So, I learned the painful truth that “Kubernetes makes everything stateless” is only half true — until you lose your cluster state. Backing up etcd is absolutely essential, and not just once; automate it and store it off-cluster. For persistent volumes, use CSI snapshots or Velero with a reliable object storage target — don’t rely on manual PVC exports, they’ll bite you later. I also found that backing up manifests from GitOps repos isn’t enough if your secrets aren’t versioned properly. Encrypt them, store the keys safely, and test your restore in a clean namespace or even a new cluster. Backups that haven’t been restored are just pretty files sitting in S3.
