The volumes will be deleted when creating a virtual machine fails with the parameter delete_on_termination being set true, which causes that the rescheduling fails

[zhangtralon]

when specifying a volume or an image with a user volume to create a virtual machine, if the virtual machine fails to be created for the first time with the parameter delete_on_termination being set “true”, the specified volume or the user volume will be deleted, which causes that the rescheduling fails.
for example:
1. upload a image
| 62aa6627-0a07-4ab4-a99f-2d99110db03e | cirros-0.3.2-x86_64-uec | ACTIVE
2.create a boot volume by the above image
cinder create --image-id 62aa6627-0a07-4ab4-a99f-2d99110db03e --availability-zone nova 1
| b821313a-9edb-474f-abb0-585a211589a6 | available | None | 1 | None | true | |
3. create a virtual machine
nova boot --flavor m1.tiny --nic net-id=28216e1d-f1c2-463b-8ae2-330a87e800d2 tralon_disk1 --block-device-mapping vda=b821313a-9edb-474f-abb0-585a211589a6::1:1
ERROR (BadRequest): Block Device Mapping is Invalid: failed to get volume b821313a-9edb-474f-abb0-585a211589a6. (HTTP 400) (Request-ID: req-486f7ab5-dc08-404e-8d4c-ac570d4f4aa1)
4. use the "cinder list" to find that the volume b821313a-9edb-474f-abb0-585a211589a6 has been deleted
+----+--------+------+------+-------------+----------+-------------+
| ID | Status | Name | Size | Volume Type | Bootable | Attached to |
+----+--------+------+------+-------------+----------+-------------+
+----+--------+------+------+-------------+----------+-------------+

[Andrew Laski (alaski)]

https://review.openstack.org/#/c/106477/

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/106477
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=6ddd9f93f82427ce909c7773f7a806361035a0b2
Submitter: Jenkins
Branch: master

commit 6ddd9f93f82427ce909c7773f7a806361035a0b2
Author: Andrew Laski <email address hidden>
Date: Fri Jul 11 14:56:22 2014 -0400

    Don't remove delete_on_terminate volumes on a reschedule

    When cleaning up volumes before a reschedule if delete_on_terminate is
    True the volume would be deleted. That's not the desired behavior so
    the volume cleanup has been moved to take place when a build is aborted.

    Change-Id: I142370c0555495b4d51736f4f6b8070a3c112a59
    Closes-bug: #1336127