I have a three node cluster with a quorum disk attached to two of them. This makes number of quorum votes = 4. It is my understanding that there should odd number of votes in a cluster. So I would like to increase the number of votes assigned to QD to 2. Is this possible? If not, what are the alternatives?
Is there any reason why you can't connect the quorum disk to all three nodes? That would be the ideal solution.
What behaviour are you trying to achieve? You don't actually need a QD in a 3 node cluster, if you don't need the cluster to survive when two nodes go down.
If you are running SC3.2 you could always implement an external quorum server instead and that would then have the right number of votes so that all nodes were capable of remaining alone in the cluster.
There was also a flexible quorum vote option but I'm not sure I like it simply for managability purposes in a production cluster. It just makes things non-standard.
Tim
The solution I'm trying to create is not restricted to 3 nodes. In the current cluster I can connect all the nodes to the QD but when the cluster grows to 4, 5 or 6 nodes or more, it is not possible to connect all of them to the shared disk. And I would still need a way adjust the QD vote count if possible. Which takes me back to the same question, is it possible?
Not using the shared disk is not an option.
Actually, I've just re-read your original message and your assumption is wrong. You don't have to have an odd number of votes, but you will need (1/2 + 1) of the total votes to survive. You can probably work out the various failure scenarios.
You cannot, as far as I remember, increase the number of votes the QD has, but you can put the QD into a maintenance state. See man scconf.
Tim
Even if this sounds mean, you should try to come up with a list of node failure scenarios and an appropriate (reconfiguration) action. E.g. in you scenario with 3 nodes and 1 QD connected to 2 nodes only, what are the scenarios that you want to cover. If one node fails, everything is fine! If the 2 nodes fail and those nodes are the ones connected to the storage, you lose your data, so no worry about the 3rd node. If 2 nodes fail and one of them is not connected to the storage, bad luck, only 2 votes out of 4. But as nodes rarely fail in parallel, you could set the first failed node into maintenance mode, which would lower your total votecount to 3.......
An external quorum server would solve all of this.
The customer engineer who I worked with a while back came back with a 20p document, describing all the various alternatives with votes settings. And in the end an engineer needed to come up with the one and only solution to solve their very specific problem. If I were to change votes on the fly, I can guarantee you I would shoot myself in the foot.
