MDEV-21117 has implemented correct binlog crash recovery in semisync replication. The server decides how to recover based on the role it has in the replication topology. In MDEV-21117 it deduces the role based on rpl_semi_sync_slave_enabled and rpl_semi_sync_master_enabled variables.

This leaves an important use case open. Typically not all slaves are semisync. If one has a mix of semysync and async slaves and the master crashes and after a failover wakes up as an async slave it won't have rpl_semi_sync_slave_enabled set and won't be able to recover correctly, because it won't be able to deduce its replication role.

We need to introduce the dedicated option that tells the server its initial¹ role in the replication topology. Perhaps the server can also automatically configure other settings, as appropriate for the specified role.

¹) initial — because the role can change dynamically, but such a change has no impact on crash recovery.