Funded upstream maintenance

pg_auto_failover Maintenance and Priority Support

Automated PostgreSQL high availability, without an external consensus cluster to babysit

pg_auto_failover runs a monitor that tracks the health of primary and standby nodes, promotes a standby when the primary fails, and reconfigures the cluster on its own. It is built on PostgreSQL’s own replication, which is what keeps its behaviour something you can reason about and test rather than trust blindly.

1,362GitHub stars v2.2Apr 2025 2vendors ship it as a documented HA option
See Pricing & Tiers View on GitHub

What pg_auto_failover Does During a Failure

Automatic promotion with zero data loss

pg_auto_failover runs a monitor that tracks the health of primary and standby nodes. When the primary fails, the monitor promotes the standby and reconfigures the cluster — no manual intervention, no data loss, no external dependencies.

Why it matters

Manual failover is slow and error-prone under pressure

  • Primary failures happen at the worst possible time
  • Manual promotion steps are risky when you're half-awake at 3am
  • External tools add operational complexity and new failure modes
  • pg_auto_failover uses PostgreSQL's own replication — nothing to learn, nothing to trust blindly

Typical use cases

Where pg_auto_failover fits

  • Production PostgreSQL clusters requiring HA without Patroni complexity
  • Kubernetes deployments needing predictable failover
  • Teams who want HA behaviour they can reason about and test
  • On-premise and cloud environments alike

Key capabilities

Built on PostgreSQL native features

  • Automatic failover with synchronous replication
  • Zero data loss on controlled failover
  • Monitor node coordinates the cluster state machine
  • Works with any number of standby nodes
  • Simple CLI for setup and operations

What’s Being Worked On Right Now

Active maintenance areas

Current work focuses on PostgreSQL version compatibility, Kubernetes operator integration, stability improvements for edge cases in network partition scenarios, and documentation.

The full backlog for this project — last release, what has merged since, and the big features queued behind it — is on the roadmap page, kept current rather than summarised here.

Install pg_auto_failover

Open source and available now

pg_auto_failover is open source and available as a PostgreSQL extension. Packages are available for Debian, Ubuntu, and RPM-based distributions, and a Docker image is provided.

The pg_auto_failover documentation covers the state machine, multi-standby setups, and the operations CLI.

GitHub Documentation

Fund pg_auto_failover Maintenance

Keep your HA setup supported through PostgreSQL upgrades

VMware/Broadcom Tanzu Postgres documents pg_auto_failover as an officially supported HA option, and Microsoft’s Citus Data is where it was born. HA software earns its keep on the day it fails over; funding its maintenance is what keeps that day boring, and what keeps compatibility current as PostgreSQL majors ship.

Work is prioritized through one queue across all four projects, triaged against your tier’s SLA, and shipped in quarterly releases — upstream, open source, for everyone.

See Pricing & Tiers See This Project’s Backlog

Get More Out of the PostgreSQL You Already Run

One learning path, four ways in — from The Art of PostgreSQL

Funded maintenance keeps the tools working. How much you get out of PostgreSQL itself is a different problem — query design, data modelling, and knowing which of its features you are not using. That is what the rest of The Art of PostgreSQL is for.

These aren’t four separate products. They’re one path: start with the book for the foundations, take the course to structure and deepen them, join a masterclass when you want to go further on a specific topic with the author in the room — and license it for the team when it should be everyone’s baseline, not one person’s.

Book

Read it yourself. SQL and data modelling from the ground up — the foundation the rest builds on.

Course

Work through it. Structured modules and hands-on exercises against real datasets, at your own pace.

Masterclass

Ask questions live. Three-hour sessions with the author on one topic, or the full curriculum.

Team

Bring the whole team. Per-seat licensing from five people up, with a dashboard to manage progress.

Copyright 2026 © Dimitri Fontaine. All rights reserved.