ImageNotFound is not correctly handled by compute manager

[Paul Murray]

When building an instance, if the glance client throws an ImageNotFound exception the compute manager will reschedule the instance without cleaning up any networks that have been built out, leaking any ports that have been created.

The correct behaviour should include cleaning up allocated networks and abandoning the instance creation, leaving it in error.

The relevant exception handling is in nova/compute/manager.py starting at line 1923. The ImageNotFound exception should be included in the list at line 1960 so the BuildAbortException is thrown, which will be caught and handled correctly further up the stack.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/384649

[Matt Riedemann (mriedem)]

Seems valid, although the API checks that the image is available, so the image would have been deleted between the time the API calls conductor which calls scheduler to pick a host to cast to to build the instance, which seems pretty slim, but we should cleanup yeah.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Paul Murray (<email address hidden>) on branch: master
Review: https://review.openstack.org/384649
Reason: I will not be working on this any more - feel free to pick it up

[Paul Murray (pmurray)]

I will not be working on this anymore, please feel free to pick it up