Provided the compute resource is slow for some reason (such as heavy load),
it can happen that the request to the compute resource takes longer than expected.
In that case, orchestration rolls back the local database. However, the
original request finishes successfully. This leads to this scenario:

1. try to delete a host with compute resource
2. compute resource timeouts on deletion of the VM
4. the local database host deletion is rolled back
5. the deletion of the VM succeeds eventually
6. the next attempt to delete the host fails with 'ActiveRecord::RecordNotFound', because
the associated VM is missing

Similar cases apply to other host orchestation (create, update), and are not limited to just
compute resources (same applies to smart proxy calls).