Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2018-04-19
13:57:20 efried jaypipes: If you want two granular groups to go into the same provider, you should combine them into a single numbered group.
13:57:21 bauzas efried: the second was 'in_tree'
13:57:37 bauzas efried: so I think it answers one of two
13:57:40 efried bauzas: We need to discuss that one further.
13:57:41 openstackgerrit Matt Riedemann proposed openstack/nova stable/queens: libvirt: slow live-migration to ensure network is ready https://review.openstack.org/557930
13:58:08 cdent mriedem: how do the functional tests become aware of that stuff? as in how does the new database string get connected to the code? that's the part I was struggling to make go. (simply overrdingn the default db connection string fell on its face)
13:58:15 bauzas efried: jaypipes: fine if I'm discussing that qpuery param by my own spec ?
13:58:26 bauzas efried: jaypipes: I tihnk we don't need a separate one
13:58:28 efried bauzas: As I stated somewhere or other, we can't use the keyword "in_tree" because that already has implications for GET /resource_providers to return *all* the providers in the same tree as the specified RP.
13:58:49 bauzas efried: I don't remember which exact param name I used but got it
13:58:53 mriedem cdent: i've always been confused by the framework for this, but you'd have to figure out how TestNovaMigrationsPostgreSQL works
13:58:55 efried bauzas: Also, it's not sufficient to say "same subtree" in any form, because we need an indication of where that subtree "starts".
13:59:09 bauzas efried: mmm k
13:59:10 mriedem the oslo.db opportunistic tests only run if a given backend is configured
13:59:27 efried bauzas: I did explain this further in one of the ML notes, but it was kinda buried. Let me find it...
13:59:31 cdent mriedem: i'll see if I can wrap my brain around that while vagueing out during the meeting
14:01:27 cdent wooph, some serious magic going on here
14:02:03 openstackgerrit Balazs Gibizer proposed openstack/nova-specs master: Suggest when to propose notification changes https://review.openstack.org/562265
14:03:07 efried bauzas: this one, scroll about 3/4 of the way down: http://lists.openstack.org/pipermail/openstack-dev/2018-April/129511.html
14:03:34 Nisha_Agarwal hi, I have stable pike setup where i was trying nova boot for ironic nodes. It always give "OrphanedObjectError" as soon as it spawns the thread for deploy. This happens due to the context getting set to None at some point. After debugging we found context is set to None here https://github.com/openstack/nova/blob/stable/pike/nova/objects/instance.py#L824
14:03:49 bauzas efried: sorry, was probably not enough caffeinated
14:03:54 bauzas or my French
14:04:09 bauzas -ETOOMUCHWORDS
14:04:10 Nisha_Agarwal When i just comment out this line ironic deploy goes through successfully,
14:04:32 Nisha_Agarwal i tried on two Pike setups, i see this behaviour consistently
14:04:33 efried Dis-moi si tu veux que je l'explique en français :)
14:05:53 Nisha_Agarwal should i go ahead and raise bug against nova?
14:07:29 Nisha_Agarwal mriedem, ^^^^
14:15:14 openstackgerrit Matt Riedemann proposed openstack/nova master: WIP: Add request_id to instance action notifications https://review.openstack.org/553288
14:15:59 sahid dansmith: arg... unsupported configuration: Unknown interface <driver name='vhost-user'>
14:16:38 mriedem Nisha_Agarwal: do you have the latest fixes for pike?
14:16:58 Nisha_Agarwal mriedem, yes
14:17:03 sahid it's not really like libvirt ignores the driver name... only way to make it working is to avoid passing name attribute
14:17:25 dansmith sahid: that's really unfortunate... I guess I should push to see more real testing...
14:17:32 Nisha_Agarwal i have pulled all the patches for stable/pike
14:17:50 sahid dansmith: nothing is merged
14:18:17 dansmith sahid: I know, I mean before +2ing.. I expected you were testing these other configs locally before review
14:18:30 sahid i expected to have that env with DPDK ready before the final +2+W
14:19:36 sahid i'm going to fix the issue in libvirt but for nova i guess we should just avoid using name attribute for vhostuser interface
14:24:26 dansmith sean-k-mooney[m]: are there jobs anywhere in neutron or experimental that use vhost/vhostuser that we can depends-on changes like this to get at least a smoke test?
14:28:58 jaypipes sahid: quick update needed on https://review.openstack.org/#/c/562565/ commit message and then I'm +2 on the whole series.
14:30:54 sahid jaypipes: cool, i will fix that, thanks
14:30:57 Nisha_Agarwal mriedem, then i will go ahhead n raise the defect?
