Skip to main content

Manage Kubernetes compute pools

Add, edit, and delete compute pools on an existing Kubernetes cluster.

Manage Kubernetes compute pools

Add, edit, and delete compute pools on an existing Kubernetes cluster.

TimeAbout 5 minutes per operation
You will needA Kubernetes cluster in ready status
OutcomeCompute pools configured for your workloads

This guide walks through adding, editing, and deleting compute pools on an existing Kubernetes cluster. Compute pools are groups of nodes with a shared instance type, role, and autoscaling range.

Add a compute pool​

Open the cluster list​

Click Clusters in the left sidebar, then click the Kubernetes tab.

Open the cluster detail page​

Click the cluster name to open the detail page.

Open the Compute Pools tab​

Click the Compute Pools tab.

Add a pool​

Click Add Compute Pool.

Configure the pool​

Configure the pool:

  • Role: control (cluster management) or worker (compute workloads)
  • Instance types (AWS): click Select Instance to browse EC2 types. You can select multiple types for Fleet diversification; set the Allocation strategy to lowest-price, diversified, or capacity-optimized.
  • Profile (Azure/GCP/on-premises): choose Small (4 vCPU, 8 GiB), Medium (8 vCPU, 16 GiB), or Large (16 vCPU, 32 GiB).
  • Min size: minimum node count (set to 0 for scale-to-zero)
  • Max size: upper bound for autoscaling

Create​

Click Create.

Success looks like this: the compute pool is created or updated, and the cluster's scheduler begins routing workloads to it.

Edit a compute pool​

Open the edit form​

On the Compute Pools tab, click the edit icon on the pool row.

Modify settings​

Modify the min/max sizes, labels, or taints.

Save​

Click Save.

Add labels​

Labels are key-value pairs used for workload scheduling. Common patterns:

  • workload-type: gpu -- schedule GPU jobs to this pool
  • team: ml-training -- restrict a pool to a specific team
  • priority: high -- mark pools for priority workloads

Vantage auto-injects the label vc.pool: <pool-name> on every node.

Add taints​

Taints prevent pods from scheduling on a pool unless they have a matching toleration. Use taints to reserve pools for specific workloads (for example, GPU-only or high-memory jobs).

Scale to zero​

Set Min size to 0 on a worker pool. Nodes are terminated when no pods are scheduled, and new nodes are provisioned when demand resumes. This is ideal for bursty workloads.

tip

Scale-to-zero pools take 2--5 minutes to provision a new node when demand arrives. For latency-sensitive workloads, keep min >= 1.

Delete a compute pool​

Delete​

Click the delete icon on the pool row.

Confirm​

Confirm the deletion.

warning

Deleting a compute pool terminates all nodes in the pool. Drain workloads first if the pool has active pods.

For field definitions and autoscaling details, see the Compute pools reference.

Ask AI
Ask a question about Vantage Compute...