| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2018-04-13 | |||
| 18:16:10 | superdan | can I or will I? | |
| 18:16:17 | mriedem | can you will it | |
| 18:21:13 | kashyap | superdan: I know you did a more robust review, I hope my -1 here is still in the spirit of it: https://review.openstack.org/#/c/484997/16/nova/conf/libvirt.py | |
| 18:24:54 | openstackgerrit | Chris Dent proposed openstack/nova master: [placement] Parse forbidden traits in query strings https://review.openstack.org/556819 | |
| 18:24:55 | openstackgerrit | Chris Dent proposed openstack/nova master: [placement] Filter resource providers by forbidden traits in db https://review.openstack.org/556472 | |
| 18:24:56 | openstackgerrit | Chris Dent proposed openstack/nova master: [placement] Filter allocation candidates by forbidden traits in db https://review.openstack.org/556660 | |
| 18:24:57 | openstackgerrit | Chris Dent proposed openstack/nova master: [placement] Support forbidden traits in API https://review.openstack.org/556820 | |
| 18:28:44 | mriedem | jgwentworth: mlavalle: posted http://forumtopics.openstack.org/cfp/details/124 | |
| 18:29:15 | jgwentworth | coolness. thanks. superdan, want to post the cells one? | |
| 18:29:49 | mlavalle | mriedem: thanks | |
| 18:30:33 | openstackgerrit | Hongbin Lu proposed openstack/nova stable/pike: Skip placement on rebuild in same host https://review.openstack.org/561015 | |
| 18:31:25 | mriedem | mlavalle: ops might want to ask about routed networsk too, i don't know how much you get for questions about that | |
| 18:32:17 | mlavalle | mriedem: I am not getting many questions lately | |
| 18:38:09 | superdan | jgwentworth: okay | |
| 18:45:51 | jgwentworth | thankee | |
| 18:46:25 | superdan | jgwentworth: http://forumtopics.openstack.org/cfp/details/125 | |
| 18:46:42 | jgwentworth | woot | |
| 18:49:13 | cfriesen | got a weird setup. we had in instance on a host in Newton, then deleted the host, then upgraded to Pike (which added uuids to the Service records but apparently not deleted ones), then created the compute node again with the same name. Now when we try to get that instance from the db it chokes, I suspect because it pulls in the deleted host record which has no uuid. | |
| 18:49:34 | cfriesen | should we have applied UUIDs to deleted hosts as well in the Pike online data migration? | |
| 18:49:39 | jgwentworth | there's a bug for that | |
| 18:50:58 | jgwentworth | cfriesen: https://bugs.launchpad.net/nova/+bug/1746509 | |
| 18:50:58 | openstack | Launchpad bug 1746509 in OpenStack Compute (nova) pike "TypeError: Can't upgrade a READER transaction to a WRITER mid-transaction" [High,Fix committed] - Assigned to melanie witt (melwitt) | |
| 18:51:36 | jgwentworth | did you do your pike upgrade before that merged or was released? | |
| 18:51:46 | openstackgerrit | Merged openstack/nova master: Add host field to InstanceActionEvent https://review.openstack.org/555146 | |
| 18:53:13 | cfriesen | jgwentworth: cool, looks interesting. I need to dig into this a bit more to get the details. | |
| 18:54:09 | jgwentworth | k. yeah, this was specifically having to do with an upgrade scenario where there were already existing service records with no UUID | |
| 18:54:22 | jgwentworth | and instance list failed | |
| 18:54:35 | cfriesen | jgwentworth: do you know why we didn't just apply uuids to deleted services for consistency? | |
| 18:55:26 | jgwentworth | cfriesen: we were but on-the-fly upon access. so the bug unfortunately was shown *during* the generation of service UUID for one that did not yet have a UUID :/ | |
| 18:55:35 | cfriesen | ah, got it. | |
| 18:56:02 | jgwentworth | because of nested database transaction contexts. unrelated to the generation but nothing hit the path until the generation was added to the instance read flow | |
| 18:56:38 | cfriesen | jg: I was thinking in service_uuids_online_data_migration() | |
| 18:56:44 | cfriesen | jgwentworth: ^ | |
| 18:57:22 | jgwentworth | yeah, I figured that's what you meant. was not done in a batch like that and I don't know why not | |
| 18:57:49 | jgwentworth | superdan might know. why we didn't do service UUID as a batch online data migration? | |
| 18:59:14 | superdan | did we not? | |
| 18:59:35 | superdan | we did | |
| 18:59:47 | superdan | we normally skip deleted records for expediency, which is what he's asking right? | |
| 19:00:17 | jgwentworth | hm, okay. so that means they didn't run nova-manage online_data_migrations? | |
| 19:00:28 | jgwentworth | no, not deleted records | |
| 19:00:38 | cfriesen | jgwentworth: pretty sure we ran it but it would have skipped the deleted records | |
| 19:00:41 | superdan | [11:54:35] <cfriesen>jgwentworth: do you know why we didn't just apply uuids to deleted services for consistency? | |
| 19:00:44 | jgwentworth | existing non-deleted service records that lacked a service UUID | |
| 19:00:50 | jgwentworth | oh, I missed that | |
| 19:01:09 | cfriesen | jgwentworth: we'll try that fix you pointed me at...hopefully that solves the on-demand issue | |
| 19:01:16 | superdan | cfriesen: for instance-related things, skipping deleted saves us a ton of work of course | |
| 19:01:31 | superdan | maybe less on services, but .. if there's no way to undelete a service there's not much point | |
| 19:01:42 | jgwentworth | okay, yeah if they service was in the deleted state then it could be skipped. but how could an active instance have a deleted service record? | |
| 19:01:46 | cfriesen | superdan: yeah, I get that. just bad luck that it triggered another bug | |
