Backup Model

Backups are stored in restic repositories with content-defined chunking (CDC) for block-level deduplication. All backups go through restic backup --stdin with a virtual filename in the snapshot.

Snapshot Tags

Every snapshot is tagged with:

Tag	Values	Description
engine	cnpg, galera	Database engine
cluster	cluster name	Database cluster CR name
namespace	namespace	Kubernetes namespace
type	backup, diverged	Snapshot type (see below)
job	20060102T150405Z	Groups diverged snapshots from the same repair (diverged only)

Snapshot Types

Type	When	Virtual Path	Description
backup	Normal backup or pre-repair escrow	<ns>/<cluster>/pgdumpall.sql	Standard database dump. Escrow backups before repair are also type=backup and follow normal retention.
diverged	Split-brain detected during repair	<ns>/<cluster>/<ordinal>-pgdumpall.sql	Per-instance capture of each diverged replica. Shared job tag groups them. Forensic record for admin review.

Engine-specific filenames: CNPG uses pgdumpall.sql, Galera uses mysqldump.sql.

Snapshot Timestamps

All snapshots use --time <job-start> so the restic timestamp reflects when the operation was initiated, not when the dump completed. This ensures all snapshots from the same job (escrow + diverged captures) share a consistent timestamp.

Retention / Prune

hasteward prune backups -e cnpg -c my-postgres -n my-ns --backups-path /backups

Flag	Default	Description
--keep-last	7	Keep the last N snapshots (or jobs for diverged)
--keep-daily	30	Keep N daily snapshots (or jobs for diverged)
--keep-weekly	12	Keep N weekly snapshots (or jobs for diverged)
--keep-monthly	24	Keep N monthly snapshots (or jobs for diverged)
-t / --type	backup	Snapshot type to prune: backup, diverged, or all

Group-aware prune for diverged snapshots: Retention policies apply to job groups, not individual snapshots. A repair job that captured 3 diverged instances counts as 1 unit for --keep-last. Snapshots sharing the same job tag are kept or removed together.