stop the pod. The GPU (or CPU) charge ceases; the disks are kept:
$ rp pod stop rp_abc123
Stopping is asynchronous — the command returns once the transition is accepted, not once the pod is stopped. Poll with:
$ rp pod get rp_abc123
- Start it again when you need it:
$ rp pod start rp_abc123
Starting is asynchronous too, and can fail later if the pod's GPU type is
out of stock in its datacentre — poll rp pod get and check stock with
rp stock gpu if the pod stays down.
Notes
- A stopped pod still bills for its storage; only the compute charge
ceases. Use
rp pod delete <id>to stop paying entirely. - A locked pod refuses to stop. Unlock it first with
rp pod update <id> --locked false. rp pod restart <id>recreates the container in one step — anything written outside/workspaceor a network volume is lost. A stop/start cycle preserves disk state; a restart does not promise to.- To tear everything down rather than pause, see Tear down endpoints, pods, and volumes in the right order.