[nova][dev] Any VMware resource pool and shares kind of feature available in openstack nova?
Hi Matt, I will define/derive priority based on the which sub network the VM belongs to - mostly Production or Development. From this, the Prod VMs will have higher resource allocation criteria than other normal VMs and these can be calculated at runtime when a VM is also rebooted like how VMware resource pools and shares features work. I appreciate any other suggestions. thanks and regards,
On 2/22/2019 2:46 AM, Sanjay K wrote:
I will define/derive priority based on the which sub network the VM belongs to - mostly Production or Development. From this, the Prod VMs will have higher resource allocation criteria than other normal VMs and these can be calculated at runtime when a VM is also rebooted like how VMware resource pools and shares features work.
It sounds like a weigher in scheduling isn't appropriate for your use case then, because weighers in scheduling are meant to weigh compute hosts once they have been filtered. It sounds like you're trying to prioritize which VMs will get built, which sounds more like a pre-emptible/spot instances use case [1][2]. As for VMware resource pools and shares features, I don't know anything about those since I'm not a vCenter user. Maybe someone more worldly, like Jay Pipes, can chime in here. [1] https://www.openstack.org/videos/summits/berlin-2018/science-demonstrations-... [2] https://specs.openstack.org/openstack/nova-specs/specs/stein/approved/enable... -- Thanks, Matt
On 02/22/2019 10:06 AM, Matt Riedemann wrote:
On 2/22/2019 2:46 AM, Sanjay K wrote:
I will define/derive priority based on the which sub network the VM belongs to - mostly Production or Development. From this, the Prod VMs will have higher resource allocation criteria than other normal VMs and these can be calculated at runtime when a VM is also rebooted like how VMware resource pools and shares features work.
It sounds like a weigher in scheduling isn't appropriate for your use case then, because weighers in scheduling are meant to weigh compute hosts once they have been filtered. It sounds like you're trying to prioritize which VMs will get built, which sounds more like a pre-emptible/spot instances use case [1][2].
As for VMware resource pools and shares features, I don't know anything about those since I'm not a vCenter user. Maybe someone more worldly, like Jay Pipes, can chime in here.
I am neither worldly nor a vCenter user. Perhaps someone from VMWare, like Chris Dent, can chime in here ;) Best, -jay
[1] https://www.openstack.org/videos/summits/berlin-2018/science-demonstrations-...
[2] https://specs.openstack.org/openstack/nova-specs/specs/stein/approved/enable...
On Fri, 22 Feb 2019, Jay Pipes wrote:
On 02/22/2019 10:06 AM, Matt Riedemann wrote:
On 2/22/2019 2:46 AM, Sanjay K wrote:
I will define/derive priority based on the which sub network the VM belongs to - mostly Production or Development. From this, the Prod VMs will have higher resource allocation criteria than other normal VMs and these can be calculated at runtime when a VM is also rebooted like how VMware resource pools and shares features work.
It sounds like a weigher in scheduling isn't appropriate for your use case then, because weighers in scheduling are meant to weigh compute hosts once they have been filtered. It sounds like you're trying to prioritize which VMs will get built, which sounds more like a pre-emptible/spot instances use case [1][2].
As for VMware resource pools and shares features, I don't know anything about those since I'm not a vCenter user. Maybe someone more worldly, like Jay Pipes, can chime in here.
I am neither worldly nor a vCenter user. Perhaps someone from VMWare, like Chris Dent, can chime in here ;)
As I understand the question the goal is to have a feature within nova itself which is similar to resource pools in vCenter (and the somewhat dynamic resource management they can do) but can be used with a group of kvm hypervisors. In which case Matt's answer is pretty much spot on: pre-emptibility, reservations, dynamic management of aggregates and the like. In which case there may be some value in seeing if Blazar can help (now or in the future). Another option would be dynamic management of the inventory and traits in placement by a custom third party agent that doesn't yet exist combined with clever flavor management. Sanjay, I suspect none of this really helps you all that much, sorry for that. If you feel like trying to describe what you want to accomplish in s slightly different way, that might help us come up with workarounds. (/me also not wordly, and only barely and sometimes a vCenter user, it's all magic in there) -- Chris Dent ٩◔̯◔۶ https://anticdent.org/ freenode: cdent tw: @anticdent
participants (4)
-
Chris Dent
-
Jay Pipes
-
Matt Riedemann
-
Sanjay K