Thursday, 2019-09-05

*** zhurong has quit IRC00:38
*** chenke has joined #openstack-cyborg01:14
*** zhurong has joined #openstack-cyborg01:21
*** s_shogo has joined #openstack-cyborg02:31
*** chunxiu has joined #openstack-cyborg02:51
*** Sundar has joined #openstack-cyborg02:55
*** xinranwang has joined #openstack-cyborg02:56
SundarHi xinranwang03:01
SundarHi chenke03:01
Sundar#startmeeting openstack-cyborg03:01
xinranwangHi Sundar03:01
openstackMeeting started Thu Sep  5 03:01:40 2019 UTC and is due to finish in 60 minutes.  The chair is Sundar. Information about MeetBot at
openstackUseful Commands: #action #agreed #help #info #idea #link #topic #startvote.03:01
*** openstack changes topic to " (Meeting topic: openstack-cyborg)"03:01
xinranwangHi chenke03:01
openstackThe meeting name has been set to 'openstack_cyborg'03:01
*** shaohe_feng has joined #openstack-cyborg03:01
Sundar#topic Who's here03:01
*** openstack changes topic to "Who's here (Meeting topic: openstack-cyborg)"03:01
xinranwang#info xinranwang03:02
s_shogo#info s_shogo03:02
chenke#info chenke03:02
SundarHi all03:02
Sundarzhurong: Welcome03:02
SundarThe main item on the agenda is, of course, the reviews03:03
SundarPlease focus on the P5-P9 patches. They and Nova notifn patch need to merge soon, before we get reviews of Nova patches. Also, they are needed for tempest to work, and that is a prereq too.03:04
Sundaryikun and all: Would you be able to review it? (Sorry for calling you out, yikun :))03:05
yikunSure, I will review it today03:05
SundarThanks :)03:05
*** wangzhh has joined #openstack-cyborg03:06
yikunand one detail question about db migrations on P5.503:06
SundarWe had a bit of a wrinkle on the Nova side too. The Cyborg's notification event would come too soon after we start the bind and would get lost. I think we have a way to solve it which (a) should hopefully be acceptable to Nova and (b) does not involve change to Nova objects or db03:06
yikunCan we append the change on existing migration script?03:06
yikunand most of other patches are okay to me03:07
Sundaryikun: There is one migration script for v2 API, so that we can show people exactly what we did. Once we releaseTrain, of course, previous migration scripts cannot be edited.03:08
SundarAs long as we don't change the migration script for Stein, I think we should be ok. Any concerns?03:08
yikunit will break upgrade from existing to now, but as you said if we still think the train relaase it's a start version,  if other people have no question,  I have no objection03:10
SundarI understand the existing practice but (a) too many migration scripts slows down upgrade for operators and (b) it is good to show onw script for all v2 API IMHO. Thanks for your understanding.03:12
wangzhhAgree with yikun. It's better to append a migration script. And other patch is fine with me.03:12
wangzhh*patches are03:12
Sundarwangzhh: Given the reasons above, would you still ask for a separate migration script?03:13
*** Coco_gao_ has joined #openstack-cyborg03:14
Coco_gao_#info Coco_gao03:14
SundarHi Coco_gao03:15
SundarWe are discussing the patch review03:15
wangzhhSo, we 'll consider T as a start version?03:15
SundarStein is the starting version -- v1 API. We should not modify or change the Stein migration script. We introduced a new migration script in Train, for v2 API. Anything unrelated to it can be in a separate script. But, so far, everything we did has been for v2 and Nova integ, I think.03:17
SundarFrom operators' point of view, the migration scripts that are part of a release should not be changed. Within a release, we usually introduce separate scripts for developers' convenience, I think.03:18
