numa_nodes=1 pined to wrong numa node

[Junbo Jiang]

There is a compute-node which have two numa nodes, each numa node have 2 cpus and 8G memory, humactl give

# numactl -H
available: 2 nodes (0-1)
node 0 cpus: 0 1
node 0 size: 7574 MB
node 0 free: 5049 MB
node 1 cpus: 2 3
node 1 size: 7874 MB
node 1 free: 7150 MB
node distances:
node 0 1
  0: 10 20
  1: 20 10

nova.conf
[DEFAULT]
ram_allocation_ratio = 1.0
[compute]
packing_host_numa_cells_allocation_strategy = True

with the above configurations, the nova-compute node try use one numa node until it exhausted, then pick next numa node. but it's not!

here is the steps to reproduce it

openstack flavor create --vcpus 2 --ram 6144 --disk 0 --property hw:numa_nodes=1 2c.6g
nova boot --image centos --flavor 2c.6g --host <this numa node> vm1
nova boot --image centos --flavor 2c.6g --host <this numa node> vm2

the flavor have 6G memory, so for this compute-node, each numa node can only have 1 vm. but the actual result is

# virsh list
 Id Name State
-----------------------------------
 21 instance-00000056 running
 22 instance-00000057 running

# virsh numatune 21
numa_mode : strict
numa_nodeset : 0
# virsh numatune 22
numa_mode : strict
numa_nodeset : 0

both vm pinned to numa node 0! even numa node 0 do not have sufficient memory!

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/c/openstack/nova/+/858495