Comment 10 for bug 1341420

I believe its an inherent part of the Nova design that compute nodes have to be able to reject and send requests back to the Scheduler (back to conductor in the new way of doing things) if they get a request for resources that they no longer have.

For sure you don't want a lot of concurrency in the scheduler (try rpc_thread_pool_size=2 , but I think compute managers in an Ironic setup will need to be able to follow the same basic approach.