mariaDB(mysql) replication after slave is up after down

Question

I have remote host (centos6) with mariadb (10.0.17-MariaDB-log - MariaDB Server) as master:

server-id   = 1
log-bin=mysql-bin
binlog_do_db = mydatabase

and local (win8.1) with mariadb (10.0.16-MariaDB-log - mariadb.org binary distribution) as a slave:

server-id = 2

As initial procedure I've dumped database on remote host, imported it on local host, then executed SHOW MASTER STATUS, get filename and offset and run:

CHANGE MASTER TO MASTER_HOST='$host', MASTER_USER='$user', MASTER_PORT = $port, MASTER_PASSWORD='$pass', MASTER_LOG_FILE='$fname', MASTER_LOG_POS=$pos
STOP SLAVE
START SLAVE

Replication starts. Everything I do with table on master is reflected to slave.

But if slave goes down, after it is up no changes (made on master) while slave was offline reflected to slave! So it looks like my slave should always be online, but it's a laptop! However after slave is up realtime replication still works - it doesn't remember all changes from offline, but if I change database on master when slave is online and started all changes are perfectly reflected to slave. Of course I know that replication is statement based, so I get not data diff but instructions. But I thought master remembers what it sent and what did not. So on next operation it just sends all non-delivered changes. Am I wrong?

My replication scenario: master server interacts with clients (mobile devices) and they change the database. From time to time I launch my laptop, start replication, get updated database and do some heavy analysis (it's too hard for my 2-core cheap server). Maybe there is a better method? Is there a way to get "offline changes" like in ICQ messenger? :) For now I can see only one solution - full db dump, but it is inconvenient, takes too much time and loads master heavily.

Answer 1

While the Slave is not connected to the Master, the Master is writing to its binlog(s). The Slave has remembered where it left off in reading from those binlogs. When the Slave reconnects, it picks up "where it left off", copies the changes from the Master's binlog(s) to the Slaves relay-log(s) and performs them. This "catchup" process will take a little time, how long depends on a lot of factors.

Do SHOW SLAVE STATUS; on the Slave to verify that it is connected and running ("Yes").

Normally, the version of the Slave should be no older than that of the Master. (I doubt if there is any issue between 10.0.17 vs 16. Nothing in the changelog for 10.0.17 jumps out at me.)

Are you using "parallel replication"? See bug fixed in 10.0.18 .