Presented by:

Deepthi has been working with databases and distributed systems for 20+ years. She has designed and implemented various database and application scaling strategies in diverse industries like Supply Chain Planning, Mobile Device Management and Managed Databases. She is currently a Founding Engineer on the Multigres team at Supabase focusing on High Availability for Postgres deployments. Prior to Supabase, Deepthi spent 7 years as a maintainer of Vitess, the open source MySQL Scaling solution that came out of YouTube, and for ~5 of those years she was also the Technical Lead for the project.

Sugu is the head of Multigres at Supabase. Multigres is a horizontally scalable Postgres architecture that supports multi-tenant, highly-available and globally distributed deployments, all while staying true to standard PostgreSQL. Sugu is also the co-creator of Vitess, a sharding solution for MySQL.

No video of the event yet, sorry!

PostgreSQL has become the database of choice for modern applications, but scaling beyond a single instance remains a significant challenge. While MySQL users have benefited from Vitess, the battle-tested horizontal sharding system born at YouTube that now powers some of the world's largest deployments, PostgreSQL users have lacked a comparable solution. Multigres changes that.

This talk introduces Multigres, an open-source project that adapts Vitess's proven architecture for PostgreSQL. Led by Sugu Sougoumarane, co-creator of Vitess, Multigres brings horizontal scalability, high availability, and intelligent sharding to PostgreSQL while preserving full protocol compatibility with existing applications.

The heart of the talk covers Multigres's two-level proxy architecture. The first layer, MultiGateway, presents your distributed cluster as a single PostgreSQL server, handling query routing, scatter-gather operations for cross-shard queries, and shielding applications from underlying failures. The second layer, MultiPooler, manages connection pooling and coordinates with individual PostgreSQL instances. Together with etcd for distributed coordination and gRPC for inter-service communication, these components form a cloud-native system designed for Kubernetes deployments.

We'll also explore MultiOrch, the orchestrator that makes high availability possible. MultiOrch monitors replication health across replicas, repairs broken streams, and coordinates failover when a primary fails, promoting an up-to-date replica automatically. MultiOrch operates on an unmodified PostgreSQL engine using synchronous replication. In multi-zone deployments, a MultiOrch instance is deployed per zone to ensure that failovers can succeed even during network partitions.

Who Should Attend This session is for PostgreSQL DBAs, backend engineers, and architects who are hitting the limits of single-instance Postgres or planning for future growth. Whether you're managing multi-tenant SaaS platforms, building globally distributed systems, or simply want to understand the next frontier of PostgreSQL scaling, you'll leave with a clear picture of how Multigres works and where it's headed.

Date:
Duration:
50 min
Room:
Conference:
Postgres Conference: 2026
Language:
Track:
Ops
Difficulty:
Medium