Comment 2 for bug 1873753

Just extra notes on this one, high nova-scheduler CPU is because of this bug.

Its complicated how it happens. There is a new task for nova-computes to request for nova-scheduler to update its instance cache, defaults to every 120 seconds (only required if you want affinity scheduler decisions to be "more efficient"). Now this message is sent to the nova-scheduler fanout queue, which then causes every worker to refresh its cache list. This quickly becomes quite a high load when you have 40 workers on every controller, and really not that many compute nodes.