namespaces 'haproxy' and 'vrouter' are empty after re-deploying cluster

[Dennis Dmitriev]

The base test case described in https://bugs.launchpad.net/fuel/+bug/1425945

Scenario:
1) Add nodes to a cluster (nova-flat): 3 controller, 2 compute, 1 cinder
2) Deploy cluster
3) Remove cinder and one of controller nodes; Add a new node and assign the 'controller+cinder' node to it.
4) Deploy cluster

Expected result: Deploy was successful.
Actual result: Deploy failed by timeout; cinder and one of controller nodes are removed from cluster, but a new node with the role 'controller+cinder' is in 'provisioned' state: http://paste.openstack.org/show/193093/

During waiting for provisioning, the Primary Controller lost all namespaces:
root@node-1:~# # ip netns list
root@node-1:~#

But vip__management was still pointing to the node-1:
root@node-1:~# crm_mon -1 | grep vip__management
 vip__management_vrouter (ocf::fuel:ns_IPaddr2): Started node-1.test.domain.local
 vip__management (ocf::fuel:ns_IPaddr2): Started node-1.test.domain.local

After deploy failed, namespaces appeared:
root@node-1:~# ip netns list
haproxy
vrouter

But there is no IP configured inside the namespaces:
root@node-1:~# # ip netns exec vrouter ifconfig
root@node-1:~# # ip netns exec haproxy ifconfig
root@node-1:~#

For example, the same namespaces on the node-2 are OK:
http://paste.openstack.org/show/193095/

The cluster remains non-functional after the failed action.

[Oleksiy Molchanov (omolchanov)]

The initial problem is not in netns I suppose, primary controller was not triggered for redeploy. So new controller is waiting for it and fails.

http://paste.openstack.org/show/193093/