[ClusterLabs] Problem with broken mysql-replication

[Niels Kobschaetzki]

Hi,

I solved it.

On 20/02/12 12:44, Niels Kobschaetzki wrote:
>Hi,
>
>I am having a problem repairing a broken mysql-replication. The way I
>want to use is to clone the database from the master with xtrabackup
>from percona, get up to date after reactivating the replication, stop
>the process and then unstandby the slave-node.
>That's my understanding after reading: https://github.com/Percona-Lab/pacemaker-replication-agents/blob/master/doc/PRM-setup-guide.rst#how-to-repair-replication
>
>The problem I am facing is that the replication immediately breaks when
>I unstandby the node.
>Pacemaker uses older positions for the binlog-file and the position
>compared to the reality and sets those when I unstandby the node.

<snip>

>How can I fix the replication of the database without taking down the
>master as well?

To fix the replication after you brought the slave up-to-date, you do a:

crm_attribute --name p_mysql_REPL_INFO --delete

Then a mysqladmin shutdown on the slave and then you can unstandby the
slave. After you did the unstandby you have to do a mysql -e 'start
slave;' because pacemake won't start it for you.

Afaik mysql_REPL_INFO is set by pacemaker when you standby the master,
thus automatic failover should still work.

Cheers,

Niels