Doctolib Needs a More Powerful Database. Okay, ... ?
summary
Doctolib describes operating a single 25 TB Amazon Aurora PostgreSQL database (one primary with ~13 readers, 96-core writer, ~250k commits/sec, largest table ~7 billion rows) and evaluates replacements: Google Spanner, YugabyteDB, and Citus. Testing with real traffic found Spanner would force application rewrites and create vendor lock‑in, Citus (on Azure) breaks many joins and amplifies connection issues, and Yugabyte requires heavy planner tuning and per‑query hacks; the chosen path is not to replace but to split data functionally, use FDW and deepen PostgreSQL expertise.
AI-written summary · about
chapters (9)
- 0:00Start
- 3:20Slow and incremental evolution
- 4:35Why AWS Aurora?
- 6:52PG Expertise on Applicative side
- 9:45Monthly Load Tests
- 11:13Turning point
- 14:05What bigger database?
- 19:12Spanner
- 37:16Final takeaway
related talks
46mBig databases, many servers, data center or cloud - it's all here!
49mPostgreSQL’s Journey to Distributed Databases
38mWhat is the right database for you?
54mBreakout Session
40mDB for next project? Why, Postgres, of course
44mContinuous Database Administration
36mPostgres at any Scale
43mJust Use Postgres.