| 19:02:03 | jgwentworth | I guess that goes back to our discussion from the other day | |
| 19:02:09 | superdan | jgwentworth: we just went over that you can delete services even when instances are on it right? :) | |
| 19:02:10 | superdan | yeah that | |
| 19:02:10 | jgwentworth | we allow service delete while instances are alive | |
| 19:02:16 | cfriesen | jgwentworth: boot instance, migrate it, delete service, upgrade, add new service with same name, migrate instance | |
| 19:02:33 | superdan | or just boot instance, delete service | |
| 19:02:38 | jgwentworth | okay, I didn't realize that was a thing | |
| 19:03:02 | superdan | jgwentworth: remember me saying I thought it was probably bad to allow this anyway? :P | |
| 19:03:04 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Make service all-cells min version helper use scatter-gather https://review.openstack.org/539590 | |
| 19:03:05 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Make host_manager use scatter-gather and ignore down cells https://review.openstack.org/539617 | |
| 19:03:06 | jgwentworth | cfriesen: so you would have a problem if we made it so services can't be deleted if they have running instances? | |
| 19:03:17 | mriedem | superdan: i've got a patch up for that, testing it in devstack atm | |
| 19:03:21 | jgwentworth | superdan: yes | |
| 19:03:24 | mriedem | https://review.openstack.org/#/c/560674/1 | |
| 19:04:12 | cfriesen | jgwentworth: I wouldn't, no. In our case we did the sequence I described...the issue is that the back reference from instance to service is by host name | |
| 19:04:29 | cfriesen | so if you delete a service then make another with the same name, they'll both get matched | |
| 19:04:57 | superdan | which is another reason why that hostname-based keying is kinda bad | |
| 19:05:36 | jgwentworth | oh, okay, so because you migrated the instance already the service delete would be okay with you as no running instances left on it | |
| 19:05:45 | cfriesen | yep | |
| 19:05:54 | jgwentworth | k, cool | |
| 19:29:19 | jgwentworth | superdan: on this patch about multi-cell query for instance group members, mriedem and I had some discussion over whether or not I should make changes to limit the query to the cell we're already in, if it's a move operation. we wanted to get your opinion if there's some reason not to do that, that we might have missed https://review.openstack.org/#/c/540258 | |
| 19:29:23 | mriedem | nice http://paste.openstack.org/show/719188/ shows that n-api doesn't barf when trying to delete allocations during server delete if the n-cpu is down and [placement] isn't configured | |
| 19:42:14 | imacdonn | mriedem: FYI, in case you run into it somewhere .... after re-enabling UCA for Oracle ZFSSA CI, I hit this - https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=891315 - 'apt-get upgrade' installed a 'libnettle6 amd64 3.4-1~cloud0' (from the UCA), and i'm hoping that'll fix it | |
| 19:42:15 | openstack | Debian bug 891315 in dnsmasq "[dnsmasq] libnettle6 dependency" [Normal,Fixed] | |
| 19:43:24 | mriedem | jgwentworth: https://review.openstack.org/#/c/557796/ lgtm, i'll +2 once you fix that typo | |
| 19:44:30 | jgwentworth | mriedem: thanks | |
| 19:44:33 | mriedem | imacdonn: which UCA? | |
| 19:44:35 | mriedem | pike or queens? | |
| 19:44:48 | imacdonn | mriedem: queens ... it's devstack master | |
| 19:45:01 | mriedem | imacdonn: haven't had any issues like that here https://review.openstack.org/#/c/554317/ | |
| 19:45:20 | imacdonn | mriedem: perhaps you got the new libnettle6 by some other means | |
| 19:45:30 | mriedem | idk, infra builds the images | |
| 19:45:55 | imacdonn | mriedem: OK, well, just a heads-up in case you see it | |
| 19:46:04 | mriedem | ack | |
| 19:46:29 | imacdonn | other than that, ZFSSA CI seems to work (with multiattach) with the UCA | |
| 19:46:47 | mriedem | outstanding | |
| 19:47:55 | superdan | jgwentworth: sure seems reasonable | |
| 19:48:12 | jgwentworth | k, thanks | |
| 19:49:47 | openstackgerrit | melanie witt proposed openstack/nova-specs master: Document Rocky review priorities https://review.openstack.org/557796 | |
| 19:56:34 | mriedem | any opinions on removing the redundant unit tests from tssurya's patch here? https://review.openstack.org/#/c/554920/10/nova/tests/unit/compute/test_host_api.py@439 the functional test in that change will now cover that | |
| 19:57:54 | openstackgerrit | Jay Pipes proposed openstack/nova master: support multiple member_of qparams for GET /r-ps https://review.openstack.org/561315 | |
| 19:58:34 | leakypipes | superdan: that should help when we need >1 placement req filters... | |
| 20:00:24 | openstackgerrit | Jay Pipes proposed openstack/nova master: support multiple member_of qparams for GET /r-ps https://review.openstack.org/561315 | |
| 20:02:31 | openstackgerrit | Jay Pipes proposed openstack/nova master: support multiple member_of qparams for GET /r-ps https://review.openstack.org/561315 | |
| 20:08:41 | jgwentworth | mriedem: yeah, seems like that unit test can go | |
| 20:09:05 | jgwentworth | I don't see it covering anything additional unless I'm missing something | |
| 20:09:28 | superdan | leakypipes: that's implementing what we said int the spec revision right? | |
| 20:10:33 | leakypipes | superdan: yessir | |
| 20:10:45 | superdan | cools | |
| 20:16:10 | jgwentworth | wow, assert_called_once() is a thing. *mind blown* | |
| 20:42:44 | mriedem | using fixtures as context managers - put that in your bag of tricks too for when a test needs to conditionally use a fixture | |