| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-07-17 | |||
| 13:03:00 | bauzas | Sara_: Nova doesn't support operators killing VMs directly | |
| 13:03:19 | bauzas | instead, you need to "nova delete" your instance | |
| 13:04:55 | openstackgerrit | Chris Dent proposed openstack/nova master: [placement] Update allocation-candidates.yaml for gabbi 1.35 https://review.openstack.org/484356 | |
| 13:05:04 | Sara_ | bauzas_: But what happens in the scenario where compute host is removed from the openstack setup and rejoined ? | |
| 13:05:21 | bauzas | Sara_: you can still delete the instances | |
| 13:05:31 | bauzas | Sara_: even if the compute is not there | |
| 13:06:05 | Sara_ | Does libvirt update the "undefine" event to nova? | |
| 13:06:14 | bauzas | Sara_: when the compute will be back, it will call the nova DB to know which instances are deleted, and then will destroy the libvirt VMs | |
| 13:07:07 | efried | Anyone savvy on nova microversion support timeframes? In particular, is pre-2.27 still supported? | |
| 13:07:08 | openstackgerrit | Gábor Antal proposed openstack/nova master: Transform HostAPI-set_enabled notifications https://review.openstack.org/482622 | |
| 13:08:28 | Sara_ | bauzas_: Does libvirt update the "undefine" event to nova? I want to get an update from libvirt to nova when the domain gets undefined..Is it supported in Mitaka? | |
| 13:09:12 | cdent | efried: the removal of nova-net kind of broke the microversion contract (which is effectively that they are around for-ever) | |
| 13:10:17 | efried | cdent Okay, so we actually need to handle https://review.openstack.org/#/c/482746/10/keystoneauth1/session.py@539 this? | |
| 13:11:58 | cdent | urgh. I don’t know, that’s a good question. Some combination of mordred, mriedem and sdague may have ideas | |
| 13:13:24 | openstackgerrit | Chris Dent proposed openstack/nova master: [placement] fix 500 error when allocating to bad class https://review.openstack.org/484162 | |
| 13:13:49 | sdague | efried: honestly, there probably should be special case code in there to handle the compute case | |
| 13:13:59 | sdague | if the point is to support what's out there | |
| 13:14:33 | efried | sdague And if I understand the docs right, we can just send both headers down and nova will pick the new one if it's post 2.27, yes? | |
| 13:15:28 | sdague | efried: yep | |
| 13:15:32 | efried | k | |
| 13:18:19 | openstackgerrit | Merged openstack/nova master: fix unshelve notification test instability https://review.openstack.org/483986 | |
| 13:24:36 | bauzas | stephenfin: not sure you're there, or if someone can answer me this question, but where can I find upstream docs about PCI passthru and SR-IOV ? | |
| 13:25:03 | jaypipes | cdent: nice! +2 from me. :) | |
| 13:25:28 | cdent | coo, thanks | |
| 13:28:00 | mordred | cdent: reading | |
| 13:28:46 | cdent | mordred: I think sdague pretty much nailed it: compute needs a special case. Unclear about other services. | |
| 13:29:01 | efried | cdent mordred Comments in the review. | |
| 13:29:12 | mordred | yah. agree - I don't think any other services made it out the door with that form, did they? | |
| 13:30:38 | bauzas | Sara_: like I said earlier, if you undefine a libvirt VM, Nova can just say that the instance is shutdown, because it can't find it | |
| 13:31:01 | bauzas | Sara_: there is no 'callback' from a specific virt driver for updating some internal Nova status about instances | |
| 13:31:14 | sdague | mordred: the only other one that might have was ironic | |
| 13:31:25 | bauzas | Sara_: that's just nova which periodically monitors host states wrt instances | |
| 13:31:36 | sdague | I think we got standardization before anything else adopted it | |
| 13:33:04 | bauzas | cdent: is there a dependent change for updating requirements to gabbi>=1.35 ? | |
| 13:33:16 | cdent | bauzas: it’s already merged | |
| 13:33:18 | bauzas | cdent: talking of https://review.openstack.org/#/c/484356/2/test-requirements.txt | |
| 13:33:45 | cdent | https://review.openstack.org/#/c/481665/ | |
| 13:34:22 | bauzas | cdent: oh ok, so I guess gabbi-1.35 supports legacy behaviour | |
| 13:34:55 | cdent | bauzas: yes, it follows semantic versioning | |
| 13:35:21 | bauzas | cdent: WAT? impossible :p | |
| 13:35:27 | cdent | ikr! | |
| 13:35:39 | bauzas | hehe | |
| 13:36:00 | bauzas | cdent: anyway, the gate works quite good today, so I guess you did your job correctly :p | |
| 13:37:12 | cdent | gabbi tests itself on gnocchi’s use of gabbi, I could probably add the same for nova’s use of gabbi too | |
| 13:40:48 | edleafe | Scheduler subteam meeting in 20 minutes in #openstack-meeting-alt | |
| 13:42:48 | efried | mordred Much as I hate to suggest further dirtying up sta, perhaps it should have legacy-microversion-header and legacy-microversion-header-before-version fields. | |
| 13:43:13 | mriedem | gmann: alex_xu: was anyone going to post a change for fixing the docs issues in https://review.openstack.org/#/c/467999/ ? if not i could work on that today | |
| 13:46:31 | efried | sdague mordred I can't even find anything about ironic that suggests microversion support *other* than via X-OpenStack-Ironic-API-Version | |
| 13:47:01 | efried | Was looking for a statement like this one that nova has: https://developer.openstack.org/api-ref/compute/#api-versions | |
