PostgreSQL Scale-Out at inMobi
summary
Postgres runs the ad stack at scale: 8 billion ads a day and ~240 TB ingested daily, growing from one Postgres 8.3 node to 100+ instances across six data centers and peaks of ~45,000 qps. Infrastructure shifts and tools are concrete — slony to streaming replication (Postgres 9.2), ELB/EC2 AMIs that spin replicas in 10–15 minutes, PG bouncer and load balancers, Hadoop for aggregation — and real pain points remain like 800 WAL files per hour and cross‑DC replication, with streaming replication largely solving slony’s bandwidth and operational fragility.
AI-written summary · about
chapters (16)
- 0:00PostgreSQL at inMobi
- 1:17Mobile ad scale and impact
- 3:06Single DB to global infra
- 7:26OLTP vs OLAP challenges
- 10:59Git + Flyway schema versioning
- 12:23US data center for low latency
- 13:24Slony replication hurdles
- 16:32Service-oriented architecture
- 20:00Load balancers and pooling
- 23:10Aggregation jobs to Hadoop
- 26:04Slony to streaming replication
- 28:14Stored procedure versioning
- 31:12Replication delays and checks
- 35:02Migrating to AWS
- 38:42Current challenges and roadmap
- 47:25Q&A
related talks
36mPostgres at any Scale
51mScaling Postgres to the next level at OpenAI
46mBig databases, many servers, data center or cloud - it's all here!
48mBuilding Petabyte-Scale PostgreSQL Deployments
13mBuilding reliable infrastructure with Postgres
37mPostgres at Any Scale
30mManaging Postgres at scale: Challenges, Tools & Techniques
54mBreakout Session