Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-08-19
10:10:43 stephenfin That'll be set on both the compute node and controller
10:11:13 stephenfin Once you've upgraded everything to Train, you'll toggle the config option on the controller, followed by the config options on the compute node
10:13:35 bhagyashris Okay and then that will translate the flavor extra specs at controller node and will report the PCPU inventory at placement side .... Right?
10:13:44 stephenfin once toggled, yes
10:13:49 stephenfin and do the reshape
10:14:19 stephenfin Until it's toggled though, the compute nodes will not report PCPUs, the extra specs will not be translated, and the reshape will not occur
10:18:09 bhagyashris And if the global config option set is false that means not toggle and user defines the cpu_dedicated_set then in that case what will happen?
10:18:28 stephenfin We error out
10:19:42 bhagyashris stephenfin: means 'pcpus_in_placement' and the cpu config options should be set properly by operator
10:20:06 stephenfin Well, by the deployment tool they're using
10:20:24 stephenfin Actuallly
10:20:37 stephenfin We don't necessarily need to set 'pcpus_in_placement'
10:20:45 stephenfin ...on the compute nodes
10:21:23 stephenfin We could just say "set 'pcpus_in_placement'" on the controller, and then set 'cpu_dedicated_set' and 'cpu_shared_set' on the compute nodes
10:22:33 stephenfin However, fwiw, I wouldn't worry about this too much right now. I assume you're still testing things out? This should only affect upgrades and nothing else
10:24:45 bhagyashris Yeah actually I was fixing the review comments that you have given patch https://review.opendev.org/#/c/674895/ and also added the functional test case for upgrade ... but looking at the recent patch set you have added the functional test ... so that makes the duplications of efforts
10:26:40 stephenfin Yeah, I've reworked the reshape to address one of my comments. I don't know how much sense the new approach makes. I know mriedem has some comments left that I need to address
10:26:45 bhagyashris stephenfin: and from above discussion it seems that upgrade will need some changes
10:27:34 bhagyashris stephenfin: okay ..
10:28:52 bhagyashris stephenfin: I just have one question from above discussion you said that "set 'pcpus_in_placement'" on the controller, and then set 'cpu_dedicated_set' and 'cpu_shared_set' on the compute nodes and We don't necessarily need to set 'pcpus_in_placement' on compute node
10:29:05 stephenfin yup?
10:29:53 bhagyashris So in that case if the 'pcpus_in_placement' = false at controller node and at co mute if I define the cpu_dedicated_set then this will report the PCPU as inventory at placement
10:30:12 bhagyashris And the instance request will ask for VCPU in that case
10:30:15 bhagyashris Right?
10:30:28 bhagyashris Then that looks like issue then
10:30:40 stephenfin Yeah, and that's an issue. I was thinking we'd check to see if PCPUs were reported by any compute node and error out if so
10:30:53 stephenfin I haven't thought through it properly yet though
10:31:32 bhagyashris stephenfin: okay
10:32:35 stephenfin There's one other thing I can do, which is add a check on the compute node to see if I'm requesting PCPUs (either directly or through 'hw:cpu_policy' and just hard fail if so. That would involve making a request to placement to get allocations though and I'm not sure we can do that.
10:32:46 stephenfin As you can see, this is very much in flux at the moment :(
10:37:11 bhagyashris stephenfin: yeah,
10:45:02 bhagyashris if any idea will comes to my mind I will let you know
11:11:15 openstackgerrit Andreas Jaeger proposed openstack/nova stable/rocky: Remove experimental job on openSUSE 42.3 https://review.opendev.org/677176
11:12:52 openstackgerrit Andreas Jaeger proposed openstack/nova stable/queens: Remove experimental job on openSUSE 42.3 https://review.opendev.org/677177
11:14:04 openstackgerrit Andreas Jaeger proposed openstack/nova stable/pike: Remove experimental job on openSUSE 42.3 https://review.opendev.org/677178
11:15:20 alex_xu stephenfin: I replied your email, but not sure I understand the problem fully
11:16:10 openstackgerrit Brin Zhang proposed openstack/nova master: Specify availability_zone to unshelve https://review.opendev.org/663851
11:26:18 openstackgerrit Adam Spiers proposed openstack/nova master: Make it easier to run a selection of tests relevant to ongoing work https://review.opendev.org/677180
11:26:23 aspiers stephenfin, alex_xu: happy Monday ;-) ^^^^
11:29:57 openstackgerrit Adam Spiers proposed openstack/nova master: Make it easier to run a selection of tests relevant to ongoing work https://review.opendev.org/677180
11:50:08 openstackgerrit Adam Spiers proposed openstack/nova master: Switch some GitHub URLs to point to opendev.org https://review.opendev.org/677185
12:21:33 openstackgerrit Takashi NATSUME proposed openstack/nova master: Fix wrong assertions in unit tests https://review.opendev.org/677102
12:23:33 yonglihe stephenfin: Hi, good morning.
