DigiNews

Summary

Jepsen analyzes MariaDB Galera Cluster 12.1.2, revealing that under default Galera configurations the cluster can lose committed transactions during coordinated node failures and exhibits anomalies like Lost Update and Stale Read even in healthy operation. It questions Galera's claimed isolation and instant replication, and recommends configuring innodb_flush_log_at_trx_commit=1 to reduce risk, while noting that MariaDB documentation can be unclear about actual guarantees.