just to summerise.

placment is a component in OpenStack which is basically a restapi for tracking resources and mapping allocation of resource to inventories.
its implemented in python and mostof the heavyliftign is done in SQL.

with the current version of mariadb in rhel 9.0 one of the primary queries we use now returned incorrect data.

this issue was also reported upstream in OpenStack 
[1] https://lists.openstack.org/pipermail/openstack-discuss/2022-July/029536.html
[2] https://lists.openstack.org/pipermail/openstack-discuss/2022-July/029567.html

the reporter upstream provided a reproduced db dump and test sql query to remove the need for placement.

placement show via a different query that the actual usage is
openstack resource provider usage show 65929119-23f6-4ba2-b98b-4eab5884633f
 
+----------------+-------+
| resource_class | usage |
+----------------+-------+
| VCPU           |    24 |
| MEMORY_MB      | 16384 |
| DISK_GB        |   200 |
+----------------+-------+
but when we execute the allocation candidate query we get

Over capacity for VCPU on resource provider 65929119-23f6-4ba2-b98b-4eab5884633f. Needed: 12, Used: 16608, Capacity: 1024.0

the used value is equal to the total of all used vales Used: 16608 == 24+16384+200



so the inner subquery that is suming the allocation per resource type is not generating the correct info.

SELECT
     rp.id, rp.uuid, rp.generation, inv.resource_class_id, inv.total, inv.reserved, inv.allocation_ratio, allocs.used
FROM
     resource_providers AS rp
     JOIN inventories AS inv ON rp.id = inv.resource_provider_id
     LEFT JOIN (
         SELECT
             resource_provider_id, resource_class_id, SUM(used) AS used
         FROM
             allocations
         WHERE
             resource_class_id IN (0, 1, 2)
             AND resource_provider_id IN (5)
         GROUP BY
             resource_provider_id, resource_class_id
     ) AS allocs ON
         inv.resource_provider_id = allocs.resource_provider_id
         AND inv.resource_class_id = allocs.resource_class_id
WHERE
     rp.id IN (5)
     AND inv.resource_class_id IN (0,1,2)
;

others also hit this https://serverfault.com/questions/1064579/openstack-only-building-one-vm-per-machine-in-cluster-then-runs-out-of-resource
and tracked it down to this bug that affect  10.3, 10.4, 10.5, 10.5.10 https://jira.mariadb.org/browse/MDEV-25714

we are schedule to have a beta of our next major release in ~ 2 weeks with full GA follow not long after.
this likely will block full GA and may block beta which is why i have set the highest severity.

the rhel/centos release version number indicate that it should be fixed but we are still seeing this in our internal build.


https://access.redhat.com/downloads/content/rhel---9.0/x86_64/11161/mariadb/10.5.13-2.el9/x86_64/fd431d51/package
centos 9 steram is using 10.5.16
https://gitlab.com/redhat/centos-stream/rpms/mariadb/-/blob/c9s/sources

i am following up with our release deliver team to make sure we are building with the latest 9.0 rpms currently but in the interim can you confirm if that should be fixed in our current packages
or if a backport/rebase is requried.

we appear to be using  mariadb-10.5.13-2.el9.x86_64,
so that imples that this is infact present in the latest version.

the jira issue suggest it fixed in 10.5.11 but based on the last comment
https://jira.mariadb.org/browse/MDEV-25714?focusedCommentId=219653&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-219653
its actually fixed in 10.5.15

ok we might want to close this as a duplicate of https://bugzilla.redhat.com/show_bug.cgi?id=2096274

Hello Sean,

If this is the duplicate as you mentioned in comment#3 feel free to close this.

We're shipping the mariadb-10.5.16 version to the RHEL-9.0, but it's still in process, so I cannot tell you how long it could take.

thanks, closing as a duplicate of https://bugzilla.redhat.com/show_bug.cgi?id=2096274
that has now been released and I believe we have already pulled mariadb-10.5.16-2.el9_0 into our latest internal build
so we should be unblocked.
http://rhsm-pulp.corp.redhat.com/content/eus/rhel9/9.0/x86_64/appstream/os/Packages/m/mariadb-10.5.16-2.el9_0.x86_64.rpm

*** This bug has been marked as a duplicate of bug 2096274 ***