Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-01-08
16:23:23 lyarwood mdbooth: not that they are tested anywhere.
16:23:30 mdbooth lyarwood: Ok, that's probably a bug too, then
16:23:52 mdbooth So it's a 3 for 1 fix
16:24:02 openstackgerrit Chris Dent proposed openstack/nova master: Optional separate database for placement API https://review.openstack.org/362766
16:24:22 lyarwood mdbooth: right, I'll respin without moving _connect_volume out of _get_guest_xml
16:24:35 mdbooth I don't think you can do that
16:24:54 lyarwood right, context
16:25:02 mdbooth Actually it's already there
16:25:19 mdbooth But you'd have to remove the encryptor stuff from _create_domain_and_network
16:25:28 mdbooth Which is just backwards
16:25:32 mriedem matrohon: no
16:25:35 mdbooth Might as well go the other way
16:25:37 mriedem bug fix is fine
16:25:50 matrohon mriedem: ack, thanks
16:25:58 mriedem matrohon: but i looked at the patch and it doesn't really make sense to me, unless a port with ip_allocation='none' results in port.get('ip_allocation') == None?
16:27:12 efried jaypipes cdent What is the behavior when I try to delete a provider that has "stuff" associated with it?
16:27:34 efried I think I've seen some cases where we automatically delete the "stuff" and some where we bounce.
16:27:44 cdent fails, if there are allocations we bounce
16:27:48 matrohon mriedem: port.get('ip_allocation') == None is the unexpected result, I'm currently trying to find the reason why. I probably need to fix that first on the neutron side
16:27:55 cdent I think that's the only one, but let confirm
16:28:28 efried cdent I'm talking about inventory, aggregates, traits - but most especially aggregates at the moment.
16:29:09 jaypipes efried: we just delete.
16:29:27 jaypipes efried: allocations are the only thing that prevents deletion (and child providers)
16:29:28 mriedem matrohon: if neutron needs changes to support this case, then things get sticky with regards to interoperability
16:29:39 cdent efried: https://github.com/openstack/nova/blob/master/nova/objects/resource_provider.py#L816
16:29:44 efried jaypipes Okay. So this is tricky.
16:29:55 mriedem because if you upgrade nova to handle this none scenario but neutron doesn't support it, things fail, and vice-versa
16:30:22 efried cdent Thanks, it's all right there!
16:31:36 edleafe mriedem: just getting caught up (sitting in traffic is *so* productive!)
16:31:43 edleafe mriedem: your changes look fine
16:32:43 edleafe mriedem: some of the issues you asked about on the review were from the original test, which was supposed to test alternate host usage; hence the 4 computes and such
16:33:02 matrohon mriedem: but ip_allocation
16:33:04 edleafe I'm going to add that test to the resize patch
16:33:09 matrohon mriedem: sorry
16:33:17 efried cdent jaypipes So this might be okay. A virt that's "leaving" an aggregate should only delete the aggregate association from self, not delete it from the aggregate or delete the shared provider. The sharing provider will still show up in the ProviderTree, so report client won't delete it. Then the next time we pull the compute RP and its associations, that sharing provider will be absent (because no longer associated by a
16:33:17 efried ggregate) so we'll... drop it from the local provider tree at that time, somehow.
16:33:22 mriedem edleafe: add the test to the resize patch?
16:33:31 mriedem edleafe: the test is already there since it comes before it in the series
16:33:51 edleafe mriedem: the regression test really doesn
16:34:06 edleafe doesn't check that the alternate hosts stuff is working as expected
16:34:16 edleafe just that it doesn't blow up :)
16:34:23 cdent efried: I think that sounds right, but as your patient, I'm going to the bath, will bbl
16:34:28 mriedem so you're going to modify the regresssion test to add to it in the resize patch
16:34:42 efried cdent Enjoy
16:35:29 edleafe mriedem: no, more or less copy parts to make a new functional test that's alternate host-specific
16:35:30 ktibi dansmith Hi do you know why consecutive_build_service_disable_threshold default value is 10 ?
16:35:36 openstackgerrit Matt Riedemann proposed openstack/nova master: Add pagination and Changes-since filter support for os-migrations. https://review.openstack.org/330406
16:35:47 ktibi because I found a method for disable all compute service with a simple user :/
16:35:57 matrohon mriedem: I'll keep on digging on the neutron side first, since not having the ip_allocation field set in port's attributes retrieved by a REST request, when it is explicitly set in the database is not really clean
16:36:03 dansmith ktibi: yes, because I set it to 10 after discussion in the room with a bunch of operators
16:36:16 matrohon mriedem: don't you think?
