Cluster Node Failure

Question

Hi,
I am a Columbia University engineering graduate student doing research on Sun Clusters. I just have two quick questions that I was unable to find answers for on the Sun Cluster Documentation website. If any1 can help me answer these two questions I would really appreciate it. The questions are as follows.
1. If a node of a cluster fails, how are resources such as file locks or client sessions recovered, rebuilt, if at all? I understand that if a node fails while talking to a client, it gets restarted on the same node if it is healthy or a backup node if it is not. But I am not clear on what happens to resources such as file locks that were originally owned by the node.
2. If a node of a cluster fails, how are internal data of the node recovered, rebuilt, if at all? I mean things like caches or internal data structures.
Please let me know.
Thanks,
Larry Chen

Answer 1

I assume by 3.0 you actually mean 3.x because 3.0 is very old technology now. We're currently on 3.2, which is the 7th or 8th release of the software. Furthermore, I don't think these are particularly simple things to give answers for, so I may have to refer you to other material for longer answers.
You may find most of the answers you want in the book I co-wrote with Richard Elling entitled "Designing Enterprise Solutions with Sun Cluster 3.0". Chapter 3 covers some of the stuff you're asking about.
From your question I'm assuming you're asking about NFS? If so, the NFS protocol together with statd and lockd co-ordinate client recovery after an NFS fail-over (although to the client, it just looks like the same server came back up quickly).
For objects that do have state in the kernel, e.g. writes to PxFS, global devices, etc, these are all handled by a highly available services framework in the kernel. They effectively use a two phase commit-like protocol to ensure that these operations can continue after failures.
Hope that helps (somewhat),
Tim
---