Got deep into Kubernetes backup and recovery this year — Velero for backup/restore orchestration, MinIO as the object storage backend, restore automation triggered through CI.
What surprised me most: "did the restore succeed" is rarely a yes/no question. Partial success is a real, common state, and treating it as a binary breaks the moment you actually need the tooling during an incident.