[nova] PTG aligning of nova spec: show-server-numa-topology

Alex Xu soulxu at gmail.com
Wed May 15 14:07:17 UTC 2019


John Garbutt <john at johngarbutt.com> 于2019年5月10日周五 下午11:28写道:

> Hi,
>
> My main worry was to not expose host related information to end users, but
> noting administrators probably do what the information.
>
> Looking again at the Stein spec we merged, the proposed policy rules
> already take care of all that.
>
> I think the next step is to re-propose the spec for the Train release. I
> couldn't find it, but maybe you have done that already?
>
> Thanks,
> johnthetubaguy
>
> On Fri, 10 May 2019 at 08:51, yonglihe <yongli.he at intel.com> wrote:
>
>> Hi,  Everyone
>>
>> I synced up with Alex about comments we got at PTG.  It's a long
>> discussion, I might lost something.
>>
>> What i got lists below,  fix me:
>>
>> *  Remove sockets
>> *  Remove thread_policy
>>
>> Not sure about following comments:
>>
>>
>> *  Remove the cpu topology from the proposal?
>> *  Using the cpu pinning info instead of cpu set?
>>
>>
>> By apply the suggestion, the API ``GET /servers/{server_id}/topology``
>> response gonna to be like this,
>>
>> and let us align what it should be:
>>
>>   {
>>           # overall policy: TOPOLOGY % 'index
>>           "nodes":[
>>                      {
>>                        # Host Numa Node
>>                        # control by policy TOPOLOGY % 'index:host_info'
>>                        "host_numa_node": 3,
>>                        # 0:5 means vcpu 0 pinning to pcpu 5
>>                        # control by policy TOPOLOGY % 'index:host_info'
>>                        "cpu_pinning": {0:5, 1:6},
>>                        "vcpu_set": [0,1,2,3],
>>                        "siblings": [[0,1],[2,3]],
>>                        "memory_mb": 1024,
>>                        "pagesize_kb": 4096,
>>                        "cores": 2,
>>                        # one core has at least one thread
>>                        "threads": 2
>>
>
I'm not sure the pagesize_kb, cores and threads. I guess Sean will have
comment on them.

We can't get cores for threads for each numa node. The
InstanceNUMACell.cpu_topology is empty only except for the dedicated cpu
policy. And the Instance.cpu_topology is for the whole instance.

And we don't support choice page size per numa node.

Thanks for John and Sean's discussion in the PTG, and sorry for loss the
conversation in PTG again. Actually, those are the questions Yongli is
looking for.


>                      }
>>                      ...
>>                     ], # nodes
>>      }
>>
>>
>> links:
>>
>> ptg:
>> https://etherpad.openstack.org/p/nova-ptg-train L334
>>
>> spec review:
>>
>> https://review.opendev.org/#/c/612256/25/specs/stein/approved/show-server-numa-topology.rst
>>
>> code review:
>> https://review.openstack.org/#/c/621476/
>>
>> bp:
>> https://blueprints.launchpad.net/nova/+spec/show-server-numa-topology
>>
>>
>> Regards
>> Yongli He
>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-discuss/attachments/20190515/fa4ffefa/attachment.html>


More information about the openstack-discuss mailing list