High Availability

By default, the SKE Operator runs one pod of Kratix and one pod of the SKE Platform Manager. When the node running a pod is drained or fails, nothing serves the platform until a new pod starts. Running more than one replica, spread across nodes, keeps SKE available while this happens.

What extra replicas give you

Kratix and the SKE Platform Manager use Kubernetes leader election:

• Only one pod is active at a time.
• If the leader pod stops, another pod takes over after about 15 to 20 seconds. Reconciliation pauses during that time.
• Every pod serves webhooks, including the pods that are not the leader.

Extra replicas do not make SKE faster

Only the leader reconciles, so adding replicas does not make Kratix or the SKE Platform Manager reconcile faster.

Set the number of replicas

Set replicas on the Kratix resource. deploymentConfig configures Kratix and platformManagerDeploymentConfig configures the SKE Platform Manager:

apiVersion: platform.syntasso.io/v1alpha1
kind: Kratix
metadata:
  name: kratix
spec:
  version: vX.Y.Z
  deploymentConfig:
    replicas: 2
  platformManagerDeploymentConfig:
    replicas: 2

If you install with Helm, set the same fields under skeDeployment, see Replicas and pod spreading.

• When replicas is not set, one pod runs.
• replicas must be at least 1. The Kratix resource is rejected otherwise.
• The Kratix resource is ready only when every Kratix pod is available.
• The SKE Operator owns these Deployments. If you change their replica count with kubectl scale or a HorizontalPodAutoscaler, the operator sets it back on its next reconcile. Set replicas on the Kratix resource instead.

Spread the pods across nodes

Use topologySpreadConstraints to spread pods across different nodes. For example:

apiVersion: platform.syntasso.io/v1alpha1
kind: Kratix
metadata:
  name: kratix
spec:
  version: vX.Y.Z
  deploymentConfig:
    replicas: 2
    topologySpreadConstraints:
      - maxSkew: 1
        topologyKey: kubernetes.io/hostname
        whenUnsatisfiable: DoNotSchedule
        nodeTaintsPolicy: Honor
        matchLabelKeys:
          - pod-template-hash
        labelSelector:
          matchLabels:
            app.kubernetes.io/component: kratix-manager

See Pod topology spread constraints for documentation of every available field.

Keep pods running during node maintenance

The SKE Operator does not create a PodDisruptionBudget. You can create one yourself. The example below keeps at least one Kratix pod running:

apiVersion: policy/v1
kind: PodDisruptionBudget
metadata:
  name: kratix-manager
  namespace: kratix-platform-system
spec:
  minAvailable: 1
  selector:
    matchLabels:
      app.kubernetes.io/component: kratix-manager
      # use app.kubernetes.io/component: ske-platform for the SKE Platform Manager

A PodDisruptionBudget can block a node drain

If minAvailable is equal to replicas, no pod can be evicted and kubectl drain never finishes. Keep minAvailable lower than replicas, and update the budget when you change replicas.