Hello Guys,

yeah, it's my first post in that great forum and I should express how it's very useful for me that's why I decided to contribute with you guys

but at the first I have a question about the Qdisk in RHCS

I have a 2 nodes cluster with ILO fencing device and using RHCS over RHEL5.4, I know that If split-brain happened it will end up with a fencing war and will end up with the 2 nodes rebooting

I heard that the Qdisk is using to solve this problem by enhancing the Quorum Majority Voting Algorithm, but I tried to understand that but so far I don't understand it

so is there anyone can explain that for me ?

Thanks a lot

Hi,

if your are using RHCS/RHEL why not open a call at RedHat?

hello mesiol,

1. I am asking for explanation not a support, so I don't need redhat in that, I need some people like you

2. I don't have a support contract with Redhat, I am only using the evaluation version to make some tests

3. If Redhat causes problem, so we can consider it as CentOS, is that good now

Thanks, and I hope to find someone who can help me

No Replies !!!

Even I am interested to see the answer to this question.
I hope we can get a reply soon.

Any update Gurus ?

Qdisk acts as a couple(*) of extra votes.

What this effectively means in a 2-node system is that if a node can't see qdisk it'll consider itself split and refuse to attempt to join the cluster until it's able to see qdisk.

It's normal to set qdisk's vote count to (Nodes-1), so that as long as you have disk and 1 node, you'll be quorate (half or more of the nodes present), (all nodes and no qdisk will get the same result)

In a 2 node system it's a special case and qdisk is the tiebreaker. For other situations it's best to have an odd number of nodes.

Qdisk can only handle 16 nodes. If you're using virtualisation it's normal to only have the physical machines talkng to qdisk and the virtuals to run as non-qdisk nodes - but in general it's better not to have virtual nodes as cluster members.