Kubernetes Storage

Kubernetes was designed for stateless, disposable pods, and storage is the layer where that assumption breaks down and has to be bolted back on carefully. A pod can be rescheduled to any node at any time; the volume it depends on has to either follow it there or already be reachable from there — get this wrong and a routine reschedule becomes a data-loss incident.

Treat every PVC as a promise about what happens when the pod above it dies, not just where the bytes live today. The reclaim policy and access mode you pick now decide what happens the day something goes wrong, not the day it's created.

1. Understand the claim as a request, not a guarantee

A PVC asks for a certain size, access mode, and StorageClass; the PV that satisfies it is either pre-provisioned or, more commonly now, dynamically created by the StorageClass's provisioner. A Pending PVC almost always means no StorageClass/provisioner can satisfy what was asked for — capacity, access mode, or a zone constraint that doesn't match available storage.