we have a Galera (wsrep_provider_version 4.23) cluster over 3 MariaDB 11.8.3 nodes. Our application has a number of migration scripts (about 50) that run on startup, via Flyway. At some point in the migration process, the scripts hang on alter table statements, waiting for a lock on a query that is in status Waiting for certification according to information_schema.processlist. As far as we can see, this is not related to node health, or load/traffic (which is currently very light in the cluster).
Our theory is that we've encountered an issue in Galera similar to https://jira.mariadb.org/browse/MDEV-34784 , but we've come across some information indicating that Galera doesn't like alter table statements in the same scripts as create table statements. So we tried splitting the migration scripts into files with only create table statements, and other files with all the alter table statements. These go through without issue.
Is it a known best practice to not have alter table statements in the same DDL scripts as create table statements in Galera?