Containers lose their files when they restart. For data that must last, Pods mount a PersistentVolumeClaim: a request for storage of a size and access mode. A PersistentVolume is the actual piece of storage, created by an administrator or by a provisioner, and the cluster binds each claim to one volume that fits its class, size and access modes.
Access modes describe what the storage allows: ReadWriteOnce (one node writes), ReadOnlyMany and ReadWriteMany (many nodes), and ReadWriteOncePod. To bind a claim to a volume made by hand, set storageClassName: ""; otherwise the default class provisions a new disk.
When a claim is deleted, the volume's persistentVolumeReclaimPolicy decides the data's fate: Delete removes the disk, Retain keeps it as Released. Volumes made by a provisioner usually start with Delete, so change the volume to Retain before you delete a claim whose data matters.
Claims and volumes, and keeping a disk before you let its claim go.