[Pacemaker] DRBD Master/Slave in a 3 node cluster

[James Oakley]

On Tuesday, October 8, 2013 "Stefan Botter" <listreader at jsj.dyndns.org> wrote:
> James, compare my configuration with yours.
> Take a look especially on the location and colocation contraints.
> 
> What did you try in the meantime?
> Start from bottom up, with a fresh configuration, and then add
> resources and constraints one by one.
> 
> 
> Anyway, I suspect the following lines are wrong:
> colocation col_res_Filesystem_arthur_ms_drbd_4 inf:
> res_Filesystem_arthur ms_drbd_4:Master
> colocation col_res_Filesystem_backup_ms_drbd_1
> inf:res_Filesystem_backup ms_drbd_1:Master
> colocation col_res_Filesystem_breadmaster_ms_drbd_3
> inf:res_Filesystem_breadmaster ms_drbd_3:Master
> colocation col_res_Filesystem_media_ms_drbd_2 inf:
> res_Filesystem_media ms_drbd_2:Master
> 
> In my mind they should look like
> colocation col_res_Filesystem_arthur_ms_drbd_4 inf: ms_drbd_4:Master
> res_Filesystem_arthur:Started

I just cleared the config and started again. I had created the config using LCMC, since it was DRBD-centric, but I recreated parts of it manually. However, since I have a 3-node cluster with a variety of different services, I had disabled the symmetric_cluster option, which tells it not to run anything that does not have an explicit positive location constraint. I was able to get a master slave working with the option off on my new config by creating location constraints just like I had before. I still can't figure out what made it refuse to run slaves before. The same colocation rules I had before are working now.

I'm thinking now that I should switch to a symmetric cluster and use negative constraints to keep the resources off the 3rd node. That will certainly make the configuration simpler. I will rebuild the whole config manually. Hopefully that will work better.

Thanks,

-- 
James Oakley
jfunk at funktronics.ca