PlanetScale launches Neki, a sharded Postgres that keeps real Postgres on every shard

PlanetScale has opened a platform preview of Neki, a horizontally sharded Postgres offering built on the sharding expertise the company developed running massive MySQL fleets (its Vitess lineage). Applications connect to a Neki router that speaks the native Postgres wire protocol, so existing drivers, ORMs, and connection strings work unchanged. The router parses queries, plans which shards to hit, and merges results, while each shard runs as a full Postgres cluster with a primary and at least two replicas spread across three availability zones. Because there is no custom storage engine, extensions, SQL semantics, and performance behave like stock Postgres — a deliberate contrast with ‘Postgres-compatible’ distributed databases that hide the shard key and drop extension support.

Users define a JSON data topology that maps logical tables to physical shards, choosing the shard key, hashing, and how tables group across shard groups. Operations that normally demand maintenance windows — schema changes, version upgrades, failovers, imports, and resharding — run as built-in online workflows over the same psql connection. Neki also carries over PlanetScale features like Insights, schema recommendations, branching, and MCP, and adds tighter connection pooling via sidecars that control both ends of the connection rather than bolting PgBouncer on front. Notably, teams can run Neki unsharded as a single primary with replicas and treat resharding as a workflow they run later, avoiding a day-one commitment to a sharding strategy.

The pitch targets teams hitting the ceiling of a single Postgres machine, where vertical scaling eventually runs out and problems like vacuuming huge tables, slow backups, connection limits, and transaction-ID wraparound compound. This is a platform preview, not production-ready: PlanetScale warns of breaking changes and advises against real workloads for now. It signals the company’s push to bring its sharding operational model — long its MySQL differentiator — to the Postgres market it entered roughly 18 months ago.