WarehousePG Disaster Recovery

WarehousePG Disaster Recovery (whpg-dr) is a command-line tool for protecting WarehousePG (WHPG) clusters against data loss and extended outages. It takes physical backups of your cluster's coordinator and segment data directories, ships write-ahead log (WAL) files to a secondary storage repository, and lets you restore the cluster to any restore point on the same infrastructure or different infrastructure.

Key capabilities

• Three DR modes:
  • Cold DR for on-demand restore at failover time
  • Warm DR for a continuously current standby with low RTO
  • Point-in-time recovery for restoring to a specific past state
• Back up the entire cluster in parallel to a POSIX-compliant filesystem such as NFS, or S3-compatible storage
• Restore to any point in time, whether to recover from data loss or inspect historical cluster state
• Advance an already-restored cluster to a later restore point without a full re-restore, enabling warm DR with low RTO
• Promote a restored cluster to a writable primary in a single command

How WarehousePG Disaster Recovery works

WarehousePG Disaster Recovery wraps Barman as its backup and restore engine, giving you a purpose-built CLI for WarehousePG while Barman handles the low-level mechanics. Each segment has its own Barman process, so backups and restores run in parallel across all segments automatically.

The storage repository, an NFS share or an S3-compatible object store, connects the primary and DR sites. whpg-dr requires no persistent link between the two clusters other than shared access to the repository. See Understanding disaster recovery for a deeper explanation of how these pieces fit together.

DR modes

WarehousePG Disaster Recovery supports three operational modes:

• Cold DR: The DR cluster doesn't need to exist until a failure occurs. Backups and WAL sit in storage, and you restore on demand. Recovery Time Objective (RTO) is higher than warm DR because provisioning and a full restore are required at failover time, but infrastructure cost is lower.
• PITR (point-in-time recovery): Restores the cluster to any restore point in storage, regardless of whether a DR cluster currently exists. Use it to recover from data corruption or to inspect the database as it was at a specific past point.
• Warm DR: A pre-provisioned DR cluster keeps pace with the primary by replaying WAL at a regular interval. The cluster remains in recovery mode (not queryable) until you promote it. RTO is low because the data is mostly current.