600k QPS on MySQL Galera cluster
summary
Claim of 600k QPS (now >1M total) corresponds to a six-node Galera cluster serving a 300GB app database using Percona + Galera + HAProxy + Puppet + xtrabackup with wsrep_slave_threads=16 and gcache≈1GB. Nightly/incremental backups run every four hours, SST uses xtrabackup and wipes receivers, monitoring via PRTG and HipChat alerts. It works at scale but exposes painful specifics: SST fragility and logs, split‑brain risk, backup storage growth, and imminent need to shard beyond a single cluster.
AI-written summary · about
chapters (8)
- 0:00Start
- 3:12What is Puppet and why do we care?
- 8:25What is Galera Cluster?
- 9:52Terminology
- 15:42Galera Features
- 21:45Galera 'Features'
- 49:07Backups
- 52:18Current issues + what's next
related talks
35mMySQL HA, Recovery and Load Balancing
38mIs Percona XtraDB Cluster Right for Any Application
51mMariaDB Scaling - From Single Instance to Multiple Clusters
29mWhat breaks when Aerospike hits 6 million QPS
22mThe Art of the Possible
32mBest Practice in Design and Implementing MySQL Geographic Distributed, High Availability Solutions
37mScale MySQL beyond limits with ProxySQL
53mGalera Cluster Data Consistency