[sheepdog] [PATCH] trivial: misc fixes of comments

[Hitoshi Mitake]

At Wed, 08 May 2013 13:18:59 +0800,
Liu Yuan wrote:
> 
> On 05/08/2013 01:12 PM, Hitoshi Mitake wrote:
> > No. The node simply leaves from the cluster. But it doesn't mean the
> > node always dies. The node keeps connection to a cluster manager can
> > resend join request again.
> > 
> > The leaved sheep marks itself as a node with 0 vnodes in
> > sd_leave_handler() (I'm not sure it is really required). But the node
> > doesn't become a gateway node. Because other nodes increment their
> > epoch and exclude the node from their views.
> 
> This is just how we handle EIO. From the users perspective, the node can
> still work as a gateway to transmit requests and it still has the
> knowledge of topology of cluster.

So do you mean there are two types of a gateway node:
1. a gateway node which is joining a cluster (sheeps executed on
   physical nodes only for executing VMs)
2. a gateway node which is not joining a cluster (sheeps faced EIO)
?

It seems strange. I think sheeps faced EIO and sheeps crashed should
be treated as different things. e.g. changing number of vnodes via
dedicated operations would be a better solution.

Thanks,
Hitoshi