Bug 2083371

Summary:	Query returns wrong result when using split optimization
Product:	Red Hat Enterprise Linux 9	Reporter:	Marian Krcmarik <mkrcmari>
Component:	mariadb	Assignee:	Michal Schorm <mschorm>
Status:	CLOSED CURRENTRELEASE	QA Contact:	Jakub Heger <jheger>
Severity:	high	Docs Contact:
Priority:	unspecified
Version:	unspecified	CC:	databases-maint, dciabrin, ljavorsk, oblaut
Target Milestone:	rc	Keywords:	AutomationBlocker, Rebase, Triaged, ZStream
Target Release:	---
Hardware:	Unspecified
OS:	Unspecified
Whiteboard:
Fixed In Version:		Doc Type:	If docs needed, set a value
Doc Text:		Story Points:	---
Clone Of:
Clones:	2096274 (view as bug list)		Environment:
Last Closed:	2022-10-19 13:40:10 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:	2092370
Bug Blocks:	2096274

Description Marian Krcmarik 2022-05-09 19:44:32 UTC

Description of problem:
There is a regression in MariaDB (which occurs in 10.5.13 build) which has a negative impact on Openstack, the related upstream bugs are:
https://jira.mariadb.org/browse/MDEV-27132
https://jira.mariadb.org/browse/MDEV-27510

The regression is fixed in 10.5.14.

The way how Openstack is impacted:
The nova placement service fails to schedule a Virtual machine because It believes all the compute nodes ran out of available resources (VCPUs to be specific). The record in the DB is wrong and does not match the reality,. Based on this discussion:
https://serverfault.com/questions/1064579/openstack-only-building-one-vm-per-machine-in-cluster-then-runs-out-of-resource It seems to be caused by the MariaDB regression which start with following bug:
https://jira.mariadb.org/browse/MDEV-25714

We retested Openstack with a scratch build og 10.5.1 + mentioned patches and The problem was not reproduced anymore.

Version-Release number of selected component (if applicable):
mariadb-10.5.13-2.el9.x86_64
OSP17.0

How reproducible:
Always

Steps to Reproduce:
1. The reproducer is within the Openstack environment where higher number of running tests will caused nova placement to stop scheduling new VMs on compute nodes due to wrong record in the DB of available capabilities

Comment 2 Michal Schorm 2022-06-08 10:57:57 UTC

Merged into the CentOS Stream 9:
  https://gitlab.com/redhat/centos-stream/rpms/mariadb/-/commits/c9s