[Sheepdog] node hot-add and auto-failover

Liu Yuan namei.unix at gmail.com
Fri Nov 18 03:34:48 CET 2011


On 11/17/2011 10:24 PM, kalos wrote:

> Hi,
> 
> first of all, congratulations for your work and thank you for sharing
> such an interesting project.
> 
> After testing it last night, a couple of questions arose:
> 
> 1 - I tried to hot-add a node to my cluster and noticed that blocks
> wouldn't be replicated on this new node: in order to make them replicate
> I had to manually restart the sheep daemon on this new added node.
> In fact not only the blocks did not replicate themselves on the new
> node, but the node itself was not seen by other nodes of the cluster.
> The command "collie node list" clearly showed this.
> 
> I followed the exact same steps found in the wiki, using myself a 64bit
> debian squeeze platform.

Hi
 I can't reproduce the issue. would you please show me how to reproduce
it reliably?

By the way, make sure you set up Corosync correctly by setting up
/etc/corosync/corosync.conf correctly.

For most cases, you only need to set 'bindnetaddr' according your network.

> 2 - Is auto-failover present in sheepdog at any rate?
> 

I am not sure if we can use VM management tools such as Openstack to
take care of auto-failover.

Anyone is welcome to hack this in sheepdog.

Thanks,
Yuan



More information about the sheepdog mailing list