PostgreSQL Distributed: Architectures & Best practices
summary
Compares concrete PostgreSQL deployments and their tradeoffs: cloud block storage (EBS) raises disk latency ~10x and drops IOPS from ~1,000,000 NVMe to thousands, hurting throughput and recovery; read replicas scale reads (examples up to ~100,000 reads/s) but introduce eventual/monotonic read anomalies and wasted cache across replicas; Aurora-like dbms-optimized storage uses six‑way WAL replication to remove local page writes but increases commit latency and IO cost; sharding (Citus) scales CPU and data but adds per-insert network roundtrips and makes shard key choice critical.
AI-written summary · about
chapters (10)
- 0:00Introduction and motivation
- 1:52Single machine PostgreSQL
- 6:07Network attached block storage
- 12:45Read replicas tradeoffs
- 19:48DBMS optimized storage
- 24:42Sharding architectures
- 31:50Active-active systems
- 35:25Distributed SQL overview
- 38:14Key tradeoffs summary
- 39:25Q&A session
related talks
43mPostgreSQL in the cloud: Theory and Practice
49mPostgreSQL’s Journey to Distributed Databases
25mData-intensive PostgreSQL: Three ways to scale
38mScaling PostgreSQL Replicas Under Load
38mPostgreSQL as a distributed computing platform
33mFive Sharding Data Models and Which is Right
39mAWS Aurora
36mBest Practices in PostgreSQL Tuning: Navigating Key Performance Bottlenecks in the Cloud