MariaDB Scaling - From Single Instance to Multiple Clusters
summary
A concrete path from a single dev instance to a cross‑DC, multi‑cluster HA deployment: deploy a Galera cluster with at least three odd nodes (quorum), front it with ProxySQL and anycast for routing and failover, handle Galera deadlock errors by retrying full application transactions, and shard when a single cluster fills—acknowledging no foreign keys across shards so a shard manager is required and using ClickHouse for cross‑shard aggregation. This setup runs in production across ~40 services.
AI-written summary · about
chapters (11)
- 0:00Start
- 8:23I'm satisfied, let's ship it.
- 10:31Hey boss, let's start doing it properly...
- 13:13Master-slave replication
- 15:21Cross DC replication - hot-standby
- 19:25Cross DC replication - both active
- 20:37Galera Cluster
- 30:41ProxySQL
- 33:38Anycast routing
- 35:18Vertical scaling summary
- 45:42Sharding
related talks
25mThe future of databases: Combining operational and analytical data at massive scale
58m600k QPS on MySQL Galera cluster
35mMySQL HA, Recovery and Load Balancing
24mComparing the High Availability of MySQL with Modern Distributed SQL Databases
38mIs Percona XtraDB Cluster Right for Any Application
52mTo Shard or Not to Shard
55mProduction with ProxySQL and Orchestrator
27mDistributed MySQL Architectures: Past, Present and Future