PV, PVC & StorageClass

A piece of storage provisioned by an admin or dynamically. Represents the actual storage — an NFS share, an EBS volume, a local disk, a Ceph RBD image. It is a cluster-level resource (not namespaced).

A request for storage by a user or pod. "I need 10Gi of RWO storage." PVCs are namespaced — they live alongside the pods that use them.

Note: Access modes are capabilities, not enforcement. RWO means the volume can be mounted read-write on one node — but if two pods land on the same node, both can mount it. RWOP closes that gap by restricting to a single pod.

• Created when the pod starts, deleted when the pod is removed
• All containers in a pod share it — useful for sidecar communication
• medium: Memory backed by tmpfs (RAM) — fast but uses the pod's memory limit
• Use cases: scratch space, cache, sharing files between init container and app container

volumes:
  - name: scratch
    emptyDir: {}              # disk-backed
  - name: cache
    emptyDir:
      medium: Memory         # tmpfs (RAM-backed)
      sizeLimit: 256Mi

• Mounts a file or directory from the host node into the pod
• Types: Directory, File, Socket, CharDevice, BlockDevice
• Data is node-specific — no portability, no replication
• If pod moves to another node, the data is gone

volumes:
  - name: host-logs
    hostPath:
      path: /var/log
      type: Directory

• Mount ConfigMap or Secret data as files inside the container
• Updates propagate eventually (kubelet sync period, ~1 minute by default)
• subPath: mount a single key as a file at a specific path (but disables auto-update)

volumes:
  - name: config
    configMap:
      name: app-config
  - name: certs
    secret:
      secretName: tls-certs

# Using subPath to mount a single key as a file
volumeMounts:
  - name: config
    mountPath: /etc/app/config.yaml
    subPath: config.yaml          # mount only this key

Note: When using subPath, config/secret updates will not be reflected in the mounted file. The pod must be restarted to pick up changes.

• Classic shared storage — the original "distributed filesystem"
• Supports RWX — multiple pods on multiple nodes can read-write simultaneously
• Good for legacy workloads, shared file access patterns
• Performance can be a bottleneck for high-IOPS workloads

volumes:
  - name: shared-data
    nfs:
      server: nfs.example.com
      path: /exports/data

• Uses a real local disk or partition — not just a directory like hostPath
• Requires node affinity — pod must schedule on the node with the disk
• K8s handles the scheduling constraint automatically
• Good for distributed databases that manage their own replication (Cassandra, CockroachDB, Elasticsearch)

apiVersion: v1
kind: PersistentVolume
metadata:
  name: local-pv
spec:
  capacity:
    storage: 100Gi
  accessModes:
    - ReadWriteOnce
  local:
    path: /mnt/disks/ssd0
  nodeAffinity:                     # required for local volumes
    required:
      nodeSelectorTerms:
        - matchExpressions:
            - key: kubernetes.io/hostname
              operator: In
              values:
                - worker-node-01

Note: A Released PV with Retain policy cannot be automatically rebound. To reuse it, an admin must: (1) delete the PV object (keeping the data), (2) optionally clean up data on the storage backend, (3) create a new PV pointing to the same storage.