Storage

Does CAST have a way to configure root volume size when using a node template?

Root volume is currently only accessible in node configurations. Currently, this would require a dedicated node configuration and node template.

We tried to simulate the scenario where Template A linked to Node config A. By default Node config A has 100 + 10 CPU per core disk configuration. The pending pod A has a selector for Template A and requests just CPU/MEM - it gets 2 CPU node with 120 GiB disk. The pending pod B has a selector for Template A and in the requests also has a request for ephemeral storage:

resources:  
          requests:  
            cpu: '1'  
            ephemeral-storage: "100Gi"

This pod gets a 2 CPU node with 213 GiB disk.

Is it possible to define zones and disk types when creating a node template? Is there an API option available for this purpose?

Node templates don’t offer this option at the moment. However, you can create a custom node config and link it to the template. The node config can have a subset of the cluster’s zones.

A side option would be to add a zone selector for the workloads:

nodeSelector:  
   topology.kubernetes.io/zone: "some-zone"

Regarding disk types, if you'd like to use storage optimized resources (local SSDs), you can select storage optimized instances only via instance constraints in the template. For more details, see Storage-optimized nodes.

How does Cast AI manage local disks on storage-optimized nodes?

Cast AI uses LVM to pool locally attached NVMe/SSD disks into a single volume on AWS (EKS) and Azure (AKS). On GCP (GKE), the local SSD mounts are managed natively by GKE. On AWS, RAID-0 is also supported for specific configurations. Kubernetes data directories (kubelet, container runtime) are placed on the local storage volume for improved I/O performance.

For a detailed breakdown of how each cloud provider handles local storage, see Storage-optimized nodes.

How does Cast AI enforce "ephemeral storage in order to limit users storage on nodes" when consolidating?

Cast AI respects Kubernetes storage requests:

ephemeral-storage: "100Gi"

This means that Cast AI will find a node with enough local storage to fit pods that have requirements defined. Read more about it in Storage-optimized nodes and Pod placement.

What is the default volume size for EBS?

The default volume size is 100GiB + CPU to Storage ratio, to increase it further for every CPU provisioned on the node.