Re: superseded: [PATCH storage,cluster,manager 0/13] multipath: cluster-wide config, storage and health overview

[Thomas Lamprecht <t.lamprecht@proxmox.com>]

Am 03.07.26 um 18:13 schrieb Dietmar Maurer:
> I just want to mention that NVMeOF has internal multipathing, so adding
> a new storage plugin for multipathing may adds some confusion (iSCSI
> need it, NVMe don't).

Yeah, I agree. That is a big part of why the type is marked as provisional.
I actually started to plan out the layered variant already before
submitting the v1: basically the transport plugins get a multipath option
(iSCSI  first) and the path handling then lives behind the transport,
dm-multipath for SCSI and native ANA for (future) NVMe-oF, so users never
have to pick a separate multipath storage at all.
But as that is quite a bit more work I basically just sent out v1 with
the simple standalone type to get early feedback on the whole direction,
and v2 to address the most grave feedback v1 got.
The cluster config and the health matrix could stay as they currently
are either way, they do not really care about the transport.
A small standalone provider might only be needed for FC, as IIRC there
the LUNs just show up on their own without any transport plugin involved.

> And I guess bad things happens if you add such nvme drive to the
> multipath storage plugin?

Nothing bad will really happen as far as I can tell, it would be just
confusing:

With native NVMe multipathing (the kernel default, ours has it built-in)
multipathd never even sees the single paths, so no dm map can get
assembled for such a WWID. One then "just" gets a 'missing' row in the
health matrix view I add here, and the storage never shows a volume for
it as it only lists actually assembled maps.
>From some basic research it seems like dm-multipath for NVMe would only
do something when explicitly booting with the nvme_core.multipath=0 on
purpose, and not sure if that's worth catering for.

I'll see if I can pick up the other approach on the weekend, but I
still think it makes sense to rebase anything from my series here,
if we actually want something from it, then on top of your work then
in any case.