Installation guide

In the event of a major network problem, cluster partitioning (aka: split-brain situation) can
occur. Each partition can no longer communicate with nodes outside its own partition. A Red
Hat cluster requires the quorum requirement be fulfilled before a status change in the cluster
is allowed. For example, quorum is required by the resource management system to relocate
cluster resources or for the CMAN module to remove nodes from the cluster.
The cluster partition is considered quorate if more than half of all votes within the entire
cluster belong the the cluster partition.
Q = V/2 + 1
where Q is the required number of votes for quorum and V is the total number of votes within
the cluster.
Although the quorum requirements calculation based on the active nodes in a cluster work
well for various cluster configurations and network issues, specific cases exist where the
cluster cannot decide or incorrect decisions have been made.
As such, the use of a quorum disk (qdisk) has been reintroduced.
4.4 Qdisk
In specific cases, the quorum requirement based on the nodes belonging to a cluster is
insufficient. In a two node cluster, the standard quorum calculation (Q = V/2 +1) would result
in two, considering one vote per cluster node. In the case of a highly available cluster, this
would make no sense. Therefore, the two node cluster is considered a special case and by
using the <two_node> configuration option, quorum is reduced to one. In this manner,
quorum is maintained and if one node should fail, the other will take over on its own.
One concern with this solution is in the case of a network loss between nodes, each node will
interpret the lack of connectivity as a failure of the other node. This problem is most
commonly referred as a split brain situation. As each node assumes it is the survivor, it will
attempt to fence the other node to prevent uncontrolled access to the shared storage
resources. In this instance, which ever node successfully fences the other first will become
the surviving member.
A quorum disk (qdisk) can be used to prevent this situation, bolstering the quorum by adding
an additional vote or votes to the cluster.
In a two node cluster configuration with a qdisk, the total expected votes would be three with
a quorum of two.
In small multi-node cluster configurations, other types of problems can occur. In a three or
four node cluster, quorum is two or three respectively, and losing two nodes can cause a
problem.
To resolve the small cluster quorum problem, a quorum disk with a vote count equaling the
number of cluster nodes minus one bolsters the quorum enough to enable the cluster to
survive with only one node remaining.
14 | www.redhat.com