Note: This bug is displayed in read-only format because the product is no longer active in Red Hat Bugzilla.

Bug 2269653

Summary:	Backport request of "Fix to implement 'pack' or 'spread' VM's NUMA cells" into OSP17.1
Product:	Red Hat OpenStack	Reporter:	Keigo Noha <knoha>
Component:	openstack-nova	Assignee:	Artom Lifshitz <alifshit>
Status:	CLOSED ERRATA	QA Contact:	James Parker <jparker>
Severity:	high	Docs Contact:
Priority:	high
Version:	17.1 (Wallaby)	CC:	alifshit, bgibizer, dasmith, egallen, eglynn, igallagh, jelynch, jhakimra, joflynn, jparker, kchamart, ltoscano, mariel, pgrist, sbauza, sgordon, smooney, vromanso
Target Milestone:	z4	Keywords:	FeatureBackport, Patch, Triaged
Target Release:	17.1
Hardware:	x86_64
OS:	Linux
Whiteboard:
Fixed In Version:	openstack-nova-23.2.3-17.1.20240614020750.2ace99d.el9ost	Doc Type:	Enhancement
Doc Text:	With this update, you can use the parameter `packing_host_numa_cells_allocation_strategy` to configure the placement strategy that the Compute service (nova) uses when scheduling instances on host nodes. You can either spread instances across NUMA nodes or pack instances on the same host NUMA node until the resources of the node are exhausted. This enhancement also improves the scheduling of instances on NUMA nodes that do not have a PCI device if the instance does not request a PCI device, which makes the PCI devices more likely to remain available for instances that request them with a NUMA affinity.	Story Points:	---
Clone Of:		Environment:
Last Closed:	2024-11-21 09:39:48 UTC	Type:	Bug
Regression:	---	Mount Type:	---
Documentation:	---	CRM:
Verified Versions:		Category:	---
oVirt Team:	---	RHEL 7.3 requirements from Atomic Host:
Cloudforms Team:	---	Target Upstream Version:
Embargoed:
Bug Depends On:
Bug Blocks:	2222869

Description Keigo Noha 2024-03-15 06:12:23 UTC

Description of problem:
OSP17.1 nova tends to fill NUMA0 cpus first. This behavior sometimes causes drawbacks in high performance VM requirement.

For example, NUMA0 has SRIOV, NUMA1 has OVS-DPDK.

In this case, the current behavior will cause the failure of creating SRIOV instance when NUMA0 is filled.

Upstream has a LP, https://bugs.launchpad.net/nova/+bug/1940668.
It seemed to try a backport to stable/wallay.

Is it possible to backport the fix into RHOSP17.1?

Version-Release number of selected component (if applicable):
OSP17.1

How reproducible:
Alwaays.

Steps to Reproduce:
1. Deploy overcloud node with SRIOV feature on NUMA0
2. Launch VMs to fill out NUMA0.
3. Launch a VM to use SRIOV.

Actual results:
NUMA0 is used first.

Expected results:
The CPU allocation will be spread out between NUMAs.

Additional info:

Comment 2 Artom Lifshitz 2024-03-22 01:07:33 UTC

We'll attempt this for 17.1.4 using the existing, but abandoned, upstream wallaby backport.

Comment 3 Artom Lifshitz 2024-05-17 16:27:01 UTC

*** Bug 2270005 has been marked as a duplicate of this bug. ***

Comment 21 errata-xmlrpc 2024-11-21 09:39:48 UTC

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory (RHOSP 17.1.4 bug fix and enhancement advisory), and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHBA-2024:9974