[DRBD-user] drbdmanage quorum control

[Roland Kammerer]

On Tue, Oct 03, 2017 at 12:05:50PM +0100, Yannis Milios wrote:
> I think you have to use 'drbdmanage reelect' command to reelect a new
> leader first.
> 
> man drbdmanage-reelect

In general that is a bad idea, and I regret that I exposed it as a
subcommand and did not hide it behind a
"--no-you-dont-want-that-unless-you-are-rck" where it then sill asks you
to prove the Riemann hypothesis before continuing...

> On Mon, Oct 2, 2017 at 2:12 PM, Jason Fitzpatrick <jayfitzpatrick at gmail.com>
> wrote:
> 
> > Hi all
> >
> > I am trying to get my head around the quorum-control features within
> > drbdmanage,
> >
> > I have deliberately crashed my cluster, and spun up one node, and as
> > expected I am unable to get drbdmanage to start due to the lack of
> > quorum,,
> >
> > I was under the impression that I should have been able to override
> > the quorum state and get the drbdmanaged process online using DBUS /
> > manually calling the service, but am drawing a blank..
> >
> > for the sake of this example it is a 2 node cluster node1 is online
> > and node2 is still powered off,
> >
> > [root at node1]# drbdmanage quorum-control --override ignore node2
> > Modifying quorum state of node 'node2':
> > Waiting for server: ...............
> > Error: Startup not successful (no quorum? not *both* nodes up in a 2
> > node cluster?)
> > Error: Startup not successful (no quorum? not *both* nodes up in a 2
> > node cluster?)
> >
> > Any advice?

Bring back the second node. In two node clusters that is the only clean
way to bring back the cluster. If you want quorum, get >=3 nodes.
Period. In two node clusters both have to be up. "reelect" is a last
resort command for the absolute worst case to bring up a 2 node cluster
where only one node survived and the other one is gone beyond repair.
"reelect" with a forced win alters internal state to make that possible.
It does not revert that internal state if, for whatever reason the
second node then shows up again. You would have to restart the "reelect"
node to get it then in a sane internal state again.

tl;tr: If you want quorum: >=3 nodes. Don't use "reelect" to force wins.

Regards, rck