Server delete tasks are incorrectly serialized

[Steven Hardy]

Currently, when deleting a number of non-dependent OS::Nova::Server resources, we serialize the deletes, which is happening because we start the task in the handle_delete, instead of handing off to the check_delete_complete, enabling subsequent tasks to be scheduled.

This means there's a noticeable delay when deleting the resources, because the first step of the task is performed by the start(), which nova_utils.delete_server means it does the delete.

Moving the start to check_delete_complete solves the problem - the deletes are all started approximately in parallel then, as expected.

The same bug appears to exist in the EC2::Instance resource too.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/109027

[Zane Bitter (zaneb)]

"we serialize the deletes, which is happening because we start the task in the handle_delete,"

Correction: we serialise the deletes, always (regardless of whether they occur in handle_delete or check_delete_complete), because we are running in a single (green)thread.

[OpenStack Infra (hudson-openstack)]

Change abandoned by Steven Hardy (<email address hidden>) on branch: master
Review: https://review.openstack.org/109027

[Robert Collins (lifeless)]

Why was this abandoned? latest heat still shows the problem.

[Robert Collins (lifeless)]

(and stevens patch *fixed it*)

[Robert Collins (lifeless)]

Did more digging, sorry for the noise, trunk is indeed fixed in this regard already.

[Steven Hardy (shardy)]

Robert - yes, after some confusion, we realized my patch actually didn't fix it, and that the key change was this patch from Randall:

https://review.openstack.org/#/c/99800/