Quick tactic: how to stop total node meltdown

Where this fits in K8s strategy

Stop replica nodes from overload caused by loads shacking up in one place

Why it’s important

Cuts risk of a domino effect of all node replicas crashing.

Problem

Kubernetes can assign pods to the same node by default. The risk comes up when you have a few too many resource-hungry pods.

That can overload the node. The whole cluster is at risk if the pattern continues.

The issue can occur even if you have a 3-replica model. That’s when you have 3 identical nodes for redundancy.

Solution

Apply inter-pod anti-affinity rules that makes sure certain (hungry) pods are not put together in the same node.

Read this story first

Then read the official documentation to implement solution

Leave a Comment