Benchmarking How High Availability Affects Managed Postgres Speed
PostgresBench benchmarks the real throughput and latency cost of high availability across managed Postgres providers.
PostgresBench, the open-source and reproducible benchmark for managed Postgres services, now extends its methodology to high-availability (HA) configurations. The test compares ClickHouse Managed Postgres, AWS RDS, Crunchy Bridge, AWS Aurora, and Neon under matched compute, targeting a common HA bar: recovery within 2 minutes and zero data loss. Rather than ranking overall HA feature sets, the goal is to isolate the performance cost of reaching comparable availability and durability.
The results highlight how architecture shapes that cost. Shared-nothing services (ClickHouse Managed Postgres, RDS, Crunchy Bridge) rely on native PostgreSQL streaming replication and physical standbys, while shared-storage forks (Aurora, Neon) achieve durability by replicating WAL writes across a distributed storage layer instead of maintaining traditional standbys. With synchronous quorum replication delivering zero RPO, ClickHouse Managed Postgres sustains roughly 4x higher TPS, 3x lower p50 latency, and 7x lower p99 latency than RDS Multi-AZ configurations offering equivalent guarantees.
For engineers, the takeaway is that HA is never free, but its price varies dramatically by vendor and design. Understanding the real throughput and tail-latency trade-offs of each HA mode is essential when sizing infrastructure and setting SLAs for mission-critical workloads.