The archive · Developer & Business Tools · Technical decision · 2025–2026
PgDog bets a Postgres proxy can shard without app changes; $5.5M seed, 2M qps
Ex-Instacart engineers built an open-source Postgres proxy that pools, load-balances and shards without app changes; $5.5M seed and 2M+ queries/s in production.
PgDog
What the business is
PgDog is an open-source Rust proxy for PostgreSQL that combines connection pooling, read/write load balancing and sharding in one executable; apps point at it via DATABASE_URL and shard Postgres without code changes or database migrations.
Starting capital：$5.5M seed round from Basis Set, Y Combinator and Pioneer Fund, announced 2026-06-10.
How it started
Lev Kokotov ran Postgres at Instacart, scaling it through the April 2020 demand spike by sharding on RDS, Aurora and EC2. After writing PgCat, a Rust Postgres pooler, he and co-founder Justin started PgDog as a fresh, fully asynchronous codebase with sharding as a first-class feature rather than an afterthought. An early Show HN ran in 2025.
What happened
The 2026-02-23 Show HN ('Scale Postgres without changing the app') drew 326 points and 63 comments, reporting the proxy in production with cross-shard aggregates, atomic two-phase-commit writes, built-in resharding over parallel logical replication, and failover-aware load balancing. On 2026-06-10 the team announced a $5.5M seed from Basis Set, YC and Pioneer Fund — that post drew 548 points and 261 comments — and disclosed 2M+ queries/s in production, 20+ TB sharded and 1.4M+ Docker pulls, with weekly releases and a growing Discord community.
No ending yet — it is still running.
Background
PgDog is an open-source Rust proxy that makes PostgreSQL horizontally scalable. It combines three tools in one executable: a connection pooler that shares a few Postgres connections among 100,000+ clients, a load balancer that routes reads to replicas and detects failovers, and a sharding layer that pulls the sharding key out of each query and sends it to the right database — all without application code changes or database migrations.
The bet is that Postgres is the only database teams need if its scaling problem is solved, and that the reason Mongo and Dynamo exist is precisely that one-primary Postgres has no easy scale-out path. Lev Kokotov, who ran Postgres at Instacart through the April 2020 demand spike, and co-founder Justin built PgDog as a wire-protocol proxy rather than an extension: sharding, aggregates, writes and resharding are handled in transit, so ORMs, prepared statements and session state keep working.
Traction built through Hacker News: the 2026-02-23 Show HN drew 326 points and 63 comments, and the June 10, 2026 funding announcement — a $5.5M seed from Basis Set, Y Combinator and Pioneer Fund — drew 548 points and 261 comments. By September 2026 PgDog reported 2M–3M+ queries per second in production, over 20 TB sharded and 1.4M+ Docker pulls, with customers including Modal, Ramp and Earth Genome.
What has to be true
- It attacked the root cause of NoSQL migrations — one-primary Postgres can't scale — rather than the symptoms, so the product promise is 'keep your database and your code'.
- Working at the wire-protocol level let it claim drop-in compatibility, which removes the single biggest switching cost in infrastructure: rewriting the application.
- It shipped open source from day one and let the community deploy it, turning Docker pulls and production queries into proof before the company even raised.
- The founder's Instacart experience gave the team a first-hand failure story (sharding by hand) that made the product credible to skeptical database engineers.
What can be applied
The most expensive part of scaling a database is often migrating to a 'better' one; a proxy that makes the incumbent scale — no app changes, no rewrite — captures teams that already trust Postgres.
Aftermath
As of September 2, 2026, PgDog is a three-person startup with seed runway, an open-source proxy released weekly on Thursdays, and a production footprint quantified on its homepage: 3M+ queries per second, dozens of deployments and testimonials from Modal, Ramp and Earth Genome. It is building an Enterprise edition for AWS with SLA-backed support. The bet is still being validated at scale — cross-shard query coverage is explicitly a work in progress — but the core claim, Postgres sharded without app changes, is already running in production.
Sources
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card