16:36:45 mriedem matrohon: i agree that None != unset and there are differences in client side behavior potentially, yes
16:36:46 dansmith ktibi: what's the method? might be something we need to exclude from the disable logic if it's not really a fault
16:36:56 mriedem matrohon: but i don't really know how that's all used on the neutron side
16:37:04 ktibi ok, 10 is good but for now, you can fool nova for disable all compute
16:37:29 openstackgerrit Matthew Booth proposed openstack/nova master: Don't connect volumes in _get_guest_storage_config https://review.openstack.org/531852
16:37:32 jaypipes efried: yeah, that sounds about right to me, too.
16:37:36 mdbooth lyarwood: ^^^
16:37:50 efried jaypipes Dangit, that still leaves us with orphan-able shared providers.
16:37:54 mdbooth lyarwood: Lets see how that does against tempest, but that change would be extremely backportable
16:38:04 mriedem ktibi: perhaps you'd like to share how users can do that
16:38:05 efried (aside from being asymmetrical)
16:38:09 ktibi dansmith if you spawn multi instances in a network with 0 free IP, nova fail to spawn and increase the trigger. if you deploy ~60 instances, you disable 6 computes
16:38:10 dansmith ktibi: what's the method?
16:38:17 matrohon mriedem: I'll have a look, but AFAIK, routed networks is the only feature really using the ip_allocation field for now
16:38:53 ktibi one user disable my 15 computes with spawn 200 instance on a /28 private network
16:39:08 jaypipes efried: don't worry about that too much.
16:39:09 dansmith ktibi: if you only have 6 computes or are packing tight I guess, but remember it has to be consecutive, so a cloud with reasonable activity will not suffer from that
16:39:20 jaypipes efried: let's focus on the nested providers handling.
16:39:26 dansmith ktibi: but if there's something we can catch about that specific case and not count it in the total, that would make sense
16:39:27 openstackgerrit Sylvain Bauza proposed openstack/nova master: WIP: libvirt: Use only existing mdevs if kernel race https://review.openstack.org/531853
16:39:38 jaypipes efried: orphaned stuff can always be cleaned up by an external perioidic.
16:39:55 mdbooth lyarwood: The importance of ^^^ is that the above is where we currently do connect_volume/attach_encryptors separately somewhat deliberately. We fix that, then everything remaining is a bug.
16:40:03 efried jaypipes Ick. But okay. Guess that's not the worst thing in the world.
16:40:09 mriedem ktibi: what is the error you get in the compute? PortLimitExceeded?
16:40:38 mriedem the api should be checking the port quota based on the number of instances and selected network before doing scheduling
16:40:41 bauzas dansmith: thanks for the reviews :)
16:40:46 mriedem which is racy, but still it's a check that happens
16:40:57 bauzas dansmith: I just added a new WIP change that will see the direction for the kernel race
16:40:58 ktibi mriedem ailed to allocate the network(s), not rescheduling
16:41:15 mriedem ktibi: yeah, but wondering what triggers that, a specific exception
16:41:18 ktibi not quota check :/ just no more IP free in the pool
16:41:20 mriedem do you have a stacktrace you can put in a paste?
16:41:26 ktibi yep wait plz
16:41:31 openstackgerrit Stephen Finucane proposed openstack/nova master: Fix typo in release note https://review.openstack.org/531854
16:41:31 mriedem NoMoreFixedIps ?
16:41:38 ktibi yes
16:41:42 bauzas dansmith: I'll also add a follow-up that will try to find the existing mdev 'allocations' (ie. which mdev is used by which instance) at init_host()
16:41:54 stephenfin jaypipes, mriedem: Could one of you push that reno typo fix through? https://review.openstack.org/531854
16:42:13 mriedem ktibi: can you report a bug? maybe we can build a whitelist of exception types that shouldn't be used to auto-disable a compute
16:42:23 dansmith bauzas: not sure what kernel race you mean, but cool
16:42:26 bauzas fun fact, I also discovered that the nvidia driver can't support the same physical GPU having more than 1 allocated mdev per guest
16:42:29 jaypipes stephenfin: done
16:42:30 lyarwood mdbooth: back sorry
16:42:37 stephenfin jaypipes: Cheers :)
16:42:40 jaypipes np
16:43:07 bauzas dansmith: the fact that when calling sysfs, libvirtd doesn't get the update so it won't change the values of availableInstances and won't show up the newly created mdecs
16:43:37 lyarwood mdbooth: that's going to break iSCSI volumes
16:43:42 ktibi mriedem https://thepasteb.in/p/zmh8qDG2ZYJIZ
16:43:48 mdbooth lyarwood: How?

Earlier   Later