Scheduling a vm to a reporting compute node cause the number of vms in the compute node exceed the limit

[tianxiang ai]

Description
===========
When scheduling a vm to a reporting compute node,
update_available_resource will clear self.stats and update compute_nodes
table in db firstly, and update this table after calculating all
instances on this node. At this moment, the scheduler filter will pass
this node because there is no num_instances in compute_nodes.stats.

Steps to reproduce
==================
1,create a new az,there is noly one compute node in this az;
2,set max_instances_per_host = 100 on the compute node;
3,boot 100 vms to this az, the vms will be scheduled to the specified node;
4,boot some vms to this az once again.

Expected result
===============
boot vm fails and return no valid host error

Actual result
=============
some vms failed but some vms will be scheduled to the compute node successfully when the compute node is reporting available resources.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/700529

[sean mooney (sean-k-mooney)]

Triaging this as medium as this is a silent failure that would allow limts on vms per host to be exceeded that would not be obvious unless the operator specifically looked for it. that said this could also be considered low for many deployments as the NumInstanceFilter is not enabled by default
https://github.com/openstack/nova/blob/65aae518f81c2754e30429a3547c5cf62a3cf598/nova/conf/scheduler.py#L320-L334

[OpenStack Infra (hudson-openstack)]

Change abandoned by tianxiang ai (<email address hidden>) on branch: master
Review: https://review.opendev.org/700529
Reason: i just found that this bug:#1857448 had been fixed in another patch:https://review.opendev.org/#/c/612293/

[tianxiang ai (aitianxiang)]

Abandoned

i just found that this bug:#1857448 had been fixed in another patch:https://review.opendev.org/#/c/612293/