Re: [suggestion]Is writing relay log necessary in slave?

[Rick James]

There are potentially two replication-related logs on a Slave:

* The "relay_log(s)" are absolutely required -- they hold the stuff sent from the Master's "binlog(s)". Yes, it is a staging area, but usually it involves nearly zero overhead. If replication gets behind, then it becomes a safety measure -- things can be sent from the Master to the relay_log rapidly even if they cannot be executed immediately. That way, if the Master dies, there is less risk of any loss on the Slave. ("Semi-Sync" adds safety to this.)

* The binlog (on a Slave) is optional. It is the same concept as the Master's binlog -- namely it holds the writes. It is useful for
** Various backup things; incremental backups; etc
** Having a Slave hanging of this Slave -- a more complex topology of servers.
** Dual-Master -- two servers, each being both a Master and a Slave to each other.

Your comments/suggestions are focused on the Slave; the relay_log is focused on keeping the Master going.

> master's binlog file name and position to an transaction safe table

This, and similar things that are currently kept in files, may be coming soon in version 5.7. The design is a tricky problem of chicken-and-egg and/or runaway allocation and/or "transaction safe".

> avoid slave disk's relay log writing load

Check it out -- I rarely see more than 2GB of data going into a relay_log per _day_. How much do you have? 2GB/day is not much burden on a disk system. The LAN/WAN will probably be overloaded before the disk.