Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-02-14
16:18:56 bauzas above is the fact that when we create a new instance, we use a random mdev which is unrelated to the pgpu which was allocated by placement
16:18:56 efried (to both)
16:19:25 mriedem i don't have a strong feeling either way
16:19:46 mriedem ProviderTree is convenient albeit heavy weight and hopefully someone doesn't abuse it somehow, but doing our own mapping dict is redundant
16:19:49 mriedem so shrug
16:20:07 bauzas yeah me too, but I'm opinionated enough to make it the smoothiest
16:20:30 bauzas and caching the value in the libvirt driver object seems the smoothiest approach
16:20:45 mriedem so the provider tree would just get reset on every upt call?
16:20:47 mriedem if so, then sure
16:20:51 bauzas yeah
16:20:54 mriedem wfm
16:21:02 bauzas i'm done with complex cache invalidation mechanisms
16:21:13 mriedem are you SO done with them?
16:25:21 bauzas mriedem: /me tries to not think hard of AZs
16:25:34 bauzas or server groups
16:25:47 mriedem hey if you like azs i have some good news for you
16:25:48 mriedem https://review.openstack.org/#/c/509206/
16:25:51 mriedem finally fixing that shitty old bug
16:26:35 bauzas \o/
16:26:46 bauzas mriedem: straight adding it into my queue
16:27:39 bauzas gibi: you'll be happy to hear that functests work now, btw. :)
16:28:07 gibi bauzas: awesome
16:29:24 gibi every time I start kicking https://review.openstack.org/#/c/635859/ I got depressed after 15 minutes about I capabilities to understand what happening
16:29:38 melwitt o/
16:30:14 melwitt gibi: me too. I tried for awhile and gave up
16:31:31 gibi melwitt: but I feel something deep is behind that so I cannot gave up :)
16:32:09 melwitt gibi: that's good, I think you will get to the bottom of it :)
16:32:20 mriedem well if you want to get really depressed http://logs.openstack.org/47/635147/11/check/nova-tox-functional-py35/5c3c2a9/job-output.txt.gz#_2019-02-13_13_33_41_103932
16:32:34 mriedem that traceback wins the grand prize
16:32:41 mriedem and i don't know what to do about it
16:32:48 mriedem looks like the db dropped in the middle of the test
16:33:32 gibi mriedem: nice :)
16:33:56 mriedem the dumping of os-traits contents to the logs on each functional test run with the PlacementFixture isn't helping either
16:34:08 mriedem http://logs.openstack.org/47/635147/11/check/nova-tox-functional-py35/5c3c2a9/job-output.txt.gz#_2019-02-13_13_33_41_193979
16:34:24 gibi mriedem: I think we can move that to DEBUG
16:35:18 mriedem yeah probably http://logs.openstack.org/59/636459/2/check/tempest-full/3226eb1/controller/logs/screen-placement-api.txt.gz#_Feb_13_13_54_30_464250
16:35:23 gibi I'm affraid that the drop of the databease could be another interfering test case. Did you managed to reproduce that locally?
16:35:31 mriedem nope
16:35:35 bauzas gibi: efried: what's the best way to copy a providertree object ? copy.deepcopy() ?
16:35:54 openstackgerrit Elod Illes proposed openstack/nova master: DNM: test pip 19 - pip freeze with git editable https://review.openstack.org/636980
16:35:56 bauzas or do we have other built-in mechanism ?
16:36:18 bauzas tbc, I want to have a separate reference to be stored on the libvirt object
16:36:54 gibi bauzas: I would be affraid of copying the self.lock = lockutils.internal_lock(_LOCK_NAME) inside the ProviderTree
16:37:25 efried bauzas: upt is already operating on a copy... but I'm actually not positive that ufpt doesn't dork with it. So yeah, copy.deepcopy is probably the best bet.
16:37:58 gibi bauzas: besides the self.lock the rest of the internals seems copiable
16:38:18 bauzas gibi: so, ideally we should provide a copy method
16:38:19 efried gibi: hm, interesting point about the lock. This impl that bauzas is doing should be read-only. But that still wouldn't stop you from bouncing off the lock if the other instance is using it.
16:38:38 bauzas efried: what I want is purely read-only
16:38:45 efried However, I would be surprised if the lockutils lock doesn't override __copy__ to *not* copy itself.
