[Pacemaker] One instance of a multistate resource unexpectedly stopped (hbreport attached)

[Florian Haas]

Andrew,

I have an arguably mildly complex configuration like the following:

* 2 DRBD M/S resources.
* 2 resource groups each consisting of a filesystem, and IP address, and
a MySQL instance.
* Order constraints to make sure that the M/S resources are promoted
before the groups are started.
* Colo constraints to make sure that the groups are started on the node
that has the corresponing M/S resource in the master role.

Thus far, everything works as expected.

Now I add another colo constraint with which I want to make sure that
the two M/S resources are never in the master role on the same node:

colocation c_mysql1_not_on_mysql2 \
	-inf: ms_drbd_mysql1:Master ms_drbd_mysql2:Master

With a configuration like that, when I start the cluster afresh, one
instance of one of my DRBD M/S resources remains stopped, with an
allocation score of -INFINITY on both nodes:

# ptest -Ls | grep res_drbd_mysql1:
clone_color: res_drbd_mysql1:0 allocation score on alice.demo: 0
clone_color: res_drbd_mysql1:0 allocation score on bob.demo: 10001
clone_color: res_drbd_mysql1:1 allocation score on alice.demo: 0
clone_color: res_drbd_mysql1:1 allocation score on bob.demo: 0
native_color: res_drbd_mysql1:0 allocation score on alice.demo: -1000000
native_color: res_drbd_mysql1:0 allocation score on bob.demo: 10001
native_color: res_drbd_mysql1:1 allocation score on alice.demo: -1000000
native_color: res_drbd_mysql1:1 allocation score on bob.demo: -1000000
res_drbd_mysql1:0 promotion score on bob.demo: 0
res_drbd_mysql1:1 promotion score on none: 0
res_drbd_mysql1:0 promotion score on bob.demo: 1000000
res_drbd_mysql1:1 promotion score on none: 0

If I remove that negative-score colo constraint, the slave instance
immediately gets started. Add the constraint again, slave instance
stops. Positive-score colo constraints, by contrast, work exactly as
expected -- no slave stoppage there.

Correct me if I'm wrong, but I am seeing no reason why this would be
expected. Can you take a look into the attached hb_report please?

Thanks!
Cheers,
Florian

-------------- next part --------------
A non-text attachment was scrubbed...
Name: hbreport.tar.bz2
Type: application/x-bzip
Size: 76811 bytes
Desc: not available
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20100224/6cc2ad1b/attachment-0001.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 260 bytes
Desc: OpenPGP digital signature
URL: <https://lists.clusterlabs.org/pipermail/pacemaker/attachments/20100224/6cc2ad1b/attachment-0003.sig>