// the take
SPQR is a stateless proxy for horizontally sharding PostgreSQL, built by Yandex Cloud's Data Platform team for their own terabyte-scale, six-figure-QPS workloads. It routes, pools, and coordinates queries across shards via separate router/coordinator/balancer processes, aimed at teams who've outgrown a single Postgres instance and want a path off the monolith without rewriting the app to talk to multiple databases directly.
Supports both hashed and range sharding with online, automatic rebalancing between shards - this is routing plus actual data movement, not just query dispatch. It also handles cross-shard queries and distributed transactions, the part most Postgres sharding proxies skip entirely. Production concerns are covered too: Kerberos/GSS and LDAP auth, TLS, hot config reload, Prometheus metrics, and a real test setup (unit, regress, e2e via Docker Compose) plus a TPC-C benchmark suite.
You're taking on a distributed system, not installing a drop-in proxy - router, coordinator, and balancer are all separate stateful processes you now have to deploy and watch. Cross-shard queries and distributed transactions on Postgres are leaky abstractions by nature; features like 'reference tables' and 'default shard' are workarounds for real limits on what you can do efficiently once sharded, and your schema has to be designed around the shard key up front. Development and production use look concentrated inside Yandex Cloud, so the external contributor base is thin relative to the star count and you're riding one team's roadmap. There's also no visible tooling for reversing course - once you shard or need to change the shard key, that's a heavy, mostly manual migration.
We read the README, the directory tree and a key source file in October 2026. Nobody pays to be listed. How we judge →