MySQL for Messaging

on YouTube ↗24m

" summary

Facebook moves mobile messaging onto a three-tier queue that keeps recent items in memory, weeks of data on flash, and long term on disk, using iris queue fronting a MySQL flash layer. It runs millions of messages per second, targets single-digit millisecond reads and writes, and relies on MySQL 5.6 GTIDs and semi-synchronous replication plus multi threaded slaves and lightweight “log tailors” to get under 30 second automated failover with no data loss.

AI-written summary · about

" chapters (14)

1. 0:00Intro
2. 0:37Traditional messages
3. 1:51Mobile-to-mobile
4. 4:16Queue
5. 6:26Tiered Storage
6. 7:27Backend needs
7. 8:21Options
8. 9:53Why not
9. 11:17MySQL Replication
10. 12:22Multithreaded Slave
11. 14:20Semi-sync
12. 16:03Master commit path
13. 19:03Fast Failover
14. 20:27Big Picture

" related talks

• 25mData in the CloudMarten Mickos · O'Reilly MySQL Conference & Expo 2011
• 01h 01mPanel: Facebook's Distributed Database EvolutionKarthik Ranganathan, Dhruba Borthakur, Harrison Fisk, Jeff Rothschild, Victoria Dudin, Vishal Kathuria · Distributed SQL Summit 2019
• 43mMulti-tenant HBase Solutions at FacebookNicolas Spiegelberg · Berlin Buzzwords 2012
• 23mScaling MemcacheHerman Lee, Rajesh Nishtala, Hans Fugal, Steven Grimm, Marc Kwiatkowski, Harry C. Li, Ryan McElroy, Mike Paleczny, Daniel Peek, Paul Saab, David Stafford, Tony Tung, Venkateshwaran Venkataramani · NSDI 2013
• 55mThe Future of MySQLEd Huang · Percona Live 2022
• 47mLessons Learned from Our Main Database Migrations at FacebookYoshinori Matsunobu · SREcon18 Asia/Australia
• 50mInnoDB to MyRocks Migration in Main MySQL DatabaseYoshinori Matsunobu · SREcon17 Asia/Australia
• 43mHow WhatsApp moved 1.5B users across data centersIgor Istocniks · Code BEAM SF 2019