On Tue, 2024-03-05 at 10:10 +0100, Peter Rajnoha wrote:
> On 3/5/24 09:47, Martin Wilck wrote:
> > On Tue, 2024-03-05 at 09:19 +0100, Peter Rajnoha wrote:
> > > Within DM and DM-subsystem rules, it's OK to use DM_SUSPENDED, if
> > > needed.
> >
> > I gather that you agree that 11-dm-mpath.rules represents a "DM
> > subsystem" rule set?
> >
>
> Sure, of course.
>
> > >
> > > We should just hide it from all the "other" rules so they don't
> > > need
> > > to
> > > bother. For them (right now), it's either "usable" or "unusable"
> > > device
> > > for whatever reason behind and we (DM+DM-subystem) should
> > > reimport
> > > whatever is needed for the state/set of variables that others may
> > > use,
> > > to stay sane. Of course, we can do this only for the state that
> > > we
> > > own.
> > >
> > > As we discussed before, this can be extended to making a
> > > difference
> > > among "usable", "temporarily unusable" (so reimport the
> > > state/variables
> > > needed) and "completely unusable" state for others.
> >
> > Yeah, but that's future work, and I doubt that it makes sense to
> > invest
> > much effort into it. I definitely wouldn't want to tie this to the
> > current patch set.
> >
>
> I agree.
>
> > As mentioned previously, it might make sense to introduce a flag
> > that
> > expresses something like "you can access this device, but you don't
> > need to" (DISK_RO={0,1} case, for example). But then, we already
> > have
> > DM_ACTIVATION to express the opposite ("you must have a look at
> > this
> > device, its properties have changed"). I wonder if you consider
> > DM_ACTIVATION a dm-internal property?
>
> Well, we added DM_ACTIVATION as a helper primarily for DM and DM-
> subsys
> rules to have a way to identify when the actual (re)activation
> happens,
> or the "add" trigger on coldplug.
>
> I think it was 69-dm-lvm.rules (or 69-dm-lvm-metad.rules at that
> time)
> where we needed to run pvscan only right after the DM dev is
> activated
> and hence avoiding running costly pvscan uselessly where it doesn't
> matter.
>
> If there's anyone else out there with similar use case, I think that
> checking DM_ACTIVATION might be useful. But it's true it is not
> advertised and shouted out somehow publicly yet.
>
> Usually, all the other rules are interested in rescanning all the
> other
> "foreign" state and attributes that is out of DM's hands, which means
> they know exactly when to do the scan or not or any other actions, it
> depends on what attributes are they watching for.
>
> DM_ACTIVATION is very useful to know when stacking devices on top of
> DM
> though, so a time when to activate the layer above. So yes, this
> variable might be useful for other to look for too.
Thanks. I agree, but it's future work and it would mostly be a thing
for authors of stacked subsystems to look into.
I think are on the same page except perhaps for the DISK_RO part of the
set. I will give that some more thought, and submit a non-RFC patch
set.
Regards
Martin
