The developer introduced tembo, an open-source HA agent for PostgreSQL written in Go. The tool integrates Raft consensus and bbolt local storage, eliminating the need for external Distributed Configuration Stores (DCS) like etcd or Consul. It operates as a single binary per node, managing both the PostgreSQL instance and cluster state.
The architecture relies on a Raft-based consensus mechanism where each node participates in elections and state replication. The agent performs health checks every second, with failover triggered after a 15-second primary node timeout. If a primary loses its quorum, it automatically stops the PostgreSQL process to prevent split-brain scenarios.
Failover selection prioritizes standby nodes based on WAL lag, with a default threshold of 1 MB. The system supports synchronous replication modes, though zero data loss depends on specific configuration settings like synchronous_commit. In the event of a node failure, the system uses pg_rewind to reintegrate the former primary as a standby.
Planned switchover operations are managed via the tembo CLI, ensuring zero data loss by waiting for the target node to reach the primary's shutdown LSN. The tool is designed to work consistently across bare-metal servers, virtual machines, and Kubernetes environments, avoiding the complexity of external dependencies.
