Running with Peacemaker about mysql-master-ha HOT 4 OPEN

And of course I meant Pacemaker ;)

If you code virtual IP failover logic inside a custom script and call it from 
MHA by "master_ip_failover_script" parameter 
(http://code.google.com/p/mysql-master-ha/wiki/Parameters#master_ip_failover_scr
ipt), you can do virual IP failover without using clustering software. I hear 
some users/customers have written such logic. 

> The document on 
http://code.google.com/p/mysql-master-ha/wiki/Using_With_Clustering_Software 
describes how to do that with a simple two node scenario but that wouldn't work 
for me...

I think that works in your case as well. By managing master and slave1 via 
Pacemaker, when master goes down you can do failover to slave1. MHA has 
parameters so that it always try to failover to slave1 (See 
http://code.google.com/p/mysql-master-ha/wiki/Parameters#candidate_master and 
http://code.google.com/p/mysql-master-ha/wiki/Parameters#no_master). If both 
master and slave1 go down at the same time, failover won't work, but in such 
serious cases automated failover should not happen.

If I don't use clustering software I may end up with two servers having the 
same IP address. 
This can happen if someone pulls out the network cable from the master (by 
accident), MHA will promote a different master and assign it the same IP.
With clustering the old master will detect that there is a new server in the 
cluster and it will never reuse the virtual IP.

The thing is I have more than just one slave and I don't want to fail over to a 
specific slave but to a slave which has the latest relay log events and this 
can't be defined with priorities but has to be determined dynamically.
With the setup described in  
http://code.google.com/p/mysql-master-ha/wiki/Using_With_Clustering_Software is 
states specifically which server would be a preferred master and which will be 
ignored. 
I want this to work with masterha_manager which will decide which slave to fail 
over to... 

> This can happen if someone pulls out the network cable from the master (by 
accident), MHA will promote a different master and assign it the same IP.

In such scenario, I highly recommend to force shutdown the original master, or 
doing nothing (do not start failover). Commands to force shutdown depend on 
H/W. Some samples are under samples/scripts/ from MHA Manager tarball. By 
shutting down the original master, you can avoid having two same IP address at 
least.

I understand your request to promote the latest slave, not a specific slave. In 
that case, implementing forcing shutdown logic and activating IP address logic 
in MHA failover script 
(http://code.google.com/p/mysql-master-ha/wiki/Parameters#master_ip_failover_scr
ipt) makes more sense to me. As described above, forcing shutdown logic highly 
depends on H/W. Activating IP address logic can be borrowed from clustering 
software.