[corosync] Allow votequorum to degrade to a single node without qdisc?

Christine Caulfield ccaulfie at redhat.com
Thu Nov 7 09:02:09 UTC 2013


On 06/11/13 20:44, Jason Harley wrote:
> Hello —
>
> I have a four node Corosync (2.3.0-1ubuntu4) cluster configured and happy.  However, I’d like to be able to allow the cluster to degrade to a single node in an (extreme) failure state — but still require (at least) two nodes to establish the cluster.  Using qdisc is not an option in my environment, so I’m curious if I can achieve this using new features in ‘Needle’ like ‘last_man_standing’ and ‘auto_tie_breaker’.
>
> This is the ‘quorum’ block from my corosync.conf:
>
>> quorum {
>>          provider: corosync_votequorum
>>          expected_votes: 4
>>          last_man_standing: 1
>>          last_man_standing_window: 10000
>>          auto_tie_breaker: 1
>> }
>
>
> Is this as simple as telling votequorum that I have a two node cluster (‘two_node: 1’) and enabling ‘allow_downscale’?
>
>

What you have should be fine with the addition of "wait_for_all: 1". You 
probably don't auto_tie_breaker unless there is a likelihood of a 2+2 
split on the network.

"two_node" means you have exactly 2 nodes in the cluster (not ever more) 
so isn't relevant and allow_downscale is solving a different problem 
than you have.

With last_man_standing the cluster will survive being degraded to 1 node 
provided there is not a large outage and more than 2 nodes fail at a time.

the man page votequorum.5 has some good examples

Chrissie


More information about the discuss mailing list