Slave in replication lags behind the master, if he is far behind

MySQL 5.6.13
~200 requests per second
ROW-Based Replication

If you run the slave with a fresh dump, i.e. when the difference in data is small, it quickly catches up to the wizard and then everything is fine.
If the slave to stop for a couple of days and then again start it somewhere for a minute or two very jauntily begins to catch up (high CPU usage, I / o, all as provisions). And then almost falls asleep — CPU load 1%, a small I / o, and Seconds_Behind_Master is reduced very slowly, standing still or even increased.

Performance iron slave and higher versions of the software are identical. Errors in the logs no. Between servers Gigabit.

Where to dig?
October 3rd 19 at 03:43
2 answers
October 3rd 19 at 03:45
it seems understood
the problem in bcache — merged base on a clean SSD and everything flew.
very strange
%iowait remained about the same — 5.4
maybe bcache problem with multithreaded access because the synthetics all the chocolate.
in any case, the issue is closed, MySQL is not to blame :)
October 3rd 19 at 03:47
1. to make sure that replication works. That is, if there are any kind of error SHOW SLAVE STATUS\G in a variable Last_Error
2. To check the age of binari logs on the master. Maybe they 2 days time to leave.
1. exactly work, because Relay_Log_File and Relay_Log_Pos are growing. There are no errors.
2. Bina on the master, for sure, by the way, I realized that for the activity in the first minutes after start the slave it is these same Bina downloads itself locally (relay-log).
i.e. replication is working, but like the speed of something is constrained. Most likely it is connected with a ROW-Based, as usual never had a problem. - lilyan.Mayer commented on October 3rd 19 at 03:50

Find more questions by tags MySQL