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

[Liu Yuan]

On Mon, Sep 22, 2014 at 03:27:58PM +0800, Ruoyu wrote:
> Thanks for your discovery.
> I think the bug is caused by the logic of recovery. We have to
> recover an object from stale directory even if it cannot be found in
> other nodes.
> It is correct because if copies == 1, we really need to find it in
> the re-joining node.
> Therefore, if a VDI is deleted after a node is left, the node will
> not know what happened to the VDI so that it will try to recover all
> objects it hold.
> To resolve it, maybe we need a mechanism of informing deleted VDI
> when sheep is starting. Any suggestion?

We don't need it. There are only two cases:

1 if we want to recover the objects from the joining node, we could simply add
  it back and wait for recovery.

2 if we don't want to recovery the objects from the joining node, we could purge
  the joining node first and then add it back.

Thanks
Yuan