[Linux-HA] Cloned resource does not fail back

[Christian Rishøj]

With heartbeat 2.1.2, I have a virtual IP resource, with 2 clones
(using CLUSTERIP). I wish the clones to run on separate nodes when
possible, permitting them to run at the same node if necessary.

The fail over part works fine (I have clone_node_max=2).

When the original node comes back up, my clone don't fail back.

I have:
   symmetric_cluster=true
   default_resource_stickiness=0
   default_resource_failure_stickiness=0

There is no fail count for the clones.

Here's the score calculation (from ptest -LVVVVVVVVVVVVV):

debug2: clone_color: Processing clone_web_ip
debug4: sort_clone_instance: resource_web_ip:0 ? resource_web_ip:1
debug3: sort_clone_instance: resource_web_ip:0 == resource_web_ip:1: default
debug3: sort_node_weight: ska (0) == dub (0) : weight
debug3: sort_node_weight: ska = dub
debug2: color_instance: Processing resource_web_ip:0
debug2: native_print: Allocating: resource_web_ip:0
(cscd::ocf:IPaddr2):    Started ska
debug3: native_color: resource_web_ip:0: Pre-Processing colocation_web_ip
debug2: native_rsc_colocation_lh: Processing colocation constraint
between resource_web_ip:0 and clone_apache2
debug3: clone_rsc_colocation_rh: Processing constraint
colocation_web_ip: 1000000
debug5: node_copy: Copying 0x713ab8 (ska) to 0x75f0b8
debug4: node_list_and: ska: 1 + 1
debug5: merge_weights: 1 + 1 = 2
debug3: node_list_and: New node weight for ska: 2
debug5: node_copy: Copying 0x714d18 (dub) to 0x75edf8
debug4: node_list_and: dub: 2 + 0
debug5: merge_weights: 0 + 2 = 2
debug3: node_list_and: New node weight for dub: 2
native_print: Allocating: resource_web_ip:0    (cscd::ocf:IPaddr2):
Started ska
debug3: native_choose_node: Choosing node for resource_web_ip:0 from 2
candidates
debug3: sort_node_weight: ska (2) == dub (2) : weight
debug3: sort_node_weight: ska = dub
native_assign_node: Color resource_web_ip:0, Node[0] ska: 2
native_assign_node: Color resource_web_ip:0, Node[1] dub: 2
native_assign_node: 2 nodes with equal score (2) for running the
listed resources (chose ska):
native_assign_node: Assigning ska to resource_web_ip:0
debug5: node_copy: Copying 0x75f0b8 (ska) to 0x714d18
debug3: native_color: Allocated resource resource_web_ip:0 to ska
debug3: native_print: Allocated resource_web_ip:0
(cscd::ocf:IPaddr2):    Started ska
debug2: color_instance: Processing resource_web_ip:1
debug2: native_print: Allocating: resource_web_ip:1
(cscd::ocf:IPaddr2):    Started ska
native_print: Allocating: resource_web_ip:1    (cscd::ocf:IPaddr2):
Started ska
debug3: native_choose_node: Choosing node for resource_web_ip:1 from 2
candidates
debug3: sort_node_weight: ska (1) > dub (0) : weight
native_assign_node: Color resource_web_ip:1, Node[0] ska: 1
native_assign_node: Color resource_web_ip:1, Node[1] dub: 0
native_assign_node: Assigning ska to resource_web_ip:1
debug5: node_copy: Copying 0x7114a8 (ska) to 0x713ab8
debug3: native_color: Allocated resource resource_web_ip:1 to ska
debug3: native_print: Allocated resource_web_ip:1
(cscd::ocf:IPaddr2):    Started ska
clone_color: Allocated 2 clone_web_ip instances of a possible 2

In other words, for the first clone, the nodes score as follows:

   ska: 2
   dub: 2

...and node ska is chosen. For the second clone, the scores are:

   ska: 1
   dub: 0

...and of course node ska is chosen again.

For my case, this is not a desirable behavior. So, how can I have the
clones distributed as evenly as possible?

Best regards,
Christian