Currently the 'running_deleted_instance_action' flag in nova/compute/manager.py defaults to 'noop'. I argue that it should default to 'log'.
In a medium to large installation, it is not immediately obvious when you have "zombie" instances running on a compute node. These instances are a hinderance to the overall health of the system. Setting the flag to "log" would be beneficial to operations staff as they would be made more aware of the problem. I do not think that setting the flag to "log" would hinder developers of Openstack.
Unfortunately, these flags are not very discoverable and so it is less likely that someone in operations would change the default value.