pg_hardstorage uses a content-addressed, chain-free backup format for PostgreSQL instead of the traditional chained-incremental model used by pgBackRest and Barman. Each backup manifest references chunks by SHA-256 hash rather than pointing to a parent backup, meaning deleting any single backup never invalidates others. This mirrors the approach taken by restic, kopia, and borgbackup. The trade-off is slightly more I/O on the source side (a full data directory read per backup) until a snapshot-aware fast path ships, but in exchange you get deduplication, a simple JSON manifest inspectable with jq, and a deletion model with no hidden ordering constraints. Real-world numbers on a 487 GiB fleet show an 87.5% dedup ratio, competitive with chained incrementals.

3m read timeFrom cybertec-postgresql.com
Post cover image
Table of contents
The chain footgunContent addressing kills the chain"But what about storage cost?"What we measuredSo what?
94.7K Impressions