Re: [dm-devel] 2.6.10-udm1

[christophe varoqui <christophe varoqui free fr>]

On Wed, 2005-01-12 at 21:13 +0000, Alasdair G Kergon wrote:
> On Wed, Jan 12, 2005 at 10:06:41PM +0100, christophe varoqui wrote:
> > Do still target 2.6.11 for inclusion ?
>  
> I've not been getting as much feedback as I'd hoped for, so I'm not 
> sure how much of the design is stable enough to submit yet.
> 
I guess you won't get the feedback (and flames) before submission.

> For example, one outstanding question is:
> 
>   If there is a local path and a remote iscsi path to the
>   same device, would those paths need different hw_handlers?
> 
What people may want is raid1 across local and iscsi/nbd devices.

What you describe seems to be a (iSCSI) developper test case : The
userspace should avoid feeding a multipath map to the DM in this case.

> In other words, does the hw_handler concept need splitting up?
> 
I don't think so : a LU is either 
o presented by a single storage controler model : a single hw_handler
suffice
o directly attached : no hw_handler needed

I can't image a valid useful scenario where a LU can be presented
through different controler models or through different kernel
subsystems.

regards,
-- 
christophe varoqui <christophe varoqui free fr>