New to KubeDB? Please start here.

Hazelcast Vertical Scaling Overview

This guide will give you an overview on how KubeDB Ops Manager updates the resources(CPU and Memory) of the Hazelcast database.

Before You Begin

• You should be familiar with the following KubeDB concepts:
  • Hazelcast
  • HazelcastOpsRequest

How Vertical Scaling Process Works

The following diagram shows how KubeDB Ops Manager updates the resources of the Hazelcast database. Open the image in a new tab to see the enlarged version.

  

The updating process consists of the following steps:

1. At first, a user creates a Hazelcast Custom Resource (CR).

2. KubeDB Community operator watches the Hazelcast CR.

3. When the operator finds a Hazelcast CR, it creates required number of StatefulSets and related necessary stuff like appbinding, services, etc.

4. Then, in order to update the resources of the Hazelcast database the user creates a HazelcastOpsRequest CR with the desired resources.

5. KubeDB Enterprise operator watches the HazelcastOpsRequest CR.

6. When it finds a HazelcastOpsRequest CR, it halts the Hazelcast object which is referred from the HazelcastOpsRequest. So, the KubeDB Community operator doesn’t perform any operations on the Hazelcast object during the updating process.

7. Then the KubeDB Enterprise operator will update the resources of the related PetSets to match the desired resources defined in the HazelcastOpsRequest CR.

8. After the successful update of the resources of the StatefulSet’s replicas, the KubeDB Enterprise operator updates the Hazelcast object to reflect the updated state.

9. After successfully updating of Hazelcast object, the KubeDB Enterprise operator resumes the Hazelcast object so that the KubeDB Community operator can resume its usual operations.

Vertical Scaling Modes

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

• 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 vertical scaling of a Hazelcast database using vertical scaling operation.