[sheepdog] [PATCH 1/3] sheep: fix a bug that read_object() fail in recovery

[Christoph Hellwig]

On Sun, Jun 24, 2012 at 09:24:54PM +0800, levin li wrote:
> From: levin li <xingke.lwp at taobao.com>
> 
> read_object() calls forward_read_obj_req() to get the object,
> but may fail with result SD_RES_OLD_NODE_VER, read_object() do
> nothing to handle this error, this patch fixed this problem
> by sending a gateway request to local node, making gateway to
> retry when error occur.

Is there any good reason to use networking to talk to the local node
instead of directly queueing it up on the workqueue?

Also is there any good reason to not treat write and remove requests
the same way?  If there is a good reason to do that it should probably
be added to the code as a comment to explain the asymetric
implementation.