wangzhhUh, Okay. I have no objection.03:19
SundarAnyways, that is just my view. Thank you, wangzhh, yikun and all.03:19
SundarAny other concerns or thoughts on P5-P9?03:19
chenkewhy are you put -2 on P5?03:21
Sundarchenke: It is only a procedure, so that the patch series does not merge piece by piece. Once we get +2 on all patches, I will remove -2, so that it will all merge together. Nova guys are following this practice for my patch series too:03:23
SundarI should have done that right at the beginning.03:23
SundarSo, even P5 can be reviewed fully.03:24
chenkeOkay. I understand it.03:24
SundarCan we hope it to merge them by tomorrow? I don't want to rush you :)  but till P5-P9 and notificaiton patch merge, we cannot even ask Nova folks for review03:25
SundarAnd Sep 9 deadline is just few days away03:25
wangzhhOne question:  Related comment in this patch.  I did't find program in update ARQ.03:26
Sundarupdate refers to bind/unbind here, and that would involve programming.03:28
SundarAre you saying the term 'update' should be changed?03:28
wangzhhNo. Just check if it is admin or not before program. As I replied.03:29
SundarHmm, it could be owner or admin, right? Whoever created the ARQ should be allowed to bind it, otherwise they should not be able to create it.03:30
wangzhhYep.  Owner or admin should be allowed to Bind and Unbind. But when  involve programming, we should check if it is admin or not.03:32
wangzhhOnly admin can program? Right?03:32
SundarBinding involves programming for FPGAs. A tenant who has the privilege to create ARQs for an FPGA should also be able to program it, right? That's why I raised the question about 'create' privileges.03:33
shaohe_fengprogram had better have sudo privilege no node03:34
SundarYes, we do sudo fpgaconf today, and have a patch to move to oslo privsep.03:34
shaohe_fengyes. let's from simple03:34
shaohe_fengFirstly it can work.03:35
shaohe_fengand let nova can call cyborg03:35
shaohe_fengthen improve cyborg03:35
Sundarwangzhh: I understand your concern. However, if only admins can create/bind ARQs for FPGAs, that will be very restrictive, right?03:35
chenkeYes. Firstly make it work.03:35
wangzhhYep. Sundar.03:36
wangzhhSo, Just make it work first.03:36
SundarOk, I agree with wangzhh that security is important and with others that we should first get it to work by Train-3. I'll bring RBAC on the agenda for the next IRC meeting. Hopefully, other things will be working by then. :)03:37
Sundar#topic Python 303:37
*** openstack changes topic to "Python 3 (Meeting topic: openstack-cyborg)"03:37
shaohe_fengthe admin or  tenant own the bit steam image, can download the image and to do program.03:38
SundarThere is a Train goal to enable Py3 by Train-3 milestone, which is next week. Thanks to s_shogo (and Ikuo) for the patch. Hope we can merge that too by next week.03:38
shaohe_fengwe should challenge the tenant03:39
Sundarshaohe_feng: Sure, let's take it up next week.03:39
s_shogoI'll check  again that soon,  after P5-P9 merged. (latest patch seems to be no problem)03:39
SundarThanks, s_shogo.03:40
Sundar#topic AoB03:40
*** openstack changes topic to "AoB (Meeting topic: openstack-cyborg)"03:40
s_shogoAbout cyborg client.03:41
xinranwangHi all, I have to drop now. Thanks for your reviews for placement patch, it is merged finally :) And please review tempest code and notify patch too.03:41
SundarOutside of patches and Py3, I don't have much for today. As i said, I think I have a way to avoid the race in Nova, and will work on getting it out. Anything else for today?03:41
Sundars_shogo: Yes03:41
SundarI saw your patch -- thanks.03:42
s_shogoI have posted a patch for OpenStack SDK, and request review to SDK team.
