Re: mysql 5.7 replication for slave_paraller_type

[Rick James]

In MVCC, the "timestamp" (I think) equates to a snapshot of the database. If two transactions start about the same time, but nothing changes between the starts, then they are both looking at the "same" snapshot.

Once those two transactions arrive at the slave, then they may as well run in parallel; it is guaranteed that nothing changed from one snapshot to the other.

In my experience, your TPS would have to be very high to see this happening much. Or have certain types of workload. What is your workload like? DW? OLTP? OLAP? Logging? Financial transactions? etc?

I don't know what "prepare" means in this context.

Splitting on database is useless if you do everything in one database. Galera took a different approach (neither database, nor timestamp); but the code under the covers is more complex.