>>> Lars Marowsky-Bree <[email protected]> schrieb am 06.02.2013 um 13:38 in 
>>> Nachricht
<[email protected]>:
> On 2013-02-06T12:22:37, Ulrich Windl <[email protected]> 
> wrote:
> 
> > > > colocation col_OCFS_cVG inf: _rsc_set_ ( cln_CFS ) cln_cLVM
> > > > order ord_cVG_CFS inf: cln_cLVM ( cln_CFS )
> > > 
> > > Why not just:
> > > 
> > > colocation col_OCFS_cVG inf: cln_CFS cln_cLVM
> > > order ord_cVG_CFS inf: cln_cLVM cln_CFS
> > 
> > The reason was "linear extensibility": If you have more than one CFS on 
> different LVs of the VG.
> 
> I'm not sure I follow what you mean by that?

Just imagine you have more than one LV per VG, each containing a filesystem. 
Then each (LV and filesystem) should be treated as independently as possible 
from other pairs. Without the resource sets, you must either define a strict 
order which in unnecessary, or you must define multiple ordering constraints. 
To me, my solution seemed to be the elegant one.


> 
> > > That ought to work. Probably clones and resource sets have a problem
> > > here?
> > From what I had read, the parenthesis don't make the difference.
> 
> I have colocated clones and I've not seen that message, so my guess was
> that this culd fix it.

It would be helpful if anybody could explain what exactly causes the original 
message
pengine: [9800]: notice: clone_rsc_colocation_rh: Cannot pair prm_A:1 with 
instance of cln_B

What if the cluster would simply pair prm_A:1 with prm_B:1? The constraint as 
complained about was not eneterd by the user; it was created by the cluster, 
and then the cluster complains it cannot handle it.

> 
> > > > DLM (Distributed Lock Manager)
> > > > O2CB (OCFS2), needs DLM
> > > > cLVM needs DLM
> > > > LVM-LV needs cLVM
> > > > OCFS2-filesystem needs both, O2CB and LVM-LV
> > > > 
> > > > The pattern should be flexible enough to allow both, OFCS on top of an 
> LV, 
> > > as well as OCFS directly on a shared disk. And the pattern should only 
> define 
> > > contraints that are necessary, i.e. do not put everything in a group and 
> > > clone that group.
> > > 
> > > The latter is the easiest solution that just works; where's the problem
> > > with that? Too simple? ;-)
> > 
> > Well, it's incomplete, as you can see simply by counting the number of 
> resources involved ;-)
> 
> I still don't follow. You add all the file systems to the cloned group.
> Problem solved?

OK, why not be concrete then: If I have, let's say, 10 filesystems for virtual 
machines that should run independently on a selection of 7 nodes, how would 
your configuration look like? 10 cloned groups? Where do you put DLM, cLVM and 
O2CB?
Obviously it cannot be inside the group. So maybe, please:

Provide a working starting configuration, and then describe how to add on that 
starting configuration.

Regards,
Ulrich


_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to