[sheepdog-users] Re-joing the cluster doesn't remove orphan objects

[Liu Yuan]

On Tue, Sep 23, 2014 at 09:08:02AM +0200, Valerio Pachera wrote:
> 2014-09-23 8:43 GMT+02:00 Liu Yuan <namei.unix at gmail.com>:
> 
> >
> > Yes, this is what I adviced. Purge the node mannually if you don't want to
> > recover the deleted objects before you re-join the node. This doesn't make
> > you
> > happy enough?
> >
> 
> For my self, no problem, but I'm considering 2 things?
> 1) A new user expects he can simply add the node back without side effects.
> If we may get orphan objects rejoining the cluster, I'd better add this to
> the documentation.

Yes, we should document that

- if we repair the node in a short time window, we can simply rejoin & no purge

  The benefit for non-purge is, we won't recover the data if that node has it
  already. But for a long time window, most of the data will be stale. (assuming
  during the repair no deletion of vdi)

- if we halt the node for a long time, it's better to purge then rejoin

> 2) I can't see a case were is preferable not to purge metadata.
 
As above mentioned, for a quick repair.

Thanks
Yuan