gke-compute-class-creator

A ComputeClass resource controls GKE autoscaling by letting you define an ordered set of priorities for workload placement. You specify hardware constraints like GPUs or Spot VMs to manage node pool scaling and configuration.

A ComputeClass in GKE is a custom resource definition that manages node configurations and autoscaling. You need it when deploying cost-optimized Spot VM clusters, performance-driven workloads, or AI/ML scenarios requiring specific GPUs and TPUs.

Yes, nodePoolAutoCreation and activeMigration are supported fields in the ComputeClass CRD specification. You can include them in your manifest to automatically create node pools and migrate workloads when applying the configuration via kubectl.

To create a GKE manifest for AI/ML workloads requiring GPUs, define a ComputeClass resource with optional GPU and TPU fields. Specify location constraints and hardware priorities to generate a valid manifest ready for kubectl apply.

The best way to optimize GKE cluster costs using Spot VMs is defining a ComputeClass manifest with cost-optimized autoscaling priorities. This declarative approach targets Spot VMs for workload placement, reducing overall compute expenses.