We could use m1, but the downside is that orchestrator doesn't support management of it's backend database cluster.
We've decided for now to use the tendril/zarcillo 'cluster' (db1115/db2093). As db1115 is running 10.1 (for tokudb support for tendril), we'll only create the orchestrator database on db2093 for now (which runs mariadb 10.4).
When we finally drop tendril we can upgrade db1115 to mariadb 10.4, and have replication. db2093 is already backed up, so we can always recover in case that node fails in the meantime.