# Percona replication controlled by Pacemaker

**URL:** <https://forums.percona.com/t/percona-replication-controlled-by-pacemaker/5707>\
**Category:** Percona Server for MySQL 5.6\
**Created:** [June 29, 2017, 7:11am UTC](https://forums.percona.com/t/percona-replication-controlled-by-pacemaker/5707 "2017-06-29T07:11:49Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![DBA](https://avatars.discourse-cdn.com/v4/letter/d/cdc98d/32.png) [@DBA](https://forums.percona.com/u/DBA)\
**Post date:** [June 29, 2017, 7:11am UTC](https://forums.percona.com/t/percona-replication-controlled-by-pacemaker/5707/1 "2017-06-29T07:11:49Z")

</div>

rpm -qa | grep -i pacemaker  
percona-pacemaker-agent-1.0.0-1.x86\_64  
pacemaker-libs-1.1.15-11.el7.x86\_64  
pacemaker-cluster-libs-1.1.15-11.el7.x86\_64  
pacemaker-cli-1.1.15-11.el7.x86\_64  
pacemaker-1.1.15-11.el7.x86\_64

rpm -qa | grep -i corosync  
corosync-2.4.0-4.el7.x86\_64  
corosynclib-2.4.0-4.el7.x86\_64

rpm -qa | grep -i percona  
percona-pacemaker-agent-1.0.0-1.x86\_64  
Percona-Server-client-56-5.6.36-rel82.0.el7.x86\_64  
Percona-Server-shared-56-5.6.36-rel82.0.el7.x86\_64  
percona-xtrabackup-2.3.8-1.el7.x86\_64  
Percona-Server-server-56-5.6.36-rel82.0.el7.x86\_64

uname -r  
3.10.0-514.10.2.el7.x86\_64

2 node Pacemaker cluster Master, Slave setup with STONITH enabled.  
Apache resource - Web service  
DRBD resource - Magento flat files for web service  
MySQL resource - Database with replication

Failover happens as it should between the servers. But we recently had an  
incident where database replication had broken and then the Master node  
failed over to the Slave (loss of data).

I have found a setting in the cib for  
\<nvpair id=“p\_mysql-instance\_attributes-max\_slave\_lag” name=“max\_slave\_lag” value=“60”/\>  
\<nvpair id=“p\_mysql-instance\_attributes-evict\_outdated\_slaves” name=“evict\_outdated\_slaves” value=“false”/\>

If I set this to “true” then it works as intended and won’t fail over to  
the Slave if the Slave replication falls 60 seconds behind the Master.  
However if I then need to rebuild the Slave from a backup of the Master, I  
will have to set the time so that I am able to rebuild the Slave before it  
fails due to the “max\_slave\_lag” value.

Does someone know of a better way of performing this? We need to perform  
patching on a monthly basis, so require failing over in the pacemaker  
cluster from Master to Slave.