My replication setup is as follows:

• a two node Galera cluster as primary and secondary master
• a Maxscale binlog router as intermediate slave
• a real slave, replicating from the binlog router

The two Galera nodes are configured as:

[mysqld]  

server-id=1

binlog-format=ROW

log-bin=binlog

log-slave-updates=1

wsrep_on=ON

wsrep_provider=/usr/lib/galera/libgalera_smm.so

wsrep_cluster_name=test_cluster

wsrep_cluster_address=gcomm://10.47.47.11:4567,10.47.47.12:4567

wsrep_sst_method=rsync

wsrep_sst_auth=root

wsrep_gtid_mode=ON

wsrep_gtid_domain_id=23

wsrep_node_address=10.47.47.11

wsrep_node_name=node-1

Second node only differs by wsrep_node_address and wsrep_node_name.

The binlog router is configured like this:

[Replication]

type=service

router=binlogrouter

user=maxscale

password=secret

server_id=4711

mariadb10_master_gtid=1

binlogdir=/var/lib/maxscale/binlogs/

[Replication-Listener]

type=listener

service=Replication

protocol=MariaDBClient

port=4711

The two nodes are set up as primary and secondary master like this:

SET @@global.gtid_slave_pos='0-1-4'; -- 

CHANGE MASTER      TO MASTER_HOST='node-1', MASTER_PORT=3306, MASTER_USER='repl', MASTER_PASSWORD='secret', MASTER_USE_GTID=Slave_pos;

CHANGE MASTER ':2' TO MASTER_HOST='node-2', MASTER_PORT=3306, MASTER_USER='repl', MASTER_PASSWORD='secret', MASTER_USE_GTID=Slave_pos;

START SLAVE;

I issue the following statements on node-1 and node-2:

node-1: CREATE TABLE t1(id int primary key);

node-1: INSERT INTO t1 VALUES (1);

node-2: INSERT INTO t1 VALUES (2);

The slave now also has table t1 with two rows in it.

Now I stop the mariadb process on node-1, and check on the maxscale node that binlog router replication has indeed switched over to node-2.

Next I insert one more row on node-2:

node-2: INSERT INTO t1 VALUES (3);

SHOW SLAVE STATUS on the slave shows that IO and SQL threads are still running, but row '3' does not show up.

I do a "STOP SLAVE; START SLAVE;" on the slave, but row '3' is still missing. Now I insert a row '4' on node-2. This row gets replicated again, but the test table on the slave only shows rows '1''2''4'. The row that was inserted after node-1 was shut down, but before restarting the replication on the slave, remains missing there.