[ SRE_ARENA ]
K8S WORKLOADS & SCHEDULING

Workloads & Scheduling: Pods are Cattle

DOWNLOADING CONCEPT: 1. THE GOLDEN RULE: NEVER RUN NAKED PODS...
  • ↳A 'Naked Pod' is a Pod you create directly using kubectl run.
  • ↳If the physical worker node it's running on catches fire and dies, the Pod dies with it and is never resurrected.
  • ↳To make your applications highly available and self-healing, you must wrap them in a Workload Controller (like a Deployment, StatefulSet, or DaemonSet).
  • ↳These controllers run in the Control Plane and constantly monitor your Pods.
  • ↳If a node dies, the controller instantly notices the missing Pods and reschedules them onto healthy nodes.
▲ LIVE INCIDENTYou scaled a deployment from 3 to 10 replicas, but the 7 new pods are stuck in Pending.
sre@arena:~/k8s-workloads
sre@arena:~$ kubectl get pods | grep Pending