16:38:49 bauzas actually, you know what ? I'm gonna make it explicit
16:38:55 bauzas by using a propertgy
16:39:11 bauzas we're not intended to manipulate this copy
16:39:12 gibi mriedem: I think I will stick to the remove-the-assert-that-blow-the-world issue, that I can reproduce locally at least
16:39:59 efried bauzas: There are methods to get a read-only copy of the data out of ProviderTree, but I'm pretty sure they only get one provider at a time.
16:40:22 efried Yeah, ProviderTree.data()
16:40:27 mriedem gibi: efried: fwiw i'm also holding off on the bw provider series until https://review.openstack.org/#/c/616239/ makes progress
16:40:40 mriedem and we've got 3 weeks from FF today
16:40:52 efried I guess that's my action
16:41:16 mriedem i'm just waiting for efried to +2 so i can nit pick a few things and make it look like i get it
16:41:17 gibi efried: thanks
16:41:21 bauzas efried: anyway, I'm rushing to propose a new rev
16:41:35 bauzas efried: you will be able to comment on the best approach by reviewing
16:41:41 bauzas not a big deal
16:41:50 gibi efried: to your surprise there is no copy overried in https://github.com/openstack/oslo.concurrency/blob/master/oslo_concurrency/lockutils.py ;)
16:41:57 efried gibi: d'oh!
16:44:11 efried uh, wtf. gibi, bauzas, when I try to deepcopy an object with a lockutils.internal_lock in it, I get an error.
16:44:35 efried so how tf would that be working in the rt? /me looks...
16:51:02 efried gibi, bauzas: http://paste.openstack.org/show/745111/ <== so how tf is https://github.com/openstack/nova/blob/master/nova/scheduler/client/report.py#L885 working??
16:51:37 bauzas heh
16:51:47 bauzas sorcery !
16:54:03 bauzas gibi: looking at the func test
16:54:05 bauzas gibi: you lost me
16:54:47 bauzas gibi: in theory, if you request a VGPU but we don't have inventories because the conf option is unset, you should end up with a beautiful NoValidHost
16:54:56 bauzas so I wonder why it works
16:55:08 bauzas oh wait
16:55:22 bauzas I missed the point you're creating the inventory manually
16:55:44 bauzas oki doki
16:55:58 bauzas so you do this to fake the fact you were running an old version
16:56:22 gibi bauzas: yeah, you are correct
16:56:51 bauzas ok, the good news is that I saw this when I looked at the provider_tree for the allocation RP UUID
16:57:02 bauzas and I was wondering why I was getting a root RP
16:57:10 bauzas gotcha
16:57:15 bauzas perfect, it works then
17:00:40 efried jaypipes: maybe you have some insight?
17:00:59 efried jaypipes: on: http://paste.openstack.org/show/745111/ <== so how tf is https://github.com/openstack/nova/blob/master/nova/scheduler/client/report.py#L885 working??
17:01:54 jaypipes looking...
17:04:09 gibi efried: sorry I'm expired for today
17:04:39 efried gibi: I think you've done enough, don't you? :P
17:05:06 efried "Ohh, I'm gibi, I notice things like lock copies, nya nya nya"
17:05:20 gibi :P
17:05:23 openstackgerrit Merged openstack/nova master: Use math.gcd starting with python 3.5 https://review.openstack.org/636669
17:05:35 openstackgerrit Merged openstack/nova master: doc: specify --os-compute-api-version when setting flavor description https://review.openstack.org/635508
17:05:46 openstackgerrit Merged openstack/nova master: Remove PLACEMENT_DB_ENABLED from nova-next job config https://review.openstack.org/634953
17:10:29 jaypipes efried: I have no idea how that works.
17:11:48 efried jaypipes: I figured we must be monkey-patching lockutils somewhere. Couldn't find it at a glance, but haven't finished looking.
17:12:02 jaypipes no idea. :(
17:13:16 openstackgerrit Stephen Finucane proposed openstack/os-vif master: WIP: Add API docs for various VIF types https://review.openstack.org/637009
17:15:01 openstackgerrit melanie witt proposed openstack/nova master: WIP Remove project_only=True from database queries https://review.openstack.org/637010
17:15:45 cdent efried: if you simply want to fix it, you can implement __deepcopy__ on the ProviderTree class
17:15:55 cdent but it's confusing that it hasn't come up before

Earlier   Later