Need a way for PRM to work with xtrabackup's --safe-slave-backup option

[jayjanssen]

http://www.percona.com/doc/percona-xtrabackup/innobackupex/innobackupex_option_reference.html#cmdoption-innobackupex--safe-slave-backup

Xtrabackup will stop the slave here repeatedly to get Slave_open_temp_tables to 0. This conflicts with PRM, which restarts the slave when it stops automatically. I can't find a way in pacemaker to cease monitoring on a given node, but leave the resources running.

[lefred]

Maybe you can modify the resource monitoring operation during the process by adding:

on-fail="ignore" ?

[y-trudeau]

The real best would be to monitor the xtrabackup pid and verify if indeed the pid is xtrabackup. In that case, if the slave is stopped, ignore the error. I don't see any option regarding pid-file for xtrabackup, maybe it is something that could be added. Does that make sense?

[jayjanssen]

Maybe, but it is quite complex and inter-dependent on a fix with xtrabackup. If there's a solution to simply mark a single node as in maintenance-mode without stopping the services, that would seem a handy thing to have in general for PRM. But, perhaps this is a pacemaker limitation.

[jnewland]

I've actually added a 'monitor_mysql' node attribute that, if present, causes the mysql monitoring check to return success. Combined with another 'xtrabackup' attribute that the xtrabackup script sets, and some location constraints, I can be sure that we don't elect a node that is running a backup as master.

On Aug 14, 2012, at 10:33 AM, Yves Trudeau [email protected] wrote:

The real best would be to monitor the xtrabackup pid and verify if indeed the pid is xtrabackup. In that case, if the slave is stopped, ignore the error. I don't see any option regarding pid-file for xtrabackup, maybe it is something that could be added. Does that make sense?

—
Reply to this email directly or view it on GitHub.

[y-trudeau]

There're 2 ways of doing this, either the xtrabackup script set a crm attribute which need to be transient to avoid a pengine run or the mysql ocf script just react to the presence of a file. The xtrabackup script would touch a file that would basically prevent the monitor ops from restarting the slave (and set master_score to 0). It would then be the responsibility of the script to remove the file. Should we also set a timeout? I could stat the touched file and if now() - creation time is larger than the timeout, the file is ignored. Thoughts?