[sheepdog-users] node recovery question

Liu Yuan namei.unix at gmail.com
Wed Jul 18 08:38:28 CEST 2012


On 07/18/2012 02:24 PM, Dietmar Maurer wrote:
>>> Is there a way to prevent that, for example when I just want to reboot a
>> node? Such node
>>> will be online again shortly so it makes no sense to start copying data.
>>>
>>> - Dietmar
>>>
>>
>> We have to think about IOs that happen between the node restarting, so
>> copying data means to 1) recover the date to the latest state 2)
>> re-balance the data in case you add more disks in with different weight.
> 
> I don't really understand that answer (does that answer my question?).
> 

No way to prevent recovery process for now at node change event.

> And what happen if someone restart a sheep in different zone (-z X).
> 

This means two node change event: node join and node leave, that is,
epoch will plus 2.

Thanks,
Yuan





More information about the sheepdog-users mailing list