Hi , I have also faced such issue long time back. Usually in openstack when you delete an instance the corresponding entry in DB is marked as deleted and nova-compute trigger the destroy instance from Hypervisor, but in your case compute service stops working in between. If you want to destroy those instance from that compute node you can try : virsh destroy <VM>

For reference this method did not work in our case /var/log/nova/compute.log showed:

2014-08-20 10:50:09.606 32739 WARNING nova.compute.manager [-] [instance: 9be70f4c-a697-49ea-b426-673d87aa1bd3] Periodic cleanup failed to delete instance: Info cache for instance 9be70f4c-a697-49ea-b426-673d87aa1bd3 could not be found.

cleanup_running_deleted_instances in nova.conf works for me but there is timeout option to i believe default is 300 so around 5 minute.

Hi, I think you have found the answer for your question (so this question can be closed now)!

Following is a detailed explanation for better understanding: Nova Compute has a periodic task '_cleanup_running_deleted_instances' which monitors any such instances which are marked deleted in (nova) DB but are running on the hypervisor. But AFAIR the default action of this periodic task is 'log' which only emits a log message.

Solution: For such instances to be cleaned up/destroyed, change the action for this periodic task to 'reap' by setting this value for 'running_deleted_instance_action' config parameter in nova.conf.