s_shogoSubsequently ,I'll post cyborg-python client patch ,please review that > All03:42
SundarGreat, sure, will do.03:42
shaohe_fengone thing.03:43
shaohe_fengwe may start several asyn jobs03:43
shaohe_fengonly all jobs are finished, the bind is successful03:44
shaohe_fengone of the jobs failed, then the bind should be failed03:44
shaohe_fengso how to sync all the state of jobs?03:45
SundarWait for all of them to finish. I think eventlet has a wait() method, IIUC03:46
shaohe_fengbut wait is block03:46
shaohe_fengso it can not in main thread.03:47
shaohe_fengalso another,  if one jobs failed, we should cancel others, right?03:47
shaohe_fengwhere should we put the wait?03:47
SundarIt can't be in main thread, I agree. We don't need to cancel the rest if one fails, at least in Train. As we discussed, if the job has already started flashing the card, we shouldn't cancel it then03:48
SundarJust a thought: May be we have one eventlet whichis a master and it launches other eventlets, and the master does the wait.03:49
shaohe_fengyes, we need a such mechanism03:50
shaohe_fengI'm thinking it.03:50
shaohe_fengthe master thread should not be from a  thread pool, right?03:50
shaohe_fengfor example, there a 2 pools.03:50
SundarSure. My suggestion is to do something super-simple for Train-3 milestone -- may be just one eventlet for all ARQs, because we won't have too many ARQs anyway03:51
shaohe_fengwe create 2 jobs, so no pool for the master thread to monitor these 2 jobs.03:51
SundarSimplcity and reliability are more important, IMHO. It will also reduce your work :)03:51
shaohe_fengyes, agree.03:52
shaohe_fengasync is more complex than sync.03:52
SundarYes. IMHO, we can even leave unbind as sync in the worst case, for now, because we don't do anything there today03:53
shaohe_fengyes. agree03:53
SundarWe just want some patch in the works by Sep 3, so that we can kickstart the review03:53
SundarI mean, Sep 903:54
SundarAnything else, anybody?03:55
SundarCool. Thank you, all. Please review the patches. Have a good day! Bye.03:56
*** openstack changes topic to "Pending patches (Meeting topic: openstack-cyborg)"03:56
openstackMeeting ended Thu Sep  5 03:56:41 2019 UTC.  Information about MeetBot at . (v 0.1.4)03:56
openstackMinutes (text):
wangzhhBye everyone.03:56
*** s_shogo has quit IRC04:00
*** tetsuro has joined #openstack-cyborg04:38
*** chunxiu has quit IRC04:48
*** tetsuro has quit IRC04:50
*** tetsuro has joined #openstack-cyborg04:52
*** Sundar has quit IRC05:13
*** shaohe_feng has quit IRC05:35
*** changzhi has joined #openstack-cyborg05:49
*** wangzhh has quit IRC06:05
*** links has joined #openstack-cyborg06:10
*** tetsuro has quit IRC06:22
*** zhurong has quit IRC06:37
*** openstackgerrit has quit IRC06:51
*** tetsuro has joined #openstack-cyborg06:55
*** tetsuro has quit IRC06:59
*** changzhi has quit IRC07:37
*** zhurong has joined #openstack-cyborg07:53
*** openstackgerrit has joined #openstack-cyborg08:20
openstackgerritpengyuesheng proposed openstack/cyborg master: Blacklist eventlet 0.25.0
openstackgerritzhurong proposed openstack/cyborg master: Fix placement report error in python3
openstackgerritpengyuesheng proposed openstack/cyborg master: Blacklist eventlet 0.23.0,0.25.0
*** Coco_gao_ has quit IRC09:35
*** tetsuro has joined #openstack-cyborg10:04
*** tetsuro has quit IRC10:08
*** tetsuro has joined #openstack-cyborg10:24
*** tetsuro has quit IRC11:07
*** chenke has quit IRC14:11
*** jraju__ has joined #openstack-cyborg14:30
*** links has quit IRC14:31
*** jraju__ has quit IRC14:35
*** links has joined #openstack-cyborg15:17
*** altlogbot_3 has quit IRC16:01
*** altlogbot_3 has joined #openstack-cyborg16:02
*** irclogbot_3 has quit IRC16:03
*** irclogbot_2 has joined #openstack-cyborg16:03
*** irclogbot_2 has quit IRC16:07
*** irclogbot_2 has joined #openstack-cyborg16:07
*** xinranwang has quit IRC16:15
*** links has quit IRC17:19
*** efried is now known as efried_afk18:52
*** efried_afk is now known as efried19:13
openstackgerritSundar Nadathur proposed openstack/cyborg master: P5: Basic changes for API layer.
openstackgerritSundar Nadathur proposed openstack/cyborg master: P5.5: Additional changes to db layer.
openstackgerritSundar Nadathur proposed openstack/cyborg master: P6: Changes for device profiles.
openstackgerritSundar Nadathur proposed openstack/cyborg master: P7: Changes for ARQs and ExtARQs.
openstackgerritSundar Nadathur proposed openstack/cyborg master: P8: Changes for programming.
openstackgerritSundar Nadathur proposed openstack/cyborg master: P9: Enable v2 API in devstack.
openstackgerritSundar Nadathur proposed openstack/cyborg master: Notify Nova when binding finished
*** efried is now known as efried_afk20:16
-openstackstatus- NOTICE: Gerrit is being restarted to pick up configuration changes. Should be quick. Sorry for the interruption.21:13
*** efried_afk is now known as efried21:58
openstackgerritSundar Nadathur proposed openstack/cyborg master: Add driver_name to deployable reported by fake driver.

Generated by 2.15.3 by Marius Gedminas - find it at!