Comment 29 for bug 1341420

I ran the attached script on a devstack instance setup with 3 Ironic nodes. I used the fake Ironic driver because it should not affect Nova scheduling behaviour. First, I ran the script on unmodified master. Second, I modified the Exact* filters to set the resource limits on the HostState objects.

Unmodified:
Total runs 100 scheds 300 rescheds 203 errors 544

With limits applied to Exact* filters:
Total runs 100 scheds 300 rescheds 223 errors 0

Interestingly, although this change removed all errors, the number of reschedules was slightly increased. This is not hugely surprising, as the race condition will be caught sooner, increasing the likelihood that it could happen again.