backup-strategy

Classify data by criticality, then assign RTO (how fast you must recover) and RPO (acceptable data loss) per system. Lower targets cost more: a 1-hour RTO with 5-minute RPO needs hot standby and real-time replication, while 24-hour targets work with daily snapshots.

The 3-2-1 rule requires three copies of data, on two different media types, with one copy stored offsite. This protects against single points of failure, ransomware, and regional disasters by combining local disk, cloud storage like S3, and an offsite datacenter.

Use pg_dump with the custom format, compress the output, and upload to S3 with the AWS CLI. The provided script also verifies archive integrity with gunzip, deletes local backups older than 7 days, and prunes S3 objects older than 30 days.

Incremental backups capture changes since the last backup of any type, minimizing storage but requiring the full backup plus all incrementals to restore. Differential backups capture changes since the last full backup, using more space but needing only two files to restore.

Use S3 Object Lock in governance mode to make backups immutable for a retention period, preventing deletion even by privileged accounts. Additionally, maintain air-gapped offline backups on disconnected external drives that ransomware cannot reach.

Untested backups frequently fail due to corruption, missing dependencies, or incomplete archives. The Skill includes an automated restore test that restores to a temporary database and verifies table counts, with quarterly restore drills recommended.