Bug 1660869

Summary:	[OSP15][Deployment] Disable NUMA live migration by default
Product:	Red Hat OpenStack	Reporter:	Martin Schuppert <mschuppe>
Component:	openstack-tripleo-heat-templates	Assignee:	Rajesh Tailor <ratailor>
Status:	CLOSED ERRATA	QA Contact:	Archit Modi <amodi>
Severity:	medium	Docs Contact:
Priority:	medium
Version:	15.0 (Stein)	CC:	lyarwood, mburns
Target Milestone:	rc	Keywords:	Triaged
Target Release:	15.0 (Stein)
Hardware:	Unspecified
OS:	Unspecified
Whiteboard:
Fixed In Version:	openstack-tripleo-heat-templates-10.3.1-0.20190225120719.92766d5.el8ost puppet-nova-14.2.1-0.20190225161149.f435ef6.el8ost	Doc Type:	If docs needed, set a value
Doc Text:		Story Points:	---
Clone Of:	1660851	Environment:
Last Closed:	2019-09-21 11:19:41 UTC	Type:	---
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:	1660851
Bug Blocks:	1660870, 1707808, 1775181, 1805828

Description Martin Schuppert 2018-12-19 13:02:16 UTC

+++ This bug was initially created as a clone of Bug #1660851 +++

Description of problem:
As noted in bug 1222414, nova does not currently support live migration of instances with NUMA topologies. The CPU and memory mapping information is not recalculated for the destination host, meaning the instance can fail to boot on a destination host with less cores/hugepages than the source or, worse, actually boot but consume resources dedicated to an existing instance.

The long term fix for this is to properly enable live migration for these instances. However, we're seeing issues with this in older versions (and every future version until that feature lands) and the fix won't be backportable to these. In addition, the previous downstream-only workaround for this, seen in bug 1585068 and related bugs, has proven problematic and is not something we want to continue to invest in. As such, we need an upstream-first approach to resolving this issue that can be backported to older versions. The easiest way to do this is to disable live migration of NUMA-based instances by default, providing a workaround flag for users who insist on being able to live migrate even with the issues seen. That is what we've done, and this bug tracks the progress in supporting that downstream.

Version-Release number of selected component (if applicable):
All supported versions of OSP.

How reproducible:
See bug 1222414.

Steps to Reproduce:
See bug 1222414.

Actual results:
See bug 1222414.

Expected results:
See bug 1222414.

Additional info:

Comment 2 Martin Schuppert 2018-12-19 13:03:55 UTC

This BZ is to track the deployment side (add a parameter to enable workarounds/enable_numa_live_migration per role and enable it on NFV roles, like dpdk/sriov)

Comment 3 Martin Schuppert 2018-12-19 13:21:01 UTC

to clarify, default is false and we only set it to true on dpdk roles

Comment 9 errata-xmlrpc 2019-09-21 11:19:41 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, 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/RHEA-2019:2811