Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-01-23
10:35:27 mdbooth sfinucan: Looking now.
10:36:35 mdbooth sfinucan: Incidentally, this one needs to get into Ocata: https://review.openstack.org/#/c/423392/ . It's a continuation of a previous code removal. Needs to be phased for upgrades, hence need for Ocata.
10:36:50 mdbooth It's very simple.
10:37:37 sfinucan mdbooth: Sure thing. I'll chuck that on my review queue :)
10:41:03 gmann alex_xu: yea, that will be nice to know for me too
10:47:19 moshele claudiub|2: sure I will take a look later today
10:51:33 cdent bauzas, jaypipes : was just reading back through the log about something to do with noauthmiddleware and fixtures and placement, but can't really piece together the gist. Were you able to work things out?
10:51:45 bauzas cdent: I fixed the problem
10:52:01 bauzas cdent: it was merged on Friday by adding a PlacementFixture
10:52:39 cdent bauzas: In any case there is a functional test of placement api that does not use gabbi: https://github.com/openstack/nova/blob/master/nova/tests/functional/api/openstack/placement/test_report_client.py
10:52:40 bauzas cdent: now I have another question about having resource usage call being optionals given the existing filters
10:52:55 bauzas cdent: I know
10:53:01 cdent bauzas: and the noauthmiddleware is _not_ something I wrote special, it is the nova one adapted
10:53:06 cdent do you have a link to the new fixture?
10:53:12 cdent (I mean the review that made it)
10:53:15 bauzas cdent: the problem was with the existing other functional tests that call the scheduler
10:53:35 bauzas cdent: see nova.fixtures.PlacementFixture
10:53:39 cdent Yeah, I've seen the mail about the optionals, it's next on my list to look at more closely
10:53:56 bauzas cdent: I'll be pushing a new PS by today
10:54:00 cdent cool
10:54:06 bauzas that would +1 the functional tests
10:55:02 bauzas I had a shit ton of problems related to the functional tests, but that's done
11:02:32 cdent bauzas: did you consider using wsgi-intercept (as in the pre-existing functional test) instead of doing a server.run()?
11:02:45 bauzas yup
11:02:59 cdent and that wouldn't work?
11:03:06 bauzas but we agreed with sdague on using the Server directly
11:03:22 bauzas cdent: honestly, that looks a nit
11:03:37 cdent I'm not complaining, I'm just trying to understand the decision process
11:03:40 bauzas cdent: given the short deadline, I don't want to discuss again about it
11:03:49 bauzas yeah, I can understand
11:03:58 bauzas but sometimes you need to find solutions that work easily
11:06:04 hachi hello
11:06:15 bauzas cdent: just one point
11:06:26 bauzas cdent: I'm not sure I like https://github.com/openstack/nova/blob/master/nova/tests/functional/api/openstack/placement/test_report_client.py#L89-L92
11:06:42 bauzas cdent: because we are mocking an internal attribute here
11:06:44 hachi how i can use the policy:
11:06:50 hachi "owner" : "user_id:%(user_id)s"
11:06:59 hachi in nova api 2.1
11:07:09 bauzas cdent: in case keystone changes its interface, we wouldn't have mock complaning at all
11:07:15 hachi with newton release
11:07:18 bauzas cdent: but then we would have huge problems
11:07:53 bauzas cdent: I rather preferred to use the keystone public way of avoiding the SC
11:08:17 cdent keystoneauth1 has made a commitment to not change the interface on its public methods without changing the entire name of the package?
11:08:21 bauzas but like I said previously, that looks a nit to me and we could discuss on that later on
11:08:38 bauzas cdent: sure, we can trust them
11:08:47 bauzas cdent: that's just riskier, that's it
11:08:52 bauzas anyway
11:09:17 cdent and even if it did break, we could just fix it when it broke: the tests will have done their job in that case
11:09:19 bauzas I'm totally up discussing on that after FF :)
11:09:52 bauzas probably we could accept not to fake our scheduler client methods and rather mock the Service object
11:10:22 bauzas but I think we could still use the Server proxy to really run an wsgi server and not intercept
11:10:47 bauzas but that's my thoughts - just coming along while tests are running on my local desktop and that I have a bit of time :)
11:11:46 cdent Out of curiosity why do you prefer the server proxy?
11:14:58 bauzas cdent: because that's running our internals
11:15:13 cdent What are "our internals"?
11:15:21 bauzas internal piece of code
11:15:59 bauzas and we're not emulating a socket, we *run* it
11:16:08 bauzas which makes possible to debug the app outside of the test
11:16:26 bauzas but meh.
