[ClusterLabs Developers] [Linux-ha-dev] Announcing crmsh release 2.1.7

[Kristoffer Grönlund]

Darren Thompson <darrent at akurit.com.au> writes:

> Just a quick question:
>
> If "scripts: no-quorum-policy=ignore" is becoming depreciated, how are we
> to manage two node (e.g. test) clusters that require this work around since
> quorum state on a single node is an odd state.
>

Hi Darren,

There are better mechanisms in corosync and Pacemaker for handling two
node clusters now while still maintaining quorum.

In corosync 2, we have the two_node: 1 setting for votequorum, which
ensures that a two node cluster doesn't suffer split brain (fencing is
required for this to work properly).

There is an explanation for how this works here:

http://people.redhat.com/ccaulfie/docs/Votequorum_Intro.pdf

Somewhat related, there used to be the start-delay meta parameter which
could be set for example for sbd stonith resources, to make a
double-fencing scenario less likely. This has now been replaced by the
pcmk_delay_max parameter. For an example of how to use this, see this
pull request for sbd:

https://github.com/ClusterLabs/sbd/pull/15/commits/ca2fba836eab169f0c8cacf7f3757c0485bcfef8

Cheers,
Kristoffer

-- 
// Kristoffer Grönlund
// kgronlund at suse.com