Is time-sharing GPU in GCP supported by CAST AI? I have a use case to run Hibernate, and pause and resume CronJob to a particular namespace instead of all namespace.

Overriding default NAMESPACES_TO_KEEP will exclude commands from hibernation. You can add there elements that you want to stay, and everything else will go down.

Set the NAMESPACES_TO_KEEP environment variable to override "opa,istio", learn more here.

If you want to shut down particular workloads/namespace (scale to 0 pods), then it's recommended that the OSS project achieves this kube-downscaler. Make sure to enable Evictor in the aggressive mode in case you have enabled Kube-downscaler - this will ensure that your cluster shrinks rapidly after Kube-downscaler removes pods.