[Linux-HA] DBRD - split brain - and HA is happily migrating

[Thomas Glanzmann]

Hello Dominik,

> In DRBD or in the entire cluster?

just the DRBD. And I think that I produced the situation exactly as you
said using (drbdadm primary).

> You didnt give your drbd.conf, but I suppose you do not use DRBD
> resource fencing. Without resource fencing, it is perfectly possible
> to execute "drbdadm primary $resource" on a disconnected secondary.

global { usage-count no; }
common { syncer { rate 100M; } }

resource postgres {
        protocol C;

        startup {
                degr-wfc-timeout 120;
        }

        disk {
                on-io-error pass_on;
        }

        on postgres-01 {
                device     /dev/drbd0;
                disk       /dev/sda3;
                address    172.17.0.21:7788;
                meta-disk  internal;
        }

        on postgres-02 {
                device     /dev/drbd0;
                disk       /dev/sda3;
                address    172.17.0.22:7788;
                meta-disk  internal;
        }
}

> Take a look at the resource fencing function in DRBD. The primary will then 
> use another communication path to set the secondary resource to an 
> "outdated" status. "drbdadm primary" on an outdated resource will not 
> succeed unless you manually force it (the DRBD RA does not do that).

> Read the "man drbd.conf" section about it and this link is also worth
> a read: 

> http://blogs.linbit.com/florian/2007/10/01/an-underrated-cluster-admins-companion-dopd/

Thanks. I adopted my configuration and test if it works as desired.

        Thomas