PoolPilot
A managed, drop-in PgBouncer fleet that auto-scales across cores using so_reuseport and peering, so connection pooling never caps your Postgres throughput.
Backend engineering teams at SaaS and e-commerce companies running Postgres-heavy workloads at scale
- One-click fleet provisioning that sizes PgBouncer processes to available cores and wires peering automatically
- Live bottleneck detector that distinguishes 'pooler is the wall' from 'Postgres is the wall' using pgbench-style probes
- Continuous throughput benchmark vs. your last config with regression alerts on cancel-request health
- Pluggable deploys: sidecar for Kubernetes, hosted endpoint, or BYO EC2
ClickHouse's 4x throughput result demonstrates the gap is wide and most teams don't know they have it; AI-driven workloads are pushing Postgres connection counts higher than ever.
PgBouncer is single-threaded and caps at one core is a widely documented scaling pain point; ClickHouse's own benchmarks show single-process PgBouncer peaks at ~87k TPS vs ~336k for the fleet, and multiple 2025-2026 articles describe '1,000+ connections hit Postgres at once' as a routine production failure mode.How we scale PgBouncer in ClickHouse Managed Postgres ↗PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale ↗
Whitespace is narrow: AWS RDS Proxy, Google Cloud SQL Managed Connection Pooling (announced at Cloud Next '25), Azure built-in PgBouncer, Akamai/Aiven managed PgBouncer, Cloudflare Hyperdrive, Supavisor, PgCat and Crunchy Data all occupy the space; ClickHouse now ships the so_reuseport+peering technique by default in its managed Postgres, commoditizing the core innovation.PostgreSQL Connection Pooling in 2026: PgBouncer vs PgCat vs Supavisor vs RDS Proxy vs Pgpool ↗Managed Connection Pooling overview | Cloud SQL for PostgreSQL ↗Postgres at Scale: Running Multiple PgBouncers - Crunchy Data ↗
Hard to monetize standalone: PgBouncer is open source, trivial to self-host, and hyperscalers bundle pooling into managed Postgres (e.g., Google's Managed Connection Pooling is part of Cloud SQL, ClickHouse ships so_reuseport+peering by default); RDS Proxy is priced around $0.015/ACU-hour, anchoring willingness-to-pay low for a pure-pooling layer.PGBouncer: Connection Pooling for Managed PostgreSQL Databases - Akamai ↗How we scale PgBouncer in ClickHouse Managed Postgres ↗
Underlying need is durable — Postgres connection pressure grows with AI/serverless workloads — but as a standalone product lifecycle is shaky because every major cloud is folding this capability into their managed Postgres offering, making a third-party pooler increasingly a feature, not a company.Connection pooling best practices - Azure Database for PostgreSQL ↗How we scale PgBouncer in ClickHouse Managed Postgres ↗
Technically feasible: PgBouncer, so_reuseport and peering are well-documented open building blocks and ClickHouse's blog lays out the exact architecture (cgroup sizing, cancel-key forwarding, max_client_conn splitting); the harder parts are operational — auth/SSL passthrough, HA, multi-region, observability, and integrating with any customer's self-hosted or cloud Postgres.How we scale PgBouncer in ClickHouse Managed Postgres ↗