bfv instance create failed due to insufficient resources, the volume is always attaching

[lijie.xie]

When using a volume to create a instance, the instance failed and in ERROR state due to insufficient resources, and the volume is always in attaching state. I must reset volume to available status through cinder reset-state and continue use it.

I am in pike version, i see https://review.opendev.org/#/c/335358/, nova-api reserve volume before attach volume to a instance. If scheduler failed for No ValidHost, the volume is always not available unless reset state. When delete the ERROR instance，the volume is still in attaching status.

Now, we should restore the state of the volume instead of resetting it every time due to scheduling failure.

Reproduce the problem through the following steps

1)cinder create 1 --image-id 7b379b4c-ad6e-4ade-81f2-94a26e3afa3a --name fake_test

()[root@busybox-openstack-556bf6df94-kkcr2 /]# cinder list
+--------------------------------------+-----------+-----------+------+-------------+----------+-------------+
| ID | Status | Name | Size | Volume Type | Bootable | Attached to |
+--------------------------------------+-----------+-----------+------+-------------+----------+-------------+
| 8ca2c77b-120a-41c1-b9f6-e8ac0f23a0f2 | available | fake_test | 1 | - | true | |
+--------------------------------------+-----------+-----------+------+-------------+----------+-------------+
2)nova boot --boot-volume 8ca2c77b-120a-41c1-b9f6-e8ac0f23a0f2 --flavor 200 --nic net-name=share_net faker_vm

()[root@busybox-openstack-556bf6df94-kkcr2 /]# nova list --all-t --fields name,host,status,task_state
+--------------------------------------+----------+------+--------+------------+
| ID | Name | Host | Status | Task State |
+--------------------------------------+----------+------+--------+------------+
| 256754e0-ed48-4acb-8293-4ac6cdae7384 | faker_vm | None | ERROR | None |
+--------------------------------------+----------+------+--------+------------+
()[root@busybox-openstack-556bf6df94-kkcr2 /]# cinder list
+--------------------------------------+-----------+----------+------+-------------+----------+-------------+
| ID | Status | Name | Size | Volume Type | Bootable | Attached to |
+--------------------------------------+-----------+----------+------+-------------+----------+-------------+
| 84534871-e327-46d4-af3f-ac347da69213 | attaching | fake_sys | 1 | - | true | |
+--------------------------------------+-----------+----------+------+-------------+----------+-------------+

3)nova delete 256754e0-ed48-4acb-8293-4ac6cdae7384

()[root@busybox-openstack-556bf6df94-kkcr2 /]# nova list --all-t
+----+------+-----------+--------+------------+-------------+----------+
| ID | Name | Tenant ID | Status | Task State | Power State | Networks |
+----+------+-----------+--------+------------+-------------+----------+
+----+------+-----------+--------+------------+-------------+----------+
()[root@busybox-openstack-556bf6df94-kkcr2 /]# cinder list --all
+--------------------------------------+----------------------------------+-----------+----------+------+-------------+----------+-------------+
| ID | Tenant ID | Status | Name | Size | Volume Type | Bootable | Attached to |
+--------------------------------------+----------------------------------+-----------+----------+------+-------------+----------+-------------+
| 84534871-e327-46d4-af3f-ac347da69213 | c7bc6a73869541c7a65b49b06fa9e21a | attaching | fake_sys | 1 | - | true | |
+--------------------------------------+----------------------------------+-----------+----------+------+-------------+----------+-------------+

[Lee Yarwood (lyarwood)]

Which version of Pike are you using? The following change was backported to stable/pike sometime ago now and should ensure that any attached/attaching volumes are removed when deleting an ERROR'd out instance:

https://review.opendev.org/q/I4dc6c8bd3bb6c135f8a698af41f5d0e026c39117

[lijie.xie (rune32bit)]

https://review.opendev.org/c/openstack/nova/+/335358

I use Newton version, but backport some features according to our needs, such as above patch, so encounter the problem.

https://review.opendev.org/q/I4dc6c8bd3bb6c135f8a698af41f5d0e026c39117

I check the patch, it cleans up the volumes attached when deleting instance. If i don’t delete this instance, then these volumes will never be used.

Why not cleanup volumes when scheduling fails，this will not affect the second use of the volume to boot instance. If follow the "Clean up network resources when reschedule fails" approach, i think it will be better than current situation.

https://review.opendev.org/c/openstack/nova/+/243477

What do you think.