New to KubeDB? Please start here.

Vertical Scaling Weaviate

This guide will give you an overview of how KubeDB Ops Manager updates the resources (for example Memory, CPU etc.) of a Weaviate cluster.

Before You Begin

• You should be familiar with the following KubeDB concepts:
  • Weaviate
  • Weaviate Quickstart

How Vertical Scaling Process Works

The vertical scaling process consists of the following steps:

1. At first, a user creates a Weaviate CR.

2. KubeDB provisioner operator watches for the Weaviate CR.

3. When the operator finds a Weaviate CR, it creates a PetSet and related necessary resources like secret, service, etc.

4. Then, in order to update the resources (for example CPU, Memory etc.) of the Weaviate cluster, the user creates a WeaviateOpsRequest CR with the desired resources.

5. KubeDB Ops Manager watches for the WeaviateOpsRequest CR.

6. When it finds one, it halts the Weaviate object so that the KubeDB provisioner operator doesn’t perform any operation on the Weaviate during the scaling process.

7. Then the KubeDB Ops-manager operator updates the resources of the PetSet’s Pods to reach the desired state, restarting the pods one at a time.

8. After successfully updating the resources of the PetSet’s Pods, the KubeDB Ops Manager updates the Weaviate object resources to reflect the updated state.

9. After successfully updating the Weaviate resources, the KubeDB Ops Manager resumes the Weaviate object so that the KubeDB Provisioner operator resumes its usual operations.

Vertical Scaling Modes

KubeDB actuates vertical scaling in one of two modes, selected through the spec.verticalScaling.mode field of the WeaviateOpsRequest:

• Restart (default): The operator patches the PetSet with the new resources and restarts the Pods (one at a time, honoring the database’s failover rules) so they come back with the updated CPU and Memory. This works on every Kubernetes cluster.
• InPlace: The operator resizes the running containers in place using the Kubernetes in-place Pod resize (pods/resize subresource) — no Pod restart, so scaling happens without downtime or failover. If a Node cannot accommodate the new resources (the resize is reported Infeasible), the operator automatically falls back to the Restart behavior for that Pod.

If spec.verticalScaling.mode is omitted, it defaults to Restart.

Note: InPlace mode relies on the Kubernetes InPlacePodVerticalScaling feature gate, which is enabled by default from Kubernetes v1.33. On older clusters, or when the feature gate is disabled, use Restart mode.

In the next doc, we are going to show a step-by-step guide on updating the resources of a Weaviate database using the vertical scaling operation.