Storage and data

Containers are disposable. Business data is not.

Persistent storage

PersistentVolumeClaims request storage through a StorageClass. GKE commonly uses Persistent Disk or Hyperdisk CSI drivers. Zonal volumes live in one zone; regional persistent disks synchronously replicate across zones where supported. Filestore provides managed NFS. Cloud Storage FUSE exposes object data through a file interface for suitable workloads, but object semantics are not a normal POSIX disk.

Database decision

Prefer Cloud SQL, AlloyDB, Spanner, Bigtable, Memorystore, Firestore, or another managed data service when it meets requirements. Running a database in GKE is reasonable only when Kubernetes portability or special software needs justify operating replication, patching, backup, failover, and capacity.

Connection design

Use private connectivity where appropriate. Prefer Workload Identity over key files. Use connection pooling and bounded concurrency. A thousand new Pods can become a database outage if every Pod opens many connections.

Backup

Backup for GKE captures selected Kubernetes resources and persistent volume data. It does not capture cluster configuration, node pools, or container images. Preserve infrastructure-as-code and images separately. Restore into a prepared target cluster and test the full application.

Official source: Backup for GKE.