cpu binding conflict when multiple VMs execute unshelve at the same time

[tangxing]

Description
===========
hw:cpu_policy of flavor is dedicated，use this flavor create six vms.Then excute 'nova shelve <server-id' command,after vm's state become to SHELVE_OFFLOAD,execute 'nova unshelve <server-id>' at
the same time.This will lead to server vms bind to same cpu.

Steps to reproduce
==================
1:nova flavor-key 1 set hw:cpu_policy=dedicated
2:[root@nail-5300-1 ~(keystone_admin)]# nova boot --flavor 1 --image 22164f51-c353-4f8a-a073-cbb2930bf25f --nic net-id=90ba0584-b2b7-4f6a-b862-f77864d8626f --max-count 6 test
3:nova shelve test-1
nova shelve test-2
nova shelve test-3
nova shelve test-4
nova shelve test-5
nova shelve test-6
4:
nova unshelve test-1
nova unshelve test-2
nova unshelve test-3
nova unshelve test-4
nova unshelve test-5
nova unshelve test-6

Actual result
=============
[root@E9000slot6 /]# virsh vcpupin 24
VCPU: CPU Affinity
----------------------------------
0: 9
1: 33

[root@E9000slot6 /]# virsh vcpupin 25
VCPU: CPU Affinity
----------------------------------
0: 9
1: 33

Logs & Configs
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager File "/usr/lib/python2.7/site-packages/nova/compute/resource_tracker.py", line 1195, in _update_usage_from_instances
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager require_allocation_refresh=require_allocation_refresh)
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager File "/usr/lib/python2.7/site-packages/nova/compute/resource_tracker.py", line 1125, in _update_usage_from_instance
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager sign=sign)
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager File "/usr/lib/python2.7/site-packages/nova/compute/resource_tracker.py", line 913, in _update_usage
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager cn, usage, free)
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager File "/usr/lib/python2.7/site-packages/nova/virt/hardware.py", line 2071, in get_host_numa_usage_from_instance
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager host_numa_topology, instance_numa_topology, free=free))
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager File "/usr/lib/python2.7/site-packages/nova/virt/hardware.py", line 1923, in numa_usage_from_instances
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager newcell.pin_cpus(pinned_cpus)
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager File "/usr/lib/python2.7/site-packages/nova/objects/numa.py", line 89, in pin_cpus
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager self.pinned_cpus))
2017-12-21 19:12:46.417 23081 ERROR nova.compute.manager CPUPinningInvalid: CPU set to pin [9, 33] must be a subset of free CPU set [32, 2, 3, 5, 6, 7, 8, 10, 11, 34, 35, 26, 27, 29, 30, 31]

[Sylvain Bauza (sylvain-bauza)]

That's a known issue that needs some design discussions in order to fix it correctly. The main upstream bug is https://bugs.launchpad.net/nova/+bug/1417667 for all the move ops.