[sheepdog-users] Cluster snapshot load not working

[Liu Yuan]

On Tue, Dec 17, 2013 at 10:28:16AM +0100, Valerio Pachera wrote:
> 2013/12/17 Liu Yuan <namei.unix at gmail.com>
> 
> > Or you can just use master tip to create some cluster snapshot and do the
> > load
> > and check if it works as expected.
> >
> 
> I did that and it behaves almost the same.
>   qemu-img convert -f qcow2 wheezy_template.qcow2 sheepdog:wheezy_template
>   qemu-img convert -f qcow2 wheezy_template2.qcow2 sheepdog:wheezy_template2
>   dog vdi snapshot wheezy_template
>   dog vdi snapshot wheezy_template2
>   dog cluster snapshot save test cluster_snapshot_test
>   dog cluster snapshot load test cluster_snapshot_test
> 
> 
> Guest do not complete boot.
> I'm able to vdi check the snapshot of the vdi (before it was not possible).
> I get error and alert messages in sheep.log.
> 
> I tried to import the same qcow2 and it works fine
>   qemu-img convert -f qcow2 wheezy_template.qcow2 sheepdog:wheezy_template3
> 

Hi Kazutaka,

   I have traced backed to your patch for this problem:

 - commit 3fbb497c572ef5339d5cbb1c1ff93b52b25e419b
   add sha1_from_buffer()

Could you take a look at this? If no time, I'll revert this patch.
P.S. I have spent some time trying to fix, but no luck. I'd prefer to remove
this tricky function because it doesn't really speedup over plainly sha1 the
whole buffer.

following test script can reproduce the problem easily:

collie/collie cluster format -c 3
qemu-img convert linux-0.2.img sheepdog:test
collie/collie vdi snapshot test -s snap
collie/collie cluster snapshot save snap1 snapshot
md5sum /tmp/store/0/obj/{007c2b2500000000,007c2b2500000001,007c2b2500000002,007c2b2500000003,007c2b2500000004}
collie/collie cluster snapshot load snap1 snapshot
md5sum /tmp/store/0/obj/{007c2b2500000000,007c2b2500000001,007c2b2500000002,007c2b2500000003,007c2b2500000004}
collie/collie vdi read test | md5sum
md5sum linux-0.2.img

Thanks
Yuan