hch's native NVMe multipathing [was: Re: [PATCH 1/2] Don't blacklist nvme]

[snitzer@redhat.com (Mike Snitzer)]

On Wed, Feb 15 2017 at  9:56am -0500,
Christoph Hellwig <hch@infradead.org> wrote:

> On Tue, Feb 14, 2017@04:19:13PM -0500, Keith Busch wrote:
> > These devices are mulitpath capable, and have been able to stack with
> > dm-mpath since kernel 4.2.
> 
> Can we make this conditional on something?  I have native NVMe
> multipathing almost ready for the next merge window which is a lot easier
> to use and faster than dm.  So I don't want us to be locked into this
> mode just before that.

You've avoided discussing this on any level (and I guess you aren't
going to LSF/MM?).  Yet you're expecting to just drop it into the tree
without a care in the world about the implications.

Nobody has interest in Linux multipathing becoming fragmented.

If every transport implemented their own multipathing the end-user would
be amazingly screwed trying to keep track of all the
quirks/configuration/management of each.

Not saying multipath-tools is great, nor that DM multipath is god's
gift.  But substantiating _why_ you need this "native NVMe
multipathing" would go a really long way to justifying your effort.

For starters, how about you show just how much better than DM multipath
this native NVMe multipathing performs?  NOTE: it'd imply you put effort
to making DM multipath work with NVMe.. if you've sat on that code too
that'd be amazingly unfortunate/frustrating.