Instances are unexpectedly being scheduled to hosts which reside outside of the matching host aggregate as per the metadata of the flavor with which the instance was created.
- Platform9 Managed OpenStack - v3.6.0-3.11.0
When adding or removing multiple hosts to an aggregate simultaneously, there was a race condition when reporting host aggregate data to nova-scheduler since an incremental update was being sent instead of querying the latest state in the Nova database.
The issue was addressed by modifying the behavior of the scheduler code to refresh the host aggregate data directly from the database. A fix for this bug (IAAS-9869) is included in Platform9 v3.11.1.