[Pacemaker] resources not migrating when some are not runnable on one node, maybe because of groups or master/slave clones?

[Phil Frost]

On 06/18/2012 10:05 AM, Jake Smith wrote:
> Why don't you have vg_nfsexports in the group? Not really any point to 
> a group with only one resource... 

> You need an order constraint here too... Pacemaker needs to know in 
> what order to start/stop/promote things. Something like: order 
> ord_drbd_master_first drbd_nfsexports_ms:promote vg_nfsexports:start 
> test:start

Indeed, a group with one resource isn't useful, but this is a minimal 
test case, not a practical configuration. My understanding is that 
putting just one resource in a group, though not a useful thing to do, 
should be a no-op with respect to how that resource is placed in the 
cluster. This is because a group is intended to be a shorthand for order 
and colocation constraints among the group, and in a group of 1, there 
is nothing to constrain. In this case, the presence of the group, which 
should do nothing but does, suggests to me a bug in the constraint logic 
or my understanding of group semantics.

Regarding the order constraint, I don't think it is necessary, because 
Dummy resources can start in any order they please. Indeed if they were 
actually the services the names suggested they would need the order 
constraint, and I did have them in my full configuration, but in my 
attempt to reduce the problem I determined that the order constraint had 
no effect on the undesirable behavior, so I removed it.