PostgreSQL

Managed instance of the PostgreSQL database server.

Components

  • PostgreSQL server (versions 10, 11, 12, 13, 14)

Configuration

Managed PostgreSQL instances already have a production-grade configuration with reasonable sized memory parameters, for example, shared_buffers and work_mem.

Warning

Putting custom configuration in /etc/local/postgresql/VERSION/*.conf doesn’t work properly starting with NixOS 20.09 and should not be used anymore. Some options from there will be ignored silently if they are already defined by our platform code. Use NixOS-based custom config as described below instead.

You can override platform and PostgreSQL defaults by using the services.postgresql.settings option in a custom NixOS module. Place it in /etc/local/nixos/postgresql.nix, for example:

{ config, pkgs, lib, ... }:
{
  services.postgresql.settings = {
      log_connections = true;
      huge_pages = "try";
      max_connections = lib.mkForce 1000;
  };
}

To override platform defaults, use lib.mkForce before the wanted value to give it the highest priority.

String values will automatically be enclosed in single quotes. Single quotes will be escaped with two single quotes. Booleans in Nix (true/false) are converted to on/off in the PostgreSQL config.

Run sudo fc-manage -b to activate the changes (restarts PostgreSQL!).

See Custom NixOS-native configuration for general information about writing NixOS modules.

Interaction

Service users can use sudo -u postgres -i to access the PostgreSQL superuser account to perform administrative commands like createdb and createuser.

Service users may invoke sudo fc-manage --build to apply configuration changes and restart the PostgreSQL server (if necessary).

Monitoring

We use the following Sensu checks to monitor the availability of PostgreSQL:

  • postgresql-alive: uses the UNIX socket in /run/postgresql to connect and logs in as fcio_monitoring database user.

  • postgresql-listen-*: connects to all configured listen addresses and checks if a TCP connection can be established.

Telegraf is used to export metrics for PostgreSQL. Telegraf connects as fcio_monitoring database user. Metrics can be viewed when a Statshost is present, using the FCIO/PostgreSQL dashboard.

Platform-created Databases

We create the fcio_monitoring database for monitoring purposes and root for the root user. In a fresh installation, the following databases are present: fcio_monitoring, postgres, root, template0, template1. Older installations still have a nagios database for monitoring which is not used anymore.

Major Version Upgrades

Upgrading to a new major version, for example from 13.x to 14.x, requires a migration of the old database cluster living in /srv/postgresql/13 to the new data directory at /srv/postgresql/14. A common way to do this is to use pg_upgrade bundled with PostgreSQL. This works on our platform but doing it properly is not trivial.

To make it easy and to prevent common errors, we provide a fc-postgresql command which prepares and runs upgrade migrations. It can also show the current state of data directories for the available major versions.

Note

fc-postgresql has to be run as postgres user. Prefix the commands with sudo -u postgres or use sudo -iu postgres to change to the postgres user. This is allowed for service and sudo-srv users.

Warning

When the PostGIS extension is used, upgrading from a version before 12 to 12 or higher isn’t possible using the fc-postgresql command. You have to create a full dump of the old cluster using pg_dumpall, switch to the new role, rebuild the system and use pg_restore.

To show which data directories exists, their migration status and which service version is running, use sudo -u postgres fc-postgresql list-versions. Add --help to see details about the meaning of the columns.

Note

Please look at the output of sudo -u postgres fc-postgresql list-versions before performing an upgrade and make sure that your assumptions about the current state (which version is active, which data dirs are there, …) are correct.

The upgrade commands need to know which databases are expected to be present in the cluster. Default databases created by PostgreSQL or our platform code are always accepted and don’t have to be specified.

If you have two databases, mydb and otherdb, for example, specify both on the command line.

To prepare an upgrade, when you use the postgresql13 role at the moment, and you want to change to postgresql14, run:

sudo -u postgres fc-postgresql upgrade --new-version 14 --expected mydb --expected otherdb

Note that this is done while the old role is still active. It’s safe to run the command while PostgreSQL is running as it does not have an impact on the current cluster and downtime is not required.

The command should automatically find the old data directory for 13, set up the new data directory and succeed if no problems with the old cluster were found. Problems may occur if the old cluster has been created with non-standard settings which are not compatible with the new cluster, the old directory has an invalid structure or multiple old data directories which need migration are found.

Warning

Depending on the machine and the amount of data, the next step may take some time. PostgreSQL will not be available during the upgrade!

To actually run the upgrade, use:

sudo -u postgres fc-postgresql upgrade --new-version 14 --expected mydb --expected otherdb --upgrade-now

This will stop the postgresql service, prevent it from starting during the upgrade, migrate data and mark the old data directory as migrated. This data directory cannot be used by the postgresql service anymore after this point.

Run sudo -u postgres fc-postgresql list-versions to see how the status of the old and new data dir has changed.

After the migration, postgresql is still stopped, and you have to change your configuration to the new major version, for example by disabling the postgresql13 role and activating the postgresql14 role, in one step.

If you really need to go back to the old version, delete the new data directory as postgres user, remove the fcio_migrated_to* files in the old data directory and switch back to the old postgresql role.

Miscellaneous

  • Our PostgreSQL installations have the autoexplain feature enabled by default.