search

zhouyh139 / mysql-master-ha

Automatically exported from code.google.com/p/mysql-master-ha
0 stars 0 forks source link

Some issues with trying to use MHA #81

Open GoogleCodeExporter opened 9 years ago

GoogleCodeExporter commented 9 years ago 
Ahoy, 

I've been having some issues with MHA getting it to fully work in an 
environment to provide a functional failover system.

Basically there are (in a testing environment) four machines, 1 MHA manager 
(MGR), 1 MySQL master (MST), 2 MySQL slaves (SLV1 & 2). 
Once I shut down MySQL on the master server, the failover kicks in immediately 
and very fast, which is great. 
Up on the completion of the failover, one of the slaves is recognized as the 
new master (SLV1) , and SLV2 now
recognizes the new master as well. At this point MHA exits. Is that intended?

From now on, when I want to restart MHA with the current setup post-failover, 
it will immediately exit requiring all machines to be started as per config 
file. 
Using the remove_dead_from_config option alleviates this a bit, but once I want 
to bring up the original MySQL server, that one's of course missing and will 
require 
readding back to the conf file.

When starting up the original MySQL master, MST, and leaving it as it is 
intending to get it set back as master for the cluster, MHA will 
fail to start up due to having two non-slave servers alive.

Using the masterha_master_switch command to run a manual failover passing the 
options to set MST back again as master, everytime SLV1 will fail to properly 
recognize MST as 
Master again and will have error messages (usually connection errors) when 
running SHOW SLAVE STATUS \G, while SLV2 will replicate without a problem.

In the case of using a machine with a VIP, how can this be specified in the 
masterha.cnf file? 
Listing each machine in the .cnf file per regular hostname I get an error as 
the hostname in the file does not match the hostname/ip in the slave settings 
on the MySQL slaves.
Replacing the hostname with the vip-name, how will this work in a failover to 
get the VIP-name functional with the new master without mysql or MHA dying?

Basically in the long run, is there a way to use MHA in a fully automated way 
where human input would be as good as not needed at all?

Will try to get some log file outputs attached soon, they've been nuked 
currently for some other testing...

Cheers :)

Achim

Original issue reported on code.google.com by achim.re...@rightster.com on 14 Apr 2014 at 5:22