Postgres Backup and Recovery

4 posts · Jan 2024 – Jul 2026

What this trail covers

PlanetScale offers robust database backup and recovery solutions, leveraging Vitess for sharding to enable parallelized backups and restores. This significantly reduces backup times for large databases and speeds up recovery processes. The system handles the complexities of backup scheduling, execution, and storage, providing users with a reliable snapshot of their data. Backups are also crucial for efficient replica creation and data recovery scenarios. This post adds considerations for building and operating backups with sharding, detailing the parallelization benefits and trade-offs in catch-up replication.

All PlanetScale trails

Jul 2026Massively parallel Postgres backupsThis post details the engineering approach to performing massively parallel backups for sharded Postgres databases (Neki). It describes the backup lifecycle, including spinning up dedicated backup nodes per shard, reusing previous backups from object storage, and replaying WAL from both object storage and the primary to minimize production impact. It also covers the initial backup process using `pg_basebackup` and explains how these parallel backup techniques are crucial for efficient database resizing and node replacement operations, highlighting the performance benefits of parallelism for large datasets.
Jul 2026Postgres backups under the hoodThis post details the mechanics of Postgres backups, contrasting logical backups (`pg_dump`) with file system backups and continuous archiving. It explains the limitations of `pg_dump`, particularly its resource consumption on large databases and its inability to perform point-in-time recovery. A significant portion is dedicated to explaining transaction wraparound and how long-running transactions, like those initiated by `pg_dump`, can inadvertently trigger read-only mode. The post then elaborates on continuous archiving, explaining the role of the Write-Ahead Log (WAL) in ensuring data durability and enabling point-in-time recovery by replaying WAL segments alongside a file system backup.
Jul 2024Faster backups with shardingThis post details how PlanetScale leverages Vitess sharding to dramatically speed up database backups. It explains the internal backup process, including restoring previous backups, catching up changes, and storing new full backups. The core contribution is demonstrating how parallelizing these steps across shards, rather than performing them serially on a single instance, reduces backup times from hours/days to minutes/hours for large databases. It also discusses the trade-off of replicating catch-up changes from the primary versus a replica.
Jan 2024Considerations for building a database disaster recovery planThis post introduces key considerations for building a database disaster recovery plan. It differentiates between High Availability (HA) and Disaster Recovery (DR), defines Recovery Point Objectives (RPOs) and Recovery Time Objectives (RTOs) as critical metrics, and explains the stateful nature of databases versus stateless code. It details MySQL replication (asynchronous vs. semi-synchronous) and backup strategies (logical vs. physical, full vs. incremental), highlighting PlanetScale's practice of using read-only replicas for backups and validating them by restoring to a backup-dedicated replica. The post also provides actionable advice on defining RPOs/RTOs with business leadership, using cross-region replication for faster recovery, prioritizing restoration efforts, quantifying downtime in terms of lost revenue, and automating the recovery process.
Jan 2024PlanetScale vs Amazon Aurora replicationThis post details PlanetScale's approach to automatically validating backups by restoring the most recent backup to a dedicated node, applying replication changes, and then creating a new backup. This process ensures that backups are healthy and restorable, a capability enabled by their traditional MySQL replication setup.
Advertisement