This Is Not Managed Postgres

Before you use Fly Postgres, here are some things worth understanding about it:

Fly Postgres is a regular app you deploy on Fly.io, with an automated creation process and some platform integration to simplify management. It relies on building blocks available to all Fly apps, like flyctl, volumes, private networking, health checks, logs, metrics, and more. The source code is available on GitHub to view and fork.

This is not a managed database. If Postgres crashes because it ran out of memory or disk space, you’ll need to do a little work to get it back.

Here’s what Fly.io manages

  • Basic provisioning & upgrade tools - fly pg provides a set of commands and building blocks that automates the provisioning of databases, connecting them to your applications, and accessing remote Postgres shells.
  • Daily volume snapshots - Fly.io takes daily snapshots of Postgres volumes and saves them for 5 days.
  • Global networking & server infrastructure - Fly.io provides the Firecracker VMs and interconnecting WireGuard network to run your Postgres database cluster.
  • Prometheus metrics - Fly.io collects and exposes various metrics via Prometheus, but you have to set up a tool, like Grafana, for aggregation and alerting.
  • Open Source templates - Templates and documentation are available at https://github.com/fly-apps/postgres-flex in the event that you need to work the Fly Postgres app to fine-tune it for your specific application.

Here’s what you manage

Deploying a Fly Postgres database means you may need to manage the following:

  • Provisioning the right cluster config for you - Consider what your availability needs are. If you run only a single instance, then anything that happens to that disk, that server or the network in that region can cause your database to be unavailable.
  • Scaling storage and memory resources - If your database runs out of disk space or memory, you’ll have to scale it up, or down if you don’t need the resources.
  • Upgrading Postgres versions & security patches - Fly.io provides tools like fly image update to upgrade your database instances to new minor versions of Postgres, but you’ll have to run the upgrades yourself. Same for security patches: you’ll have to apply those to running Postgres clusters.
  • Developing a database backup & restoration plan - Fly.io’s self-managed Postgres ships with a basic daily volume snapshot tool that keeps snapshots around for 5 days. It does not manage off-site backups, etc.
  • Monitoring & alerts - Fly.io collects and exposes relevant Prometheus metrics, but you’ll have to configure your own monitoring and alerts to keep tabs on the performance and resource utilization of your database instances.
  • Recovering from outages - If the volume in your database fills up, a replica fails, etc. you’ll have to do a little bit of work to bring your database back to health.
  • Global replication - You can add read-only replicas outside the primary region to speed up read-heavy globally distributed apps, by scaling your Fly Postgres app. It’s up to you to tweak your application to get writes to the leader instance, but the Fly-Replay header simplifies that.
  • Configuration Tuning - You may need to tune your Postgres configuration to match your application’s needs. There are a lot of knobs to turn, but fly pg config only supports a few of them out of the box. For more details, see Postgres Configuration Tuning.
  • Advanced customization - TimescaleDB is included in the default image and can be enabled with these instructions. If your application demands additional Postgres extensions or something else in the VM, you can fork and maintain your own branch of Fly’s open source Postgres Flex app.

Fully managed Postgres

Supabase Postgres, a fully managed Postgres database on Fly.io, is now in beta. Learn more about Supabase Postgres on Fly.io.

Recommended External Providers

If you want a fully managed database solution for your Fly Apps, there are many great options, including:

Other Places

You can connect your Fly Apps to the usual suspects, too:

Report an issue or edit this page on GitHub