Taking multi-master MariaDB for a spin, reproducing a Jepsen report

This article discusses Kyle Kingsbury's Jepsen report on MariaDB's Galera cluster, which revealed data loss during process crashes and lost updates/stale reads in healthy conditions. The author explains Galera as a multi-master database, how to run a three-node cluster locally, and translates transaction histories from the research paper into SQL. It covers differences between MySQL and MariaDB regarding transaction isolation levels and builds a small reproducer for lost updates, porting the reproduction outside Jepsen's Clojure ecosystem. The article also notes MariaDB's acquisition of Galera and the end of MySQL support, providing practical notes for testing Galera clusters.