[sheepdog-users] is --nohalt dangerous?

[Liu Yuan]

On 07/18/2012 03:10 PM, Dietmar Maurer wrote:
> Please re-read the code. It does not 'relax anything'.

Ah, yeah.

But which is the best value for cluster to switch to halt state?

current_vnode_info->nr_zones <= (sys->nr_copies/2)

means if we have 2 copies, and one node fails will switch the cluster to
halt, probably some people don't want to see it. Some people might be
pleased to run the cluster, even there is only one node left.

So I think nohalt option can serve those people in all.

Thanks,
Yuan