| 13:47:04 | sdague | efried: oh, maybe they predated and never fixed it | |
| 13:51:43 | alex_xu | mriedem: here is https://review.openstack.org/483170 | |
| 13:51:53 | alex_xu | sorry, I forget to tell you the link | |
| 13:53:59 | mordred | efried: ugh | |
| 13:54:05 | efried | agree | |
| 13:55:35 | cdent | mordred, efried: sorry not sorry | |
| 13:56:08 | openstackgerrit | Steven Webster proposed openstack/nova master: Fix instance evacuation with PCI devices https://review.openstack.org/484381 | |
| 13:56:08 | efried | cdent I think you're off the hook - it was cmurphy brought it up. | |
| 13:56:12 | mordred | efried: another option is to not support legacy headers in the ksa microversion helper method, since sending this at all is a new ksa feature in the first place - and we can have python-novaclient and python-ironicclient also add the old headers if they want to | |
| 13:56:26 | cdent | efried: I was the person who proposed the new form | |
| 13:56:47 | efried | cdent Oh, the generic OpenStack-API-Version ? | |
| 13:56:48 | cdent | mordred: I’d like that, but is that fair? | |
| 13:56:50 | mordred | I'm not crazy about either option tbh - just bringing it up | |
| 13:56:52 | cdent | efried: yes\ | |
| 13:57:15 | mordred | well - my concern is that there's no way for the user to know then that get('/servers', microversion='2.13') isn't going to work | |
| 13:57:18 | efried | cdent Yeah, it's the better way of doing it. You just got there too late. | |
| 13:57:33 | jangutter | bauzas: do you still need docs about PCI passthrough and SR-IOV? | |
| 13:57:41 | efried | mordred Right, python-*client isn't necessarily the only API consumer, right? | |
| 13:57:41 | cdent | story of my life | |
| 13:57:47 | mordred | cdent: the original form is basically unworkable in general - so thank you very much for making it the change | |
| 13:58:00 | mordred | efried: exactly | |
| 13:58:11 | efried | mordred So we hardcode it in ksa... or make it part of sta. | |
| 13:58:31 | efried | Cause I really don't like that other option - making "consumers" know about this quirk. | |
| 13:59:32 | bauzas | jangutter: I'd appreciate, yes | |
| 13:59:42 | bauzas | jangutter: I mean, I can find other docs, and I know most of that | |
| 14:00:01 | bauzas | jangutter: but we had some upstream docs that give me now 404s and I'd like to make sure we still have those somewhere | |
| 14:00:38 | jangutter | bauzas: I always refer back to some very old IOMMU/VFIO/PCI-ASSIGN presentation waay back when. | |
| 14:01:15 | bauzas | I know we had some admin-guide doc | |
| 14:01:28 | bauzas | but those are giving me 404s because of the doc migration | |
| 14:01:29 | mordred | efried: yah. me either. although maybe the answer for today is to not add the helper method in this release since this is getting in to a kind-of ugly place and needs some thought/design | |
| 14:01:44 | jangutter | bauzas: aah, end-user stuff. | |
| 14:01:55 | edleafe | cdent: bauzas: scheduler meeting now running | |
| 14:02:28 | bauzas | edleafe: snap, caught by the police :p | |
| 14:02:53 | mordred | efried: or - we make sure that the "add microversion helper parameter" bit to the end of the stack so that if we can't all agree on the approach we don't block anything else | |
| 14:02:54 | jangutter | bauzas: https://wiki.openstack.org/wiki/SR-IOV-Passthrough-For-Networking <--- very stale. | |
| 14:03:38 | efried | mordred Probably at least that last thing. | |
| 14:03:44 | bauzas | asettle: hola, unsure if you can answer my question | |
| 14:04:02 | efried | mordred Cause I imagine it may take a little while to figure out all the special cases, whatever approach we land on. | |
| 14:04:35 | jangutter | bauzas: and the one for Ocata: https://docs.openstack.org/ocata/networking-guide/config-sriov.html <--- something like this, or do you want more low-level? | |
| 14:04:52 | bauzas | asettle: but I can see a 404 on http://docs.openstack.org/admin-guide/compute-pci-passthrough.html | |
| 14:04:58 | bauzas | jangutter: yeah, that will work out | |
| 14:05:02 | mordred | efried: yes - and I think we need to get a ksa release cut that hasall the other things this week so that it makes the cutoff and you can use it in nova | |
| 14:05:18 | bauzas | jangutter: it's not really for my knowledge, it's rather more about providing guidance to others | |
| 14:05:52 | efried | mordred Yes, and I'm a tad worried about that. Working hard on getting https://review.openstack.org/#/c/483604/ polished off, cause that affects the initial APIs of the new methods. | |
| 14:06:16 | efried | mordred Also concerned about getting that rename (min/max_version => min/max_endpoint_version) in place for same reason. | |
| 14:06:28 | mordred | efried: ++ | |
| 14:06:39 | efried | mordred Prolly should move this conversation to -dev/-sdks | |
| 14:12:33 | openstackgerrit | Jay Pipes proposed openstack/nova master: add dict of allocation requests to select_dests() https://review.openstack.org/483565 | |
| 14:12:33 | openstackgerrit | Jay Pipes proposed openstack/nova master: claim resources in placement API during schedule() https://review.openstack.org/483566 | |
| 14:12:41 | jaypipes | edleafe: done ^ | |
| 14:12:59 | edleafe | jaypipes: gracias | |