[Sheepdog] sheep recovery

[MORITA Kazutaka]

CCed to the sheepdog mailing list.

On 2011/05/01 3:07, Ski Mountain wrote:
> I have some questions about what happens when a sheep node fails, and the VM 
> (Virtual Machines) are started up on a different sheep node because of the 
> automatic object recovery that is built into sheep dog.  
> 
> 
> Is it based on, disk, memory, idle CPU, response time, number of VM currently on 
> the other sheep nodes?  
> 
> 
> 
> Or would it be possible to add it in a future feature using some or all of the 
> above configuration options.  IE if one of the nodes failes in a cluster then 
> they will automatically be booted back up on the machine that has the most ilde 
> CPU currently.  Or bring the VM from the filed node up on the machine that has 
> the least amount of memory used at that used...
> 
> Would love to start a discussion with other members of sheepdog as to show the 
> best was to accomplish this would be.  

Sheepdog doesn't restart virtual machines which ran on the failed
nodes, and there is no plan to add such feature now.  The management
of virtual machines is out of work of the storage system, isn't it?
I think it should be done in another layer (e.g. OpenStack Nova).

Thanks,

Kazutaka