Elkin, serg I need your opinions about this feature request. I see negative impacts for other threads, when the replication thread would get higher OS priority. If you see this as a useful enhancement, would it be a high effort to get it implemented?

No, it's a rather small task, low effort.

So this would be as a CHANGE MASTER TO option? The WL mentions IO starvation which is less of a problem as at least Linux nice levels correspond to IO priority levels for blocking relay log reads, however applying changes means io_reader threads could still cause delay. Seems xtradb had os_thread_set_priority.

Is there the case for the master's replication thread to also increase priority too?

I didn't think of that yet, first I'd like to understand whether thread priority will actually help. Like, create a load where CPU is 100% utilized so that replication lags. Then just hard-code pthread_setschedprio for slave threads, and see if that helps. If it does, we can come up with some syntax for that.

ralf.gebhardt@mariadb.com serg's practical suggestion makes sense. Needs to be done at least to learn about effects. An even if our result (of always limited tries) would be not so encouraging, they could be contrary elsewhere I think.