Postgres High Availability

Replication / failover orchestrators

• ★ Patroni (Zalando) — distributed-coordination-backed orchestrator (etcd / Consul / ZooKeeper / k8s); auto-failover; the default for self-host Postgres HA.
• ★ CloudNativePG — k8s operator (CRD-driven); the default for Postgres on k8s in 2026.
• Stolon (Sorint.lab) — older Patroni alternative.
• repmgr (2ndQuadrant / EDB) — repmgr / repmgrd; older but still used.
• pg_auto_failover (Citus / Microsoft) — simpler 3-node setup; one monitor.
• pgpool-II — connection pool + load balance + failover; older.

Connection poolers

• ★ PgBouncer — universal; tiny; transaction- or session-pooling. Default in front of every Postgres.
• ★ Pgpool-II — adds load balancing across replicas; heavier.
• Supavisor (Supabase) — modern Elixir-based pooler; multi-tenant.
• Odyssey (Yandex) — multi-threaded PgBouncer alternative.

Logical replication / CDC

• Postgres native logical replication — built-in; per-table; for upgrades / migrations / data fan-out.
• pglogical — older external replication.
• Debezium — CDC into Kafka / NATS / Pulsar.
• Sequin / Materialize / Striim — managed CDC platforms.

Backups

• ★ WAL-G — continuous WAL archiving + base backups; PITR. Default for self-host.
• pgBackRest — full-featured; great for big DBs; more setup.
• Barman — older WAL archiver.
• pg_dump / pg_dumpall — logical backups; complement, not replace, WAL.
• See Backup.

Hosted / managed Postgres

See Web Dev Databases — Neon (branching), Supabase, Crunchy Bridge, AWS RDS / Aurora, GCP Cloud SQL, Azure DB.

Performance / tuning

• pgtune — generate postgresql.conf for your hardware.
• PgHero — easy performance dashboard.
• pg_stat_statements — built-in; what's slow.
• auto_explain — log slow query plans.
• pgBadger — log analyzer.
• @embarrassingly/cloud-pg-tune style tools — many.

Extensions worth knowing

• pgvector — embeddings + vector search.
• pgvectorscale — for huge vector workloads.
• TimescaleDB — time-series.
• Citus — sharding.
• PostGIS — spatial.
• pg_partman — auto-partitioning.
• pgaudit — audit logging.
• pg_cron — schedule jobs in-DB.
• pgmq — message queue extension.

HA topologies

Classic 3-node Patroni:

• 3 Postgres nodes (1 primary, 2 replicas).
• etcd / Consul cluster (3 or 5 nodes) for leader election.
• HAProxy or similar pointing at the current primary based on Patroni's API.
• PgBouncer in front for connection pooling.

CloudNativePG on k8s:

• 3-replica Cluster CRD.
• Operator handles failover, backups (to S3), upgrades.
• service-rw and service-ro services route to primary / replicas.

Cheap-and-cheerful:

• Single primary + a logical replica for read-only reporting.
• WAL-G to R2 for PITR.
• Restore from WAL-G if primary dies. RTO ~minutes.

Patterns to adopt

• ★ Always run PgBouncer. Connection storms kill Postgres before anything else.
• Use logical replication for major-version upgrades. Replicate to a new-version replica, then switch.
• WAL archive to object storage from day 1.
• Test restore drills. A WAL archive you've never restored is a hope.
• Pin extensions in IaC; don't CREATE EXTENSION ad-hoc in prod.
• Monitor replication lag — alert if > a few seconds.

Pick this if…

• Default self-host HA: Patroni + etcd + HAProxy + PgBouncer + WAL-G.
• Default k8s: CloudNativePG.
• Just want managed Postgres: Neon or Crunchy Bridge.
• Backup + PITR: WAL-G or pgBackRest.
• Connection pool: PgBouncer (always).