Building Distributed OLTP with Azure Cosmos DB for PostgreSQL
summary
Demonstrates building OLTP on Azure Cosmos DB for PostgreSQL using the open‑source Citus extension and Postgres SQL: default 32 shards sharded by a distribution key, ability to start on one node and scale to many, and online rebalance and tenant isolation operations that move hot tenants between nodes without downtime. Shows a measured case: single node ~15–16k orders/min at ~368ms vs distributed cluster ~214k orders/min at ~39ms (≈13× throughput), plus limits like no global unique index on non‑shard keys.
AI-written summary · about
chapters (19)
- 0:00Intro
- 1:33Agenda
- 2:10Evolution to Distributed SQL
- 3:48Comparing Sale-Out Architectures
- 5:28Cosmos DB for PostgreSQL - Sharded Nothing Data Architecture
- 8:21Distributed Systems are NOT ONLY for large data volumes
- 10:06Distributed Systems are not "big" single node systems
- 13:30Creating distributed tables-2 step process
- 14:28Creating distributed tables- INSERT reords
- 19:23Building distributed databases
- 21:14Distributed Table Types
- 23:35Filter on shard key
- 27:56Primary/foreign key checks within shard
- 35:15Unique constraints within shard/node
- 35:28Joins within compute node
- 43:01Workload Rebalancing
- 44:08Online Tenant Isolation
- 47:35High Availability
- 47:45Built-in Monitoring
related talks
27mMulti-tenant Saas apps made simple on Azure Cosmos DB for PostgreSQL
16mOnline Scale-Out and Cloud Data Ingestion
24mCitus from the Customer eyes
08mFoundations of Distributed PostgreSQL with Citus
02h 02mAzure Cosmos DB: Liftoff
13mHow to do Data Modeling for Distributed Postgres in Azure Cosmos DB
18mBuilding Performant Data Architecture for Multi-Tenant SaaS Apps
31mKEYNOTE: The Distributed PostgreSQL Problem & How Citus Solves it