On 1/8/2019 12:38 PM, Stephen Finucane wrote:
I have (1) fixed here:
https://review.openstack.org/#/c/629281/
That said, I'm not sure if it's the best thing to do. From what I'm hearing, it seems the advice we should be giving is to not mix instances with/without NUMA topologies, with/without hugepages and with/without CPU pinning. We've only documented the latter, as discussed on this related bug by cfriesen:
https://bugs.launchpad.net/nova/+bug/1792985
Given that we should be advising folks not to mix these (something I wasn't aware of until now), what does the original patch actually give us?
I think we should look at it from the other direction...what is the ultimate *desired* behaviour? Personally, I'm coming at it from a "small-cloud" perspective where we may only have one or two compute nodes. As such, the host-aggregate solution doesn't really work. I would like to be able to run cpu-pinned and cpu-shared instances on the same node. I would like to run small-page (with overcommit) and huge-page (without overcommit) instances on the same node. I would like to run cpu-shared/small-page instances (which float over the whole host) on the same host as a cpu-pinned/small-page instance (which is pinned to specific NUMA nodes). We have a warning in the docs currently that is specifically for separating CPU-pinned and CPU-shared instances, but we also have a spec that plans to specifically support that case. The way the code is currently written we also need to separate NUMA-affined small-page instances from non-NUMA-affined small-page instances, but I think that's a bug, not a sensible design. Chris