The Question is:
Can you recommend a quorum file scheme to allow 1 of my 3 clustered nodes to
remain up, while 2 are down for maintenance? Would a Quorum file vote of 3,
system votes of 1, and expected votes of 4 do the trick and prevent
partitioning? Many thanks!
The Answer is :
What you seek is a three node cluster able to be reduced down to
one node with the orderly shutdown of two of its members.
There is no way to perform the shutdown of two arbitrary voting
nodes in a three-voting-nodes cluster, without direct operator
intervention. This is a variant of the classic two-node cluster
configuration, of course -- when you move from three to two nodes,
the clsuter will continue operating. When you have a two-node
configuration, you must have a primary-secondary configuration
or (if you have a shared interconnect) a quorum disk, or you
must have a plan to determine which lobe should be manually
resumed through operator intervention.
The two-node configuration is a matter of making sure that the
cluster quorum value is set to number of votes of the remaining
node after the other two nodes are removed. (For details on
EXPECTED_VOTES and VOTES, please see the cluster documentation
and the OpenVMS FAQ.) Also see the SET CLUSTER/EXPECTED_VOTES
command, and the Availability Manager (or AMDS) quorum adjustment
When you perform an orderly shutdown of a cluster member, the
specification of the REMOVE_NODE option will cause the quorum
value to adjusted downward.
See the OpenVMS Cluster Systems manual, where this is explained
in detail. The OpenVMS documentation is available at:
Having all members with EXPECTED_VOTES of 1 would allow any single
member to boot while the others are down. As the other nodes boot
into the cluster, the quorum value will automatically increase.
(This is effectively a partitioned cluster, and has the inherent
corruption-related risks of such configurations.)
Having all members with EXPECTED_VOTES of 2 would require you to have
two VOTES present at boot time, so that a single node with a bad
cluster interconnect could not form it's own single node cluster.
This still allows you to use the REMOVE_NODE option on an orderly
shutdown to get down to a single node cluster for maintenance.
The OpenVMS Wizard would tend to configure the cluster using the
expected and documented settings for VOTES and EXPECTED_VOTES (see
the OpenVMS manuals and in the OpenVMS FAQ), and would configure
and use Availability Manager or AMDS to manually resume the remaining
node for the specific case of the two-node shutdown.
If this process must be performed unattended and you know which
lobe will remain, bootstrap that lobe with the VOTES. You can
then reset the volumes and roll the lobes using the REMOVE_NODE
and/or SET CLUSTER/EXPECTED_VOTES commands.