[Starlingx-discuss] Patch Elimination Train version

Alonso, Juan Carlos juan.carlos.alonso at intel.com
Wed Nov 6 20:19:04 UTC 2019


Hello Ya,

I have an standard configuration local storage (2+2) baremetal environment, I can access to nova-compute service for both computes but I did not found a list of 'cpu_models', just found 'cpu_mode' and it is not set as 'custom' as documentation says:
[libvirt]
cpu_mode = host-model

So, I executed:
$ system helm-override-update --set livbirt.cpu_mode=custom stx-openstack nova openstack
$ system helm-override-update --set cpu_models=SandyBridge,IvyBridge,Haswell,Broadwell stx-openstack nova openstack

I check again /etc/nova/nova.conf inside pod, and still saw just 'cpu_mode = host-model'
Is this configuration be modified? If yes, how can I do it?

Also I could create flavors and instances with below configurations:
trait:HW_CPU_X86_AVX='required'
trait:HW_CPU_X86_AVX2='required'
trait:HW_CPU_X86_AVX512F='required'

One instance for each configuration and one instance with those three configuration in a same flavor.
But, how can I check the CPU model in the instance?

In the test cases I mention at the beginning of this threat it says that the instances with flavor property trait:HW_CPU_X86_AVX512F='required' should get ERROR, but it could be launched as ACTIVE.

Regards.
Juan Carlos Alonso

-----Original Message-----
From: wang.ya <wang.ya at 99cloud.net> 
Sent: Monday, November 4, 2019 9:01 PM
To: Hu, Yong <yong.hu at intel.com>; Jones, Bruce E <bruce.e.jones at intel.com>; Alonso, Juan Carlos <juan.carlos.alonso at intel.com>; Troyer, Dean <dean.troyer at intel.com>
Cc: Cabrales, Ada <ada.cabrales at intel.com>
Subject: Re: Patch Elimination Train version

Hi, yong:

*vCPU Models selection* feature has been completed in train release.

The feature's workflow is as follow:

1. Configure the *cpu_models* option under [libvirt] module of nova-compute service.
     The *cpu_models* option is an ordered list of CPU models. It's sorted by CPU flags, from less to more. For example:
     	kvm64, IvyBridge, Haswell
     In above example, kvm64 only has base flags, IvyBridge has flag *avx*, Haswell have both *avx* and *avx2*.
     Please note that the configured CPU models must compatible with compute node's host CPU.

2. Create/Set flavor with required traits. The nova compute will select the CPU model from [libvirt]/cpu_models which first meet the flavor's requested flags. For example:
     	openstack flavor set 1 --property trait:HW_CPU_X86_AVX=required
     In above example, we mark the flavor request flag *avx*, if the CPU of compute node has flag *avx*, then the instance can be scheduled success and created with CPU model IvyBridge, otherwise the API call will get an error: "No valid host". 

Spec: http://specs.openstack.org/openstack/nova-specs/specs/train/approved/cpu-model-selection.html
Code: https://review.opendev.org/#/q/status:merged+project:openstack/nova+branch:master+topic:bp/cpu-model-selection

Hope the above info can help you. Also, I'll online in today's meeting :)

Best Regards


On 2019/11/5, 9:13 AM, "Yong Hu" <yong.hu at intel.com> wrote:

    + Wang Ya.
    
    Hi Ya,
    I remember your team has been upstreaming the related patches.
    Could you share some info with us here or in the meeting tonight?
    
    regards,
    Yong
    
    On 2019/11/5 6:37 AM, Jones, Bruce E wrote:
    > + Dean and Yong (the new PL for the OpenStack team)
    > 
    > AFAIK the CPU model changes that went into Train are internal only and 
    > don’t impact OpenStack operation.
    > 
    > Suggest you bring all of this to the distro.openstack meeting tomorrow 
    > where it can be discussed with the right people.
    > 
    >       brucej
    > 
    > *From:* Alonso, Juan Carlos
    > *Sent:* Monday, November 4, 2019 2:30 PM
    > *To:* Jones, Bruce E <bruce.e.jones at intel.com>
    > *Cc:* Cabrales, Ada <ada.cabrales at intel.com>
    > *Subject:* Patch Elimination Train version
    > 
    > Hi Bruce,
    > 
    > In last test plan for Patch Elimination Stein version, there was some 
    > *vCPU Model (Nova)* feature test cases that was marked as Unlikely status:
    > 
    > 1. Create a flavor without any extra specs. Launch instance using this 
    > flavor. Check instance cpu_model is 'kvm64'
    > 
    > 2. Create a flavor with extra specs: 'trait:HW_CPU_X86_AVX2=required'. 
    > Launch instance using this flavor. Check instance is Haswell and is in 
    > compute node 2.
    > 
    > 3. Create a flavor with extra specs: 
    > 'trait:HW_CPU_X86_AVX512F=required'. Launch instance using this flavor. 
    > Check API call get an error because no compute node has property 'avx512F'.
    > 
    > 4. Create a flavor with multiple extra specs: 'trait:xxx'. Launch 
    > instance using this flavor.
    > 
    > 5. Create a flavor and ignoring cpu_model. Launch instance using this 
    > flavor.
    > 
    > For Train version, is such *vCPU Model (Nova)* feature still out?
    > 
    > Also, there was some Live Migration Bugs table related to Neutron:
    > 
    > Enable delete bound trunk for linux bridge agent
    > 
    > 0040-rpc-removing-timeout-backoff-multiplier.patch (2e4ca00)
    > 
    > 0062-dvr-force-admin-state-update-before-distri.patch (edf9731)
    > 
    > 0065-dvr-do-not-create-agent-gateway-ports-unle.patch (2857911)
    > 
    > 0077-api-reject-routes-with-invalid-network-val.patch (6ed8fe6)
    > 
    > 0014-add-support-for-querying-quotas-with-usage.patch (71c07d7)
    > 
    > Prevent DHCP from processing stale RPC messages
    > 
    > 0039-dhcp-handle-concurrent-port-creation-error.patch (a38f89)
    > 
    > 0047-wsgi-prevent-accepting-socket-without-a-gr.patch (8e72491)
    > 
    > I need to know if such features/tests are going to be testes or not, can 
    > you help me to verify?
    > 
    > Test Plan for Stein: 
    > https://docs.google.com/spreadsheets/d/1jsTyzy1HT7VR4EUsbBRaTYvs0BlXeGWkhLD0pJmFChc/edit#gid=1511025137
    > 
    > Test Plan for Train: 
    > https://docs.google.com/spreadsheets/d/1n9P7as4iqO55DytIRuIGdyc7x0n2WM_q-cMfHcItJlY/edit?pli=1#gid=184151983
    > 
    > Regards.
    > 
    > Juan Carlos Alonso
    > 
    
    




More information about the Starlingx-discuss mailing list