[dm-devel] [PATCH v2] multipathd: fix missing persistent reseravtion for active path

[Martin Wilck]

On Mon, 2021-09-13 at 10:32 -0500, Benjamin Marzinski wrote:
> On Mon, Sep 13, 2021 at 09:01:11AM +0200, Martin Wilck wrote:
> > Hello lixiaokeng,
> > 
> > On Mon, 2021-09-13 at 10:43 +0800, lixiaokeng wrote:
> > > There are two paths(sucu as sda and adb) for one LUN. The two
> > > paths log in, but before the two uevents have been processed
> > > (for example there are many uevent), users use multipathd add
> > > path /dev/sda to cause mpatha and use mpathpersist -o -I to
> > > register prkey for mpatha. The add map uevent is after add path
> > > uevent, the the uevent(add sdb) will delay and missing persistent
> > > reseravtion check.
> > > 
> > > Here, we add persistent reseravtion check in update_map() which
> > > is called ev_add_map().
> > > 
> > > Signed-off-by: Lixiaokeng <lixiaokeng at huawei.com>
> > 
> > Thank you, this looks ok to me. Have you tested it?
> > 
> > I'll wait for Ben's opinion nonetheless, because he's more
> > exprerienced
> > with this part of the code than myself.
> > 
> > This said, I would like to have multipathd record which paths have
> > already registered the key, to avoid doing that repeatedly.
> > 
> Other than adding this, the patch looks fine.

I would say we can take the patch, then. We can add the record-keeping
later, I suppose it needs some deeper considerations. I wouldn't be
against lixiaokeng giving it a shot ;-)

> > Additional question to Ben in this context: what's the reason that
> > we
> > don't actively register keys (that we found in multipath.conf or
> > prkeys) during multipathd startup / reconfigure?
> 
> Like I said in my reply to the first patch, the goal was to make
> persitent reservations to multipath devices work just like with scsi
> devices.

There's no obvious way to do it for SCSI other than writing a custom 
udev rule. For multipath, we could. I see no problem with adding the
automatic registration, making it depend on a new configuration setting
(well _almost_ no problem - yet another configuration option). The
question is whether anyone would be interested in such a feature.

Regards
Martin