14:32:01 mriedem Nisha_Agarwal: i'm in a meeting so don't have time to investigate this, nor the ironic background, so for tracking it yes if you want to report a bug that's fine
14:32:37 openstackgerrit Merged openstack/nova stable/ocata: Avoid unnecessary lazy-loads in mutated_migration_context https://review.openstack.org/515581
14:32:45 openstackgerrit Merged openstack/nova stable/ocata: Only attempt a rebuild claim for an evacuation to a new host https://review.openstack.org/550560
14:49:05 openstackgerrit zhangyangyang proposed openstack/nova master: Deprecate auth_uri option https://review.openstack.org/562537
15:00:47 stephenfin jaypipes, bauzas: Tell me more about https://review.openstack.org/#/c/557065/2/specs/rocky/approved/vgpu-rocky.rst
15:01:13 stephenfin jaypipes, bauzas: That's a very similar issue we have for NUMA-aware vSwitch
15:01:26 bauzas it's about configuring things
15:01:43 stephenfin I've been trying to get dynamic '[physnet_$PHYSNET_NAME]' groups created on the fly but oslo.config isn't designed that way
15:01:57 bauzas stephenfin: yeah I remember your spec
15:02:19 stephenfin So what I've done is ugly enough to not even push to Gerrit, heh https://github.com/stephenfin/nova/commit/b2c0421e64d627aaad2cf9ac71c6e1c79357a970
15:02:23 bauzas looks like we're limited for describing resources or topologies
15:02:51 bauzas stephenfin: the problem is that oslo.config is pretty freeform
15:03:06 bauzas stephenfin: and IMHO the pci whitelist is a terrible interface
15:03:50 stephenfin Right, well I'd be interested in figuring out what you do. I don't think what I have is good enough so I'm open to ideas
15:04:09 stephenfin Until then, I'll probably revert back to a JSON option just so I can proceed
15:05:08 jaypipes stephenfin: dansmith is gonna hate that.
15:05:16 mriedem stephenfin: as in StrOpt with serialized json?
15:05:18 mriedem i also hate that
15:05:20 takashin python-novaclient stable maintainers for pike, would you review https://review.openstack.org/#/c/562500/ ? This patch fixes pike gate job failure.
15:05:25 mriedem i would rather us load in a yaml file
15:05:28 stephenfin jaypipes: Oh, I'm not planning to keep it
15:05:31 mriedem if that's the alternative
15:05:51 melwitt takashin: I think you need to backport that to queens first, then to pike
15:06:06 stephenfin It just seems like we don't have a decided-upon alternative and I need to load _something_
15:06:11 mriedem takashin: done
15:06:13 dansmith stephenfin: I thought I suggested the cinder model for encoding single-level dicts in config with sections?
15:06:21 mriedem melwitt: fix came from queens
15:06:21 dansmith it's stupid simple
15:06:32 takashin melwitt: it has already merged in queens.
15:06:37 stephenfin dansmith: Yeah, it seems horrible though -> https://github.com/stephenfin/nova/commit/b2c0421e64d627aaad2cf9ac71c6e1c79357a970
15:06:44 melwitt mriedem, takashin: oh, okay. my bad
15:07:05 takashin mriedem: Thank you.
15:07:12 stephenfin dansmith: With cinder, they still know the group names ahead of time. We're having to build that based on another value
15:07:21 stephenfin It works, but it seem...wrong (IMO, anyway)
15:07:42 dansmith stephenfin: eh? they let you set your own section names.. backend_foo and what not
15:08:42 bauzas dansmith: I don't have the context, what were you suggesting and do you think it could apply for https://review.openstack.org/#/c/557065/ ?
15:09:18 dansmith bauzas: stephenfin: https://wiki.openstack.org/wiki/Cinder-multi-backend
15:09:24 mriedem bauzas: http://logs.openstack.org/54/554154/27/check/tempest-full/fbfe7c5/controller/logs/etc/cinder/cinder_conf.txt.gz
15:09:43 mriedem enabled_backends = lvmdriver-1
15:09:47 mriedem [lvmdriver-1]
15:09:51 dansmith exactly that
15:10:03 bauzas I see
15:10:13 bauzas so dynamically importing opt groups ?
15:10:16 mriedem http://logs.openstack.org/54/554154/27/check/legacy-tempest-dsvm-full-devstack-plugin-ceph/b88aa0b/logs/etc/cinder/cinder.conf.txt.gz for ceph
15:10:19 stephenfin dansmith: I need to figure out the heck that is working so. All I could find was this https://github.com/openstack/cinder/blob/master/cinder/volume/configuration.py
15:10:27 stephenfin and all calls to that were statically defined
15:10:44 bauzas mriedem: dansmith: okay, I need to consider that for https://review.openstack.org/#/c/557065/
15:10:47 stephenfin dansmith, mriedem, bauzas: but yeah, my configuration was looking like this http://paste.openstack.org/show/719567/
15:11:09 stephenfin think that was what was proposed in the review
15:11:25 dansmith stephenfin: right I think that's highly desirable over a separate yaml or json blob unless there's a really good reason not to
15:11:34 stephenfin dansmith: I concur
15:12:19 mriedem you could always....ask someone in -cinder about how that impl works
15:12:26 dansmith WAT
15:12:29 dansmith can you do that?
15:12:34 mriedem or if it can be shared into oslo.config so we can re-use it
15:12:39 mriedem remember oslo-incubator?
15:12:44 stephenfin but what if they're mean to me?

Earlier   Later