I have master/slave setup with two MariaDB 10.x instances. They were bidirectional but for the past two months it was only one direction.
When re-enabling the bi-directional it looks as if the master (where slave was being enable) lost data on some tables for exact period of time it was not setup as a slave. These are InnoDB tables, replicated with statements.
The master, where slave was being enabled, appears to be rolling back to an earlier snapshot - on a table by table basis. The resultant effect was a set of record disappearing from the table up to a certain PK as well as PK counter was set back to the same position.
The exact period of time lost for the effected tables was two months. The binary logs were expiring after 3 days.
I haven't found any documentation which shows this possibility, maybe somewhere in the replication code?