Definitely, we can do a better job in determining correctness of the given
cluster stack configuration (in the form of cluster.conf). What currently
validates successfully is a very rough overapproximation of actually
correct and sound configuration. Granted, the ideas behind original
Relax NG schema modularization ([bug 680930], virtually first milestone
on this road) are to preserved.
This bug is to both (1) track the work in this direction in the cman's
package scope and (2) to be blocked by subordinate bugs for other packages
involved (fence-agents, resource-agents, etc.). This bug should serve as
a central point for any discussion on the (generalized) topic.
For (1), current progress is captured in the cluster-devel ML  (with
many back-and-forth iterations due to the complex nature of the changes)
and in branches following STABLE32  and (dependent) RHEL65  ones.
> ([bug 680930], virtually first milestone on this road)
because of the vital decentralization of the "correctness" knowledge,
shifting it towards particular components
Another thing that should have been captured in the schema:
Another discovery (thanks to Piyush Sharma):
/cluster/rm/resources/action -> wrong
/cluster/rm/resources/SOMETHING/action -> ok (for SOMETHING in clusterfs, ip, ...)
As a side effect, wrongly-alleged-valid first case is what makes luci
choke: [bug 1173942].
re [comment 7]:
...as per rgmanager's code examination
Definitely 6.8 material.
[comment 10] applies just to [comment 7] that should rather be tracked
separately: [bug 1272587].
Not worth to pursuit at this point in the release cycle.
Unlike related [bug 1173942] that I am going to reopen.