[sheepdog] [PATCH] zk: delete zk node before join cluster

[MORITA Kazutaka]

At Fri, 31 May 2013 10:03:32 +0800,
Kai Zhang wrote:
> 
> When will the 0.6.0 be released?

I'm going to release it today as I announced last week.

> 
> And would you like to share some details of removing master?

Currently, we assume that all the sheeps must have the latest epoch
information when starting sheepdog.  To avoid the older node joining
to the cluster, we elect a master from the nodes who have the latest
epoch, and make the master check whether the joining node has the
latest one, too.

My idea removes this assumption; older nodes can join but are linked
to the delayed list, and all the nodes share the latest epoch
information by using corosync multicast or zookeeper znode.  Because
all sheeps have the latest epoch, any node can check the joining node.
I've implemented almost all of the codes, and group.c and the cluster
drivers are very simplified.

Anyway, I'll send the patch set this weekend or the next week.  Let's
discuss this issue on the patches.


> As I discussed with Yuan, we both believe that current zookeeper driver is not quite stable in a bad network environment.
> Maybe corosync also has this problem, Wenhao has tested this.

I believe corosync is much more stable on smaller environment less
than twenty nodes.

Thanks,

Kazutaka