Current anti-affinity only allows instances equals to number of hypervisors

[Prashant Zanwar]

1. anti-affinity allows instances upto only number of hypervisors in an aggregate. It's a limitation, and possibly can be complimented with some round-robin filter, and computecapacity filter. Once all hypervisors rule is exploited, using anti-affinity rule, it can in round robin fashion choose host based on filter defined .. Should allow to choose from scheduler filters too, along with anti-affinity ..

[Michael McCune (mimccune)]

this sounds like a request for a feature improvement, we could definitely add more to improve the anti-affinity behavior. marking as wishlist for a future improvement.

[Prashant Zanwar (pzanwar)]

thanks. I was hoping this to be part of basic approach, but if not, it's fine for me it to e enhanced .. # of hypervisors can be limited, but i would still have # of instances many, so i should be able to apply affinity group rule for all of my vms.

[Michael McCune (mimccune)]

Prashant, we are discussing pursuing a path where the user could define the number of nodes per hypervisor when requesting anti-affinity for a process. Would this be sufficient for a first pass at this problem?

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/282506

[Vitalii Gridnev (vgridnev)]

We have blueprint for that: https://blueprints.launchpad.net/sahara/+spec/improving-anti-affinity