12:27:46 openstackgerrit Takashi NATSUME proposed openstack/nova master: Tests: autospecs all the mock.patch usages https://review.opendev.org/470775
12:37:46 sean-k-mooney yonglihe: i have not seen stephenfin online yet today
12:38:17 sean-k-mooney he proably is around but im guessing he is still catching up on email/gerrit after the weekend
12:42:04 stephenfin no, I'm here
12:42:28 stephenfin (sean-k-mooney: See discussion with bhagyashris from earlier :))
12:42:57 sean-k-mooney stephenfin: ah ok sorry
12:43:07 sean-k-mooney yep i missed that
12:44:10 stephenfin yonglihe: Those patches are on my list, but there's a big queue of stuff ahead of them. I'll get to it eventually :)
12:47:12 openstackgerrit Brin Zhang proposed openstack/python-novaclient master: Microversion 2.77: Support Specifying AZ to unshelve https://review.opendev.org/665136
12:49:49 sean-k-mooney stephenfin: did you reach any decision with the path forward for PCPU upgrades
13:01:05 stephenfin sean-k-mooney: I'm still on the fence, tbh
13:01:37 sean-k-mooney i think the global + 3 cycle process will be the safest
13:01:38 stephenfin I'm leaning towards the global config option still, but we're going to need some way to ensure people don't toggle it on the compute node first
13:01:55 sean-k-mooney ya
13:03:18 sean-k-mooney i think our best way to prevent that is to work with the intaller project to support enabling it
13:03:27 stephenfin I think so too
13:03:36 stephenfin But I've no idea how to fix things if people do it wrong
13:03:49 stephenfin We're kind of handing people a loaded gun here
13:03:53 sean-k-mooney e.g how to rever a reshape
13:04:19 sean-k-mooney ya it is. its not the only one we have given people in the past
13:04:27 stephenfin Pretty much
13:04:31 stephenfin Or how to keep reshaping
13:05:18 sean-k-mooney have we talked to the OSA,kolla or triplo folk about it yet?
13:06:07 sean-k-mooney i know we plan to go over the triplo stuff internally with our deployment folks but did any of the install project comment on your ml post
13:06:41 stephenfin efried: I've started discussing it with TripleO guys, of course
13:06:54 stephenfin But not OSA or Kolla. I don't think Kolla supported pinned CPUs as a rule atm
13:07:21 stephenfin I recall being in a call where there was talk of it at a previous summit but I don't know how much further it went
13:09:46 stephenfin efried: I've noticed that there are many more gaps in test coverage with that privsep series. Are you okay with a giant follow-up to close said gaps?
13:09:57 openstackgerrit Adam Spiers proposed openstack/nova master: Add extra spec parameter and image property for memory encryption https://review.opendev.org/664420
13:10:14 stephenfin Cos I'll have to recuse myself from reviews if I go add said missing tests to each patch
13:10:38 stephenfin unless you're happy to single +2 the tests, that is
13:10:43 efried stephenfin: I have no doubt there are gaps in test coverage overall, but what bugs me is the coverage being reduced by each patch. Is that what you're talking about?
13:10:44 aspiers stephenfin: responses to your feedback there ^^^
13:10:56 stephenfin efried: Yeah, that specifically
13:11:11 stephenfin i.e. missing tests in nova/tests/unit/privsep
13:11:13 stephenfin aspiers: sweet
13:11:17 efried In that case, yes, a fup is good for me, but I'll want to wait to merge a given patch until its coverage is at least proposed.
13:11:50 aspiers stephenfin: but you might find more enjoyment in https://review.opendev.org/#/c/677180/
13:12:00 efried and yes, good plan wrt reviews, since it's tough to convince other cores to care about that series, we should take advantage of the two of us being willing to dig into it :P
13:12:01 aspiers efried: you too :)
13:12:41 aspiers my dream is never having to *manually* run tests locally again
13:12:52 aspiers just edit files and watch things turn red or green
13:13:07 aspiers it's entirely doable
13:13:14 aspiers I've done it with many other projects
13:13:17 stephenfin efried: Yeah, of course. Maybe just keep an eye out for gaps as you're reviewing so I've something to hit "Done" on
13:14:25 efried wfm
13:15:59 efried aspiers, stephenfin: Love this idea, add it to the list of workflow improvements it's hard to get people to pay attention to: https://review.opendev.org/#/c/673356/
13:16:29 aspiers efried: personally I *never* tire of workflow improvements
13:16:39 aspiers you can always count me in for reviewing those
13:18:04 efried aspiers: what's mybranch^! ?
13:18:21 aspiers the commit at the tip of mybranch
13:18:31 openstackgerrit Matt Riedemann proposed openstack/nova stable/ocata: Exclude build request marker from server listing https://review.opendev.org/677201
13:19:02 aspiers efried: https://git-scm.com/docs/git-rev-parse#_other_rev_parent_shorthand_notations
13:19:21 stephenfin Well it's not a workflow improvement, but this would save us some time https://review.opendev.org/#/c/470775/
13:19:52 stephenfin (make mock fail if you call a mocked function incorrectly or a function that doesn't exist)
13:20:35 stephenfin and it's been around in one form or another for...2 years, 2 months and 17 days :'D
13:21:05 efried yeah, I thought claudiub had been working on this for ages and was never able to get it quiiite right.

Earlier   Later