Scale and run highly available#

This guide shows you how to run more than one replica of the Plone backend and frontend, protect them with a PodDisruptionBudget, and add a HorizontalPodAutoscaler.

Prerequisites#

A working Plone deployment created with cdk8s-plone.
A PostgreSQL backend reachable by every backend replica.

The backend uses RelStorage on PostgreSQL, so several backend replicas share one database and scale horizontally. The Volto frontend is stateless and scales horizontally as well.

Set the replica count#

Set replicas on backend and frontend independently.

import { Plone, PloneVariant } from '@bluedynamics/cdk8s-plone';

new Plone(chart, 'plone', {
  variant: PloneVariant.VOLTO,
  backend: {
    image: 'plone/plone-backend:6.1.3',
    replicas: 3,
  },
  frontend: {
    image: 'plone/plone-frontend:16.0.0',
    replicas: 2,
  },
});

Both default to 2 replicas when you omit the option.

Protect availability with a PodDisruptionBudget#

A PodDisruptionBudget keeps a minimum number of pods running during voluntary disruptions such as node drains.

Set minAvailable or maxUnavailable on the component. Each accepts an absolute number or a percentage string such as "50%".

backend: {
  image: 'plone/plone-backend:6.1.3',
  replicas: 5,
  minAvailable: 3,
}

cdk8s-plone creates the PodDisruptionBudget automatically when replicas is 2 or more, or when you set minAvailable or maxUnavailable explicitly. A single replica with no explicit setting gets no PodDisruptionBudget. The generated PodDisruptionBudget always sets unhealthyPodEvictionPolicy: AlwaysAllow, so unhealthy pods never block a node drain.

Add a HorizontalPodAutoscaler#

cdk8s-plone emits plain Deployments and does not configure autoscaling. To scale on load, add your own HorizontalPodAutoscaler that targets the generated backend Deployment.

Find the Deployment name from the synthesized manifests:

cdk8s synth
grep -A2 'kind: Deployment' dist/*.yaml | grep 'name:'

Add the autoscaler as a Kubernetes manifest and apply it alongside your deployment:

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: plone-backend
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: <backend-deployment-name>
  minReplicas: 2
  maxReplicas: 8
  metrics:
    - type: Resource
      resource:
        name: cpu
        target:
          type: Utilization
          averageUtilization: 70

The HorizontalPodAutoscaler needs metrics-server in the cluster. Keep minReplicas at 2 or more so the PodDisruptionBudget stays effective.

Verify#

Confirm the PodDisruptionBudget is present in the synthesized manifests:

cdk8s synth
grep -l 'kind: PodDisruptionBudget' dist/*.yaml

Inspect the running objects on the cluster:

kubectl get deploy -n <namespace>
kubectl get pdb -n <namespace>
kubectl get hpa -n <namespace>