Lessons from 4 Years of writing the driver for MongoDB
summary
Four years of driver work reduces to concrete lessons: a JavaScript BSON parser proves the hotspot so a C++ parser gets written and speeds parsing but creates maintenance pain until parsing stays native while types revert to JS objects; 64-bit ints use the Google long closure library; node Mongo usage jumps 20,000 to 200,000 npm installs per month; slab-allocate buffers beat many small allocations; connection buffering trades latency for durability and invites disk-offload ideas; add a toBSON hook, run tests in a version grid, and build reverse code coverage.
AI-written summary · about
chapters (23)
- 0:00Intro
- 2:32First Commit
- 3:20IT was crap
- 3:52Node was liquid
- 4:38Types Oh Types
- 5:43Features
- 7:51Making JS BSON faster Buffers
- 9:58Pool handling
- 10:14Write concerns
- 11:01Reconnecting
- 12:01Custom Serialization off Object to BSON
- 12:51Replicaset
- 15:41Windows Build Hell
- 17:46Regrets
- 17:56Should have
- 19:36Testing it ALL
- 19:503D Reality hits
- 20:18A New Beginning
- 22:47Documentation
- 24:28On Backwards Compatibility
- 25:30On Managing The Project
- 27:43Take aways
- 28:18In Conclusion
related talks
48mGetting Started with MongoDB & Rust
22mNode.js Applicative DoS Through MongoDB Injection
33mReaching Ludicrous Speed
34mThe Road to Hype: How NoSQL became Serverless
35mPainless Data Storage with MongoDB and Go
53mMastering Node.js Performance
51m5 boring performance improvements in MongoDB 3.4 that matter
25mA JavaScript Parody: A Fun Journey through Hot Topics of Node.js and JS