Autostop and Autodown

The autostop (or autodown) feature automatically stops (or tears down) a cluster after it becomes idle.

With autostop, users can simply submit jobs and leave their laptops, while ensuring no unnecessary spending occurs: after jobs have finished, the cluster(s) used will be automatically stopped (which can be restarted later).

With autodown, the cluster(s) used will be automatically torn down (i.e., terminated) instead.

To schedule autostop for a cluster, use sky autostop or sky launch -i <idle minutes>:

# Launch a cluster with logging detached (the -d flag)
sky launch -d -c mycluster cluster.yaml

# Autostop the cluster after 10 minutes of idleness
sky autostop mycluster -i 10

# Use the default, 5 minutes of idleness
# sky autostop mycluster

# (Equivalent to the above) Use the -i flag:
sky launch -d -c mycluster cluster.yaml -i 10

To schedule autodown for a cluster, pass the --down flag to either sky autostop or sky launch:

# Add the --down flag to schedule autodown instead of autostop.

# This means the cluster will be torn down after 10 minutes of idleness.
sky launch -d -c mycluster2 cluster.yaml -i 10 --down

# Or:
sky autostop mycluster2 -i 10 --down

Note

The autostop/autodown logic will be automatically executed by the remote cluster. Your local machine does not need to stay up for them to take effect.

To cancel any scheduled autostop/autodown on the cluster:

sky autostop mycluster --cancel

To view the status of the cluster, use sky status [--refresh]:

$ sky status
NAME         LAUNCHED    RESOURCES            STATUS   AUTOSTOP       COMMAND
mycluster    1 min ago   2x AWS(m4.2xlarge)   UP       10 min         sky launch -d -c ...
mycluster2   1 min ago   2x AWS(m4.2xlarge)   UP       10 min(down)   sky launch -d -c ...

# Refresh the statuses by querying the cloud providers
$ sky status --refresh
I 06-27 13:36:11 backend_utils.py:2273] Autodowned cluster: mycluster2
NAME         LAUNCHED    RESOURCES            STATUS   AUTOSTOP  COMMAND
mycluster    11 min ago  2x AWS(m4.2xlarge)   STOPPED  10 min    sky launch -d -c ...

Note that sky status shows the cached statuses, which can be outdated for clusters with autostop/autodown scheduled. To query the latest statuses of those clusters, use sky status --refresh.