Avoid overloading the master NIC on restarting IO_THREAD on lagging slave.

When GTID slave negotiation is enabled, the relay logs are wiped on starting the IO_THREAD. This might not be a big issue in most cases, but it is very bad on a lagging slave. I recently ran "STOP SLAVE; START SLAVE UNTIL ...;" on a GTID enabled slave, and it saturated the network interface of the master for more than one hour (this slave was lagging by ~4 days and had more than 250 GB of unprocessed relay logs).

I could have been more careful only restarting the SQL_THREAD (not the IO_THREAD), but there are still situations where restarting the IO_THREAD cannot be avoided (restarting MariaDB as an example).

If would be much better to avoid re-downloading binary log on starting the IO_THREAD as much of the relays logs are good on disk.

Some more details in the following:
http://jfg-mysql.blogspot.nl/2015/10/bad-commands-with-mariadb-gtids-2.html

Thanks,

JFG