[sheepdog] [PATCH v4] sheepdog: selectable object size support

[Liu Yuan]

On Fri, Feb 06, 2015 at 04:57:55PM +0900, Teruaki Ishizaki wrote:
> (2015/02/06 11:18), Liu Yuan wrote:
> >On Wed, Feb 04, 2015 at 01:54:19PM +0900, Teruaki Ishizaki wrote:
> >>(2015/02/02 15:52), Liu Yuan wrote:
> >>>On Tue, Jan 27, 2015 at 05:35:27PM +0900, Teruaki Ishizaki wrote:
> >>>>Previously, qemu block driver of sheepdog used hard-coded VDI object size.
> >>>>This patch enables users to handle "block_size_shift" value for
> >>>>calculating VDI object size.
> >>>>
> >>>>When you start qemu, you don't need to specify additional command option.
> >>>>
> >>>>But when you create the VDI which doesn't have default object size
> >>>>with qemu-img command, you specify block_size_shift option.
> >>>>
> >>>>If you want to create a VDI of 8MB(1 << 23) object size,
> >>>>you need to specify following command option.
> >>>>
> >>>>  # qemu-img create -o block_size_shift=23 sheepdog:test1 100M
> >>>
> >>>Is it possible to make this option more user friendly? such as
> >>>
> >>>  $ qemu-img create -o object_size=8M sheepdog:test 1G
> >>
> >>At first, I thought that the object_size was user friendly.
> >>But, Sheepdog has already the value of block_size_shift
> >>in the inode layout that means like object_size.
> >>
> >>'object_size' doesn't always fit right in 'block_size_shift'.
> >>On the other hands, 'block_size_shift' always fit right in
> >>'object_size'.
> >>
> >>I think that existing layout shouldn't be changed easily and
> >>it seems that it is difficult for users to specify
> >>the object_size value that fit right in 'block_size_shift'.
> >
> >I don't think we need to change the layout. block_size_shift is what QEMU talks
> >to sheep, and QEMU options is what users talks to QEMU. We can convert the user
> >friendly object size into block_size_shift internally in the driver before
> >sending it tosheep daemon, no?
> >
> For example, users specify 12MB for object size, block_size_shift
> doesn't fit exactly to an integer.

In this case, we should abort and print the error message and notify the users
the acceptable range like erasure coding option.

> 
> I suppose that normally an administrator do format sheepdog cluster
> with specifying block_size_shift and users usually do qemu-img command
> without a block_size_shift option.

block_size_shift is too developer centered and has a direct relation to the
underlying algorithm. If in the future, e.g, we change the underlying algorithm
about how we represent block size, block_size_shift might not even exsit. So
use object_size would be more generic and won't have this kind of problem.

Secondly, it is not hard to parse the object_size into block_size_shift, so I'd
suggest we'd better try our best to make it user friendly.

Thanks
Yuan