A MySQL backup orchestrator with point-in-time recovery

The brief

Multiple databases, multiple environments, mixed Docker + bare-metal — and the backup story was a tangle of cron jobs and assumptions. We needed point-in-time recovery, tested restores, and an honest answer to “what’s our RPO and RTO” for every database we run.

What we built

• Full + incremental snapshots built on Percona XtraBackup with binlog application for point-in-time recovery to any second within the retention window.
• Docker + host-based MySQL support with a single config schema — the same orchestrator runs against containers and hardware.
• Automated retention — daily, weekly, monthly policies, enforced by the orchestrator, never by humans editing crontabs.
• zstd compression with verified integrity hashes; bandwidth costs dropped by ~60% vs. gzip without measurably impacting backup duration.
• Restore verification — every backup is restored into a sandbox automatically and validated against checksum + sample queries. Untested backups don’t count.
• Operator playbook for the top-five recovery scenarios, kept inside the same repo as the tooling.

Outcome

Documented RPO of 5 minutes and RTO of 30 minutes for every database in the system. First time the team had a backup playbook they actually trusted enough to run from cold.