[sheepdog] [PATCH v2 2/6] write inode at one time in vdi_create() and vdi_clone()

[MORITA Kazutaka]

At Wed, 16 Oct 2013 15:55:35 +0800,
Liu Yuan wrote:
> 
> On Wed, Oct 16, 2013 at 03:21:47AM +0900, MORITA Kazutaka wrote:
> > At Tue, 15 Oct 2013 14:45:36 +0800,
> > Robin Dong wrote:
> > > 
> > > It will improve performance for writing whole inode once after changing all vids
> > > instead of writing each vid every time.
> > > 
> > > Signed-off-by: Robin Dong <sanbai at taobao.com>
> > > ---
> > >  dog/vdi.c |   50 ++++++++++++++++++++++++++++++++------------------
> > >  1 files changed, 32 insertions(+), 18 deletions(-)
> > > 
> > > diff --git a/dog/vdi.c b/dog/vdi.c
> > > index 28509e2..02267ab 100644
> > > --- a/dog/vdi.c
> > > +++ b/dog/vdi.c
> > > @@ -536,15 +536,18 @@ static int vdi_create(int argc, char **argv)
> > >  		}
> > >  
> > >  		sd_inode_set_vdi(inode, idx, vid);
> > > -		ret = sd_write_object(vid_to_vdi_oid(vid), 0, &vid, sizeof(vid),
> > > -				      SD_INODE_HEADER_SIZE + sizeof(vid) * idx,
> > > -				      0, inode->nr_copies, inode->copy_policy,
> > > -				      false, true);
> > > -		if (ret) {
> > > -			ret = EXIT_FAILURE;
> > > -			goto out;
> > > -		}
> > >  	}
> > > +
> > > +	ret = sd_write_object(vid_to_vdi_oid(vid), 0, inode,
> > > +			      sizeof(uint32_t) * MAX_DATA_OBJS,
> > > +			      SD_INODE_HEADER_SIZE, 0,
> > > +			      inode->nr_copies, inode->copy_policy,
> > > +			      false, true);
> > > +	if (ret) {
> > > +		ret = EXIT_FAILURE;
> > > +		goto out;
> > > +	}
> > > +
> > 
> > This improves performance, but generates lots of orphan data objects
> > if we interrupt vdi creation.  Do you have any performance number?
> > 
> 
> I guess we can firstly update inode and then write data objects. if operation is
> interrupted, we can delete the vdi and no orphan.

Who fixes the corrupted inode object after vdi creation is
interrupted?

Thanks,

Kazutaka