Thursday, 2020-11-26

*** swp20 has joined #openstack-cyborg02:32
*** Yumeng has joined #openstack-cyborg03:02
swp20hi Yumeng03:02
Yumenghi wenping03:03
*** xinranwang has joined #openstack-cyborg03:06
xinranwangHi all03:07
Yumenghi xinranwang03:07
swp20the question about un-hotplug GPU is because the image is not available for cirros, when i use centos, it's ok for re-hotplug.03:07
swp20hi xinranwang03:07
Yumengswp20: what's the question?03:08
swp20i asked you on wechat03:09
swp20un-hotplug not success with qemu driver in use.03:09
Yumenglet's start the meeting first :)03:09
Yumeng#startmeeting openstack-cyborg03:09
openstackMeeting started Thu Nov 26 03:09:37 2020 UTC and is due to finish in 60 minutes.  The chair is Yumeng. Information about MeetBot at
openstackUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.03:09
*** openstack changes topic to " (Meeting topic: openstack-cyborg)"03:09
openstackThe meeting name has been set to 'openstack_cyborg'03:09
Yumeng#topic Roll call03:09
*** openstack changes topic to "Roll call (Meeting topic: openstack-cyborg)"03:09
Yumeng#info Yumeng03:09
xinranwang#info xinranwang03:10
swp20#info swp2003:10
Yumeng#topic Agenda03:10
*** openstack changes topic to "Agenda (Meeting topic: openstack-cyborg)"03:10
Yumeng# topic vgpu03:11
Yumeng#topic vgpu03:11
*** openstack changes topic to "vgpu (Meeting topic: openstack-cyborg)"03:11
Yumengswp20: pls continue03:13
Yumengare you saying the detach failure in hotplug is because of cirros mssing?03:14
swp20yeah, the cirros vm process crash when detach the gpu device03:14
swp20the un-hotplug is not real success in fact.03:15
swp20cirros image is not support and centos is well.03:15
Yumengso vgpu hotplug is not supported in cirros but supported in centos, right?03:18
swp20i am not sure.03:18
swp20i means re-hotplug03:18
Yumengdid you find out why un-hotplug is not successful?03:19
swp20attach, detach and reattach03:19
swp20i search the vm log03:20
swp20there is process crash problem.03:21
Yumengis it a occasional case or it crash every time?03:23
swp20it's high probability03:23
YumengHas this crash ever happend in Centos?03:24
swp20hasn't met yet.03:25
Yumengok. got taht.03:25
Yumenglooks like hotplug is image sensitive.03:26
YumengThanks wenping for the sharing03:26
swp20maybe the driver is important.03:26
Yumengdo you mean nvidia virtualization driver?03:27
swp20no, i means the driver in image03:27
swp20gpu is not support well for cirros03:28
swp20include vgpu03:28
swp20you can test for vgpu about detach by 'virsh detach-device'03:28
Yumengyes, the VFIO mdev driver is very important. nvidia virtualization driver version must be well match the image version03:29
Yumengok. will try when I got time03:29
YumengI also have a vGPU issue to discuss with you03:30
Yumengabout the vGPU support03:30
swp20i think in the time of bind arq is better03:30
swp20to create mdev03:31
swp20attach_handle is too early03:31
swp20and maintain task is heavy03:32
Yumengyes, I also think so.03:32
swp20so let's confirm this.03:33
Yumengxinranwang what do you think?03:34
YumengSylvain prefer create mdev in generate attach_handle. sean and gibi is fine with either03:35
Yumengfrom my perspective, I also prefer creating medv when arq bind03:36
xinranwangif gpu's type is determined, the max number of vfs is also determined, right?03:37
Yumengbut if it is changed, we need to delete all the created ones and create new ones03:38
xinranwangif we do not create mdev at attach_handle generation step, how many vfs should we report?03:38
Yumengeven if they were never used.03:38
Yumengxinranwang: the maximum number03:40
Yumengin the inventory, we always report the maximum number03:40
xinranwangok, got it.03:40
xinranwangit seems create mdev during binding is more efficient. we just create it when we use it.03:41
Yumengyes, that's also how I mentioned in nova spec.03:42
xinranwangdoes mdev creation spend much time?03:42
Yumengnot much.03:42
xinranwangwill it fail in some cases?03:43
YumengI tested in my env, but it was not a big number of VMs. create mdev is very fast03:44
Yumengbut not sure what's the results when VM is a large number03:44
xinranwangmdev creation is a serial task, i think.03:45
xinranwanganyway, i think at binding step is more efficient, if there is no obvious gap.03:46
Yumengcreation failure is at very Low frequency.03:46
Yumenghasn't met yet03:46
Yumengxinranwang: ok. cool03:47
YumengSo we agreed on at binding step.03:47
YumengI will go back to sync with nova guys03:47
Yumengok. nothing else from side.03:47
YumengIs ther anything else you guys what to mention?03:48
xinranwangnothing from my side03:49
Yumenglunch time~~03:51
xinranwangbon appetit03:51
Yumengso let's wrap up today's meeting03:52
*** openstack changes topic to "Pending patches (Meeting topic: openstack-cyborg)"03:52
openstackMeeting ended Thu Nov 26 03:52:41 2020 UTC.  Information about MeetBot at . (v 0.1.4)03:52
openstackMinutes (text):
*** swp20 has quit IRC04:15
*** xinranwang has quit IRC06:06
openstackgerritXinran WANG proposed openstack/cyborg master: refresh device profile link in API doc
*** swp20 has joined #openstack-cyborg07:46
openstackgerritXinran WANG proposed openstack/cyborg master: refresh device profile link in API doc
openstackgerritYumengBao proposed openstack/cyborg-specs master: Cyborg NVIDIA GPU Driver support vGPU management
*** swp20 has quit IRC13:35
*** swp20 has joined #openstack-cyborg13:36
*** songwenping_ has joined #openstack-cyborg15:34
*** swp20 has quit IRC15:37
*** Yumeng has quit IRC17:12

Generated by 2.17.2 by Marius Gedminas - find it at!