The best practice for ensuring application high availability (HA) is to run multiple replicas (pods) of the application and distribute them across different worker nodes. This strategy mitigates the risk of a single point of failure. If one worker node becomes unavailable due to hardware failure or maintenance, the application instances running on other healthy worker nodes can continue to serve traffic, ensuring service continuity. This is typically managed by a Kubernetes Deployment or a Red Hat OpenShift DeploymentConfig, which works with the scheduler to enforce this distribution.

B. Use multiple pods across different master nodes.

Application workloads should not be scheduled on master (control plane) nodes. These nodes are reserved for running essential cluster management services, and adding application pods can compromise cluster stability and security.

C. Use multiple small pods on a master node.

This is incorrect for two reasons: it violates the principle of not running applications on master nodes, and it creates a single point of failure by placing all pods on one node.

D. Use multiple pods across both worker and master nodes.

While this includes the correct practice of using worker nodes, it also includes the incorrect and risky practice of using master nodes for application workloads, making it a flawed HA strategy.

1. IBM Cloud Documentation - High availability for your cluster: "To protect your app from a zone failure

you must have multiple worker nodes in different zones in your cluster and deploy multiple replicas of your app... When you have multiple app pods

they are spread across worker nodes in different zones." This document explicitly states that spreading app pods (replicas) across different worker nodes is the key to high availability.

Source: IBM Cloud Docs

"High availability for your cluster"

Section: "Single zone clusters".

2. Kubernetes Documentation - Concepts > Architecture > Nodes: "The worker nodes in a cluster are the machines (VMs

physical servers

etc) that run your containerized applications... The control plane manages the worker nodes and the Pods in the cluster." This official documentation establishes the fundamental architectural separation where applications run on worker nodes

not control plane (master) nodes.

Source: Kubernetes Documentation

"Components of Kubernetes"

Section: "Worker Nodes".

3. Red Hat OpenShift Container Platform Documentation - About high availability: "For application high availability

you should have multiple replicas of a pod running. If you have multiple replicas of a pod running on different nodes and a node fails

the pods on the failed node are recreated on other nodes." This directly supports the principle of using multiple pods on different nodes for application HA.

Source: Red Hat OpenShift Container Platform 4.14 Documentation

"About high availability"

Section: "Application high availability".