MySQL/MariaDB Parallel Replication: inventory, use cases and limitations
summary
Inventory of MySQL/MariaDB parallel replication, specific mechanics and real Booking.com benchmarks: MariaDB uses domain ID tagging (10.0) and master-side group commit IDs (10.0/10.1), MySQL 5.6/5.7 use schema or intervals, and 10.1 adds optimistic parallel replication that kills conflicters and retries. Benchmarks show one production batch workload moving from ~100 tps single‑thread to ~600 tps with tuned 10.0 (6×), while other environments see 1.2–2.5× depending on durability, big transactions, intermediate masters and grouping quality; an InnoDB race is found and fixed.
AI-written summary · about
chapters (36)
- 0:00Introduction
- 0:58Single threaded replication
- 1:27Bookingcom
- 1:52Agenda
- 2:29Why parallel replication
- 3:29Structure of the talk
- 4:39Running transactions
- 6:10Out of order commit impacts
- 7:56Slave checkpoint
- 9:02Parallel replication
- 19:19Big transactions
- 20:53Intermediate masters
- 21:25Avoid intermediate master
- 21:36Slave group commit
- 22:41Issue parameters
- 24:09Bin log
- 25:54Slave deadlock
- 27:51Optimistic replication
- 29:11Limitations
- 29:35Logical clock
- 30:36Slave commit order
- 30:52Write sets
- 31:04Right set
- 31:24Testing
- 32:27Commit rate
- 34:11Speedup
- 35:51Other environments
- 37:14Conservative
- 40:40Booking
- 40:56Real deployment
- 43:52Bugs
- 45:20Questions
- 48:41Benchmarks
- 48:53Still bugs
- 49:18More talks
- 49:48QA
related talks
29mEverything You Ever Wanted to Know About MariaDB Replication but Were Afraid to Ask
40mMySQL at Scale: Booking MariaDB to MySQL Migration
35mMySQL HA, Recovery and Load Balancing
22mMySQL Replication: Latest Developments
41mMariaDB The New MySQL
49mMySQL Replication — Advanced Features
14mThe State of the Art of MySQL Replication
17mState of MariaDB