[dm-devel] Re: [PATCH 5/5] dm-mpath: convert to request-based

[Mike Snitzer <snitzer redhat com>]

On Fri, Aug 28 2009 at  1:00am -0400,
Kiyoshi Ueda <k-ueda ct jp nec com> wrote:

> Hi Alasdair,
> 
> On 08/28/2009 02:54 AM +0900, Alasdair G Kergon wrote:
> > On Tue, Jun 02, 2009 at 04:03:25PM +0900, Kiyoshi Ueda wrote:
> >> This patch converts dm-multipath target to request-based from bio-based.
> >  
> > How much effort would it be to retain the old mpath implementation
> > in parallel?
> > 
> > I'm rather concerned that we're losing some useful functionality in
> > 2.6.31 with this patch - stacking over bio-based devices (test beds use
> > this and it's helpful for debugging), barrier support - and supporting
> > both would make it easier for people to compare the two implementations
> > and stick to the old one if in their particular circumstances it worked
> > better.
> > 
> > Perhaps, dm-mpath could just register two targets (like snapshot does),
> > one bio-based, and one rq-based, sharing most of the functions with
> > wrappers to indicate which is which where necessary?
> 
> Such wrappers need to be made very well to share codes as much as
> possible.  Otherwise, we won't be able to maintain the non-default
> (bio-based?) code, then people won't be able to use it even for
> testing/debugging.
> Also we need to consider the user interface so that user-space tools
> won't be confused.
> 
> I'll look into this when I finish the barrier implementation of
> request-based dm. (hopefully 2.6.32 timeframe, maybe 2.6.33)
> 
> By the way, only for testing/debugging purposes, making request-based
> error/linear targets (e.g. named like error_rq/linear_rq) may be an
> alternative.

As a stop-gap I'd imagine Linux's 'scsi_debug' could be used for testing
and debugging purposes.

Mike