New to KubeDB? Please start here.

Weaviate Compute Resource Autoscaling

This guide will give you an overview of how KubeDB autoscales the compute resources (CPU and Memory) of a Weaviate cluster using a WeaviateAutoscaler.

Before You Begin

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

How Compute Autoscaling Works

KubeDB provides a WeaviateAutoscaler CRD to automatically scale the compute resources of a Weaviate cluster. It is backed by a VerticalPodAutoscaler (VPA) that observes the actual resource usage of the Weaviate pods (requires metrics-server).

The compute autoscaling process consists of the following steps:

1. The user creates a WeaviateAutoscaler CR with a spec.compute.weaviate block describing the trigger, the min/max allowed resources, and the controlled resources.

2. The KubeDB Autoscaler operator creates a VerticalPodAutoscaler for the cluster and watches the recommendations it produces.

3. When the recommended resources differ from the current resources by more than resourceDiffPercentage (and the pods are older than podLifeTimeThreshold), the Autoscaler operator creates a WeaviateOpsRequest of type VerticalScaling.

4. The KubeDB Ops Manager applies the VerticalScaling ops request, updating the pod resources within the minAllowed/maxAllowed bounds.

The relevant fields under spec.compute.weaviate are:

• trigger — On or Off, enables/disables compute autoscaling.
• podLifeTimeThreshold — the minimum age of a Pod before a recommendation can be applied.
• resourceDiffPercentage — the minimum percentage change required before a new recommendation is applied.
• minAllowed / maxAllowed — the lower and upper bounds for the autoscaled resources.
• controlledResources — the resource types to autoscale (e.g. cpu, memory).
• containerControlledValues — whether to control RequestsAndLimits or just Requests.

In the next doc, we are going to show a step-by-step guide on autoscaling the compute resources of a Weaviate cluster.