[dm-devel] [PATCH v2 1/2] libmultipath: don't reject maps with undefined prio

[Martin Wilck]

On Fr, 2018-03-23 at 11:30 +0100, Martin Wilck wrote:
> On Do, 2018-03-22 at 18:52 -0500, Benjamin Marzinski wrote:
> > On Wed, Mar 21, 2018 at 10:34:18AM +0100, Martin Wilck wrote:
> > > libmultipath's prio routines can deal with pp->priority ==
> > > PRIO_UNDEF
> > > just fine. PRIO_UNDEF is just a very low priority. So there's
> > > no reason to reject setting up a multipath map because paths have
> > > undefined priority.
> > > 
> > 
> > The problem with this is that presumably, if there is a prioritizer
> > set,
> > then the paths aren't supposed to be in the same pathgroup.  If
> > this
> > is
> > the case and you end up setting all the paths to the same pathgroup
> > because the prioritizer is wrong, you will likely have a bad
> > experience.
> 
> Sure. This is obviously an unfortunate corner-case. The good news is
> that it happens very rarely. The default setting for prio is "const",
> which can't fail, so (disregarding FAILED paths) this can happen only
> for hardware that claims to support ALUA (or some other algorithm)
> but
> actually doesn't (or fails ALUA repeatedly). There's one concrete
> example of such hardware (certain variants of IBM IPR) that has
> motivated this patch.

It turns out that this patch didn't fix the problem with that
particular controller, because hwhandler "alua" would fail during
domap() and thus kernel rejected multipath creation even though
multipathd didn't.

So, while I still think the arguments below are correct in principle,
the reason why I wanted this patch is gone.

Martin

-- 
Dr. Martin Wilck <mwilck at suse.com>, Tel. +49 (0)911 74053 2107
SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)