Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-10-06
20:05:45 mtreinish before the osprofiler tests were added it wasn't an issue
20:05:51 jgwentworth right
20:06:15 jgwentworth oh well, sorry for the noise. I thought something had changed
20:06:54 mtreinish no worries, I'm just glad other people are looking at this stuff :)
20:08:04 mtreinish jgwentworth: fwiw, if you want to fix that output you could probably backport 501842
20:08:36 mtreinish or just rip the osprofiler test out on the stable branches, tbh I'm not sure what value it adds
20:10:03 jgwentworth yeah, I like running profiler last locally so tests I'm working on run and fail first without having to wait for profiler
20:10:08 openstackgerrit Matt Riedemann proposed openstack/nova stable/pