Re: [PATCH 1/6] libmultipath: signal device with no table in libmp_mapinfo

[Benjamin Marzinski <bmarzins@redhat.com>]

On Thu, Nov 21, 2024 at 09:57:28AM +0100, Martin Wilck wrote:
> On Wed, 2024-11-20 at 16:59 -0500, Benjamin Marzinski wrote:
> > 
> > How would you feel about adding a parameter to do_foreach_partmaps()
> > to
> > say whether or not it should remove empty partmaps (or possibly just
> > checking if the partmap_func is remove_partmaps).  Your argument
> > makes
> > sense when you are removing a device. But what about functions like
> > dm_cancel_remove_partmaps() and dm_rename_partmaps()? I'm not sure
> > that
> > these should automatically empty (a possibly being created) partition
> > devices.
> 
> We're moving far into corner case land here :-) 
> 
> I think we should keep it as simple as possible. For me, that means
> that an empty partition map (with UUID matching the current mpath map)
> should be treated like any other partition map with matching UUID, no
> matter what the current operation is.
> 
> There will be some situations where the outcome will be suboptimal. By
> keeping it simple, we'll at least be able to understand the outcome.

That's reasonable. I misunderstood your suggestion as asking to always
remove empty partition devices in do_foreach_partmaps(). We'd need to
make sure all the partmap_func()s work with empty devices. But that
shouldn't be too bad.

-Ben 
 
> Martin