PgDog is funded and coming to a database near you
Points and comments are a snapshot, not live.
PgDog, a Postgres sharding proxy written in Rust, announced $5.5M in funding and claims 2M queries per second in production.
PgDog is a proxy layer that makes PostgreSQL horizontally scalable through sharding, load balancing, and connection pooling. The three-person startup has deployed it across dozens of production systems handling over 2M queries per second and 20TB of sharded data. The product is open source (1.4M Docker pulls) and released weekly. Founder Lev Kokotov previously scaled Postgres at Instacart during a 5x growth period. PgDog runs anywhere: on-premise, in cloud accounts, or on laptops with no dependencies or hidden serverless costs. The funding from Basis Set, Y Combinator, and Pioneer Fund provides years of runway. An Enterprise edition with SLA-backed support is also available.
The core value proposition is enabling single-machine Postgres to handle workloads that previously required alternatives like MongoDB or DynamoDB. The technology addresses Postgres bottlenecks including WAL writer throughput limits, vacuum contention, and single-primary failure points through bulkheading across multiple primaries.
What commenters are saying
Commenters confirmed the legitimacy of the 2M QPS production claim and asked clarifying technical questions. Key findings emerged: cross-shard writes use two-phase commit with consistency trade-offs, reads are eventually consistent, and the proxy caches query ASTs (memory usage ranges from minimal for pooling-only to 1GB per pod with heavy query parsing). A developer shared successful multi-tenant Kubernetes deployment using a custom controller to manage pgdog configuration atomically across instances with 2-3 second downtime.
Comparison with pgbouncer and Citus drew discussion. One user switched from pgbouncer without measurable performance difference at low traffic. Citus comparison noted that PgDog's threaded Tokio concurrency outperforms Citus's process model for OLTP, though Citus remains stronger for OLAP with its advanced query engine. Requests included easier distribution methods (binary downloads, Homebrew, crates.io) and better Kubernetes-native configuration handling.