On 06/18/2014 07:05 AM, Sean Dague wrote: > Because this is the way this degrades when we are using all our quota, > I'm really wary of adding these back until we discuss the expectations > here
This seems fair > We actually had 0 nodes in use or ready of the type at the time. Firstly I'm trying to understand why this happens before we grapple with any solution. Forgive me, as I'm new to nodepool code. I created a small snippet to try and understand the most basic behaviour [1]. e.g. --- (py27)$ python test.py 10 10 10 10 <AllocationGrant of 5 of fedora from hpcloud> <AllocationGrant of 5 of ubuntu from hpcloud> <AllocationGrant of 5 of fedora from rackspace> <AllocationGrant of 5 of ubuntu from rackspace> --- this seems correct with all nodes evenly distributed across two providers. The case of interest here is when the total requests is over the quota. That's when something has to give. As jeblair mentioned to me, if the requests are 50/50 then neither gets starved, e.g. if we want 50 of each node, but only have 20 total quota, we still see an even distribution: -- (py27)$ python test.py 10 10 50 50 <AllocationGrant of 5 of fedora from hpcloud> <AllocationGrant of 5 of ubuntu from hpcloud> <AllocationGrant of 5 of fedora from rackspace> <AllocationGrant of 5 of ubuntu from rackspace> --- The question for me is what happens as the ratio of requested ubuntu nodes to fedora nodes increases when over-quota. e.g. at 20:1, with 20 total nodes available, we still get our fedora node: --- (py27)$ python test.py 10 10 1 20 <AllocationGrant of 10 of ubuntu from hpcloud> <AllocationGrant of 1 of fedora from rackspace> <AllocationGrant of 9 of ubuntu from rackspace> --- but eventually, at 30:1, the fedora node gets dropped --- (py27)$ python test.py 10 10 1 29 <AllocationGrant of 10 of ubuntu from hpcloud> <AllocationGrant of 1 of fedora from rackspace> <AllocationGrant of 9 of ubuntu from rackspace> (py27)$ python test.py 10 10 1 30 <AllocationGrant of 10 of ubuntu from hpcloud> <AllocationGrant of 10 of ubuntu from rackspace> --- So, as a first step, would you agree that is a very simplified but otherwise accurate reflection of what happened? -i [1] https://gist.github.com/ianw/541d2b7ad8983acbf225 _______________________________________________ OpenStack-Infra mailing list [email protected] http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-infra
