Manage GPU instances

Each account has one current instance, and every rcc instance command acts on it. There’s no instance ID to pass. To work with another account’s instance, switch accounts with -A (see Sign-in and accounts).

Check credits and available resources

rcc status           # account, credits and instance at a glance
rcc credits status   # credit balance
rcc catalog list     # environments, resource pools and templates you may use

Filter by kind, or look up and search for an entry:

rcc catalog list --kind environment
rcc catalog list --kind resource_pool
rcc catalog list --kind resource_template
rcc catalog show ID
rcc catalog search TEXT

Being listed means your account may request the resource. It does not reserve a GPU or promise free capacity right now.

Add an SSH key

Before your first instance with SSH, register an SSH key for the current account:

rcc instance add-key

The CLI checks the key you already have and, if there is none, offers to generate an Ed25519 key. Only the public key is uploaded.

To pick a specific existing key:

rcc instance add-key --identity-file ~/.ssh/id_ed25519

SSH is on by default when you create an instance; add --no-ssh if you don’t need it.

Create an instance

Run it in a terminal and the CLI walks you through choosing an environment and resources:

rcc instance create

In a script, or anywhere without an interactive terminal, spell the choices out. The values come from rcc catalog list:

rcc instance create \
  --env ENVIRONMENT \
  --pod-type RESOURCE_POOL \
  --resource-template RESOURCE_TEMPLATE

Option	Value
--env	the ID, name or image of an environment entry
--pod-type	a resource_pool entry
--resource-template	a resource_template entry

By default the instance gets a temporary workspace that is deleted with it. To keep /workspace after the instance is deleted, add --use-pvc. See Keep your data.

The CLI creates notebook instances. To serve a model, use the console. See Serve a model.

By default the CLI waits until the instance is ready. Add --no-wait to return as soon as the request is accepted, then follow the progress:

rcc instance create \
  --env ENVIRONMENT \
  --pod-type RESOURCE_POOL \
  --resource-template RESOURCE_TEMPLATE \
  --no-wait

rcc instance status --watch
rcc instance logs

If a request went out but no response came back, retry with the same --idempotency-key KEY so it is not created twice. Never reuse a key for a different create request.

A timeout or Ctrl+C does not cancel the creation

When the local wait times out or you press Ctrl+C, the CLI only stops waiting; the service keeps creating the instance. Carry on with rcc instance status --watch and rcc instance logs.

Open JupyterLab or connect over SSH

rcc instance open   # open JupyterLab in your browser
rcc instance ssh    # connect over SSH

On a machine without a browser, print the link or the SSH command instead:

rcc instance open --print
rcc instance ssh --print

The JupyterLab link works once and expires quickly. Keep it out of logs, tickets and chat.

Status and logs

rcc instance list             # the current instance (--all: one per saved account)
rcc instance status           # its current state
rcc instance status --watch   # keep refreshing, every 3 seconds by default
rcc instance logs             # startup events and container logs

You can only see the instances of the account you are signed in as.

Delete an instance

rcc instance delete --yes
rcc instance status

Right after deletion the status may say the instance is shutting down. That means deletion has started, not that cleanup is finished. Keep checking until no running instance is left. Signing out does not delete an instance.