// the find
EnterpriseDB/repmgr
A lightweight replication manager for PostgreSQL (Postgres)
repmgr is a PostgreSQL replication and failover manager that wraps Postgres's built-in streaming replication with higher-level tooling: registering nodes, cloning standbys, executing switchovers, and running an automatic failover daemon (repmgrd). It's for teams running self-managed Postgres clusters who want failover automation without going all-in on Patroni or an orchestrated container setup.
The CLI is genuinely well-structured — separate subcommands per concern (standby clone/register/follow/promote, cluster show/matrix) make cluster operations scriptable and auditable. The witness node concept is a real solution to split-brain: a lightweight non-data node that participates in quorum without holding replica state. Backed by EnterpriseDB with commercial support available, so it's not abandonware. The SQL migration files from 4.0 through 5.5 are present and numbered, which means upgrades have a tracked path rather than 'just reinstall'.
Automatic failover requires repmgrd running as a daemon on every node, and fencing is not built in — you have to configure external fencing scripts yourself or accept the split-brain risk; the docs acknowledge this with a separate markdown file but don't solve it for you. No consensus store (unlike Patroni's etcd/Consul/ZooKeeper backend), so cluster state lives in a Postgres metadata schema on the same cluster you're trying to protect — during a hard failure, that metadata may be unreachable at exactly the wrong moment. Last push was March 2026 and the repo is at 5.5 with no activity since; it supports Postgres 17 but there's no indication 18 support is in progress. Testing infrastructure is thin: one regression test file, no visible CI matrix in the directory tree.