11:16:58 bauzas anyway, back to work
11:17:00 cdent okay, but when the placement api runs for real, it doesn't use any of the nova Server() handling
11:23:11 openstackgerrit Bob Ball proposed openstack/nova: Use physical utilisation for cached images https://review.openstack.org/353862
11:25:00 openstackgerrit Claudiu Belu proposed openstack/nova: hyper-v: Adds support for VMs with nested virtualization https://review.openstack.org/398509
11:26:42 cdent bauzas: I've got a couple questions on the CoreFilter issue that are probably easier to track here than email. You got time to answer them now or should I just post a response to the email?
11:27:25 bauzas cdent: just throw me a bit of them
11:29:02 cdent In the world before Placement API existed, if I disabled the CoreFilter, that means that select_destinations could return some compute nodes that would reject a "placement" and cause a retry because it didn't have enough cores. Is that right?
11:31:28 openstackgerrit Alex Xu proposed openstack/nova: Limit the min length of string for integer JSON-Schema https://review.openstack.org/415103
11:31:47 cdent bauzus: If it is right, then asking the placement api for vcpu resources is simply short circuiting that retry and we should do that?
11:31:57 johnthetubaguy alex_xu: I would expect no policy.json file in a new install, empty I guess is OK too.
11:32:12 alex_xu johnthetubaguy: got it, thanks
11:32:13 johnthetubaguy alex_xu: in the future I was thinking a fully commented out yaml file, a bit like what we do for config
11:32:27 alex_xu johnthetubaguy: ok
11:33:26 alex_xu johnthetubaguy: I probably figure out a way to implement the 3 cases you pointed out. so I will work with Kevin_Zheng to update the patch.
11:35:00 johnthetubaguy alex_xu: Kevin_Zheng: let me know how that goes, for me the warning message is an optional extra, as we could leave the old rules ill we can get that into olso.policy
11:40:36 bauzas cdent: sorry, missed your question
11:40:54 bauzas cdent: in a world where an operator disables the CoreFilter, we don't verify VCPUs
11:41:14 bauzas cdent: most of them for example are disabiling the DiskFilter for the exact reason it doesn't count correctly
11:41:34 cdent we don't verify vcpus at the scheduler level but do at the compute node level, or we don't very vcpus _anywhere_?
11:41:51 bauzas cdent: if you don't run the CoreFilter, we don't return a limit
11:42:12 alex_xu hachi: we didn't suggest to use user_id based policy check. That was added back to the code just for backward-compatible, it may be removed in the future
11:42:16 bauzas cdent: if so, that's coming to the compute, and then we verify the claim without verifying the limit
11:42:44 bauzas cdent: which means no allocation ratio
11:42:45 cdent interesting
11:43:18 bauzas cdent: so, I'm a bit afraid that a lot of operators have disabled the DiskFilter
11:43:58 bauzas cdent: if now we check the disk size directly without having the aggregates created automatically for the RPs, then they could have problmes
11:44:05 cdent so your idea is to remove resource classes from the request to /resource_providers based on which filters are disabled
11:44:11 bauzas cdent: yup
11:44:23 bauzas cdent: lemme upload the change this afternoon and you'll see
11:44:26 cdent and the change would solely be on the nova-scheduler side
11:44:33 bauzas yup
11:45:03 cdent cool. It makes sense. My main concern was that we should change something on the placement server-side based on nova's config
11:45:08 bauzas I just have a problem with our FakeDriver creating the same hypervisor hostname for all our computes
11:45:28 bauzas so, once I fix that, I'll upload
11:46:26 openstackgerrit Chris Dent proposed openstack/nova: Optional separate database for placement API https://review.openstack.org/362766
11:49:07 openstackgerrit Stephen Finucane proposed openstack/nova: Fix and add some notes to the cells v2 first time setup doc https://review.openstack.org/421421
11:55:49 openstackgerrit Alvaro Lopez Garcia proposed openstack/nova: Ensure that periodic reclaim cleans DB deleted instances https://review.openstack.org/323250
12:00:27 openstackgerrit Zhenyu Zheng proposed openstack/nova: Fix device taging after API microversion 2.37 https://review.openstack.org/423952
12:28:16 openstackgerrit Wang Qing wu proposed openstack/nova: WIP(5): Introduce base PowerVM driver (ovs vif) https://review.openstack.org/422512
12:39:16 openstackgerrit Wang Qing wu proposed openstack/nova: WIP(5): Introduce base PowerVM driver (ovs vif) https://review.openstack.org/422512
12:48:38 openstackgerrit Alex Szarka proposed openstack/nova: Transform instance-live_migration_abort notification https://review.openstack.org/420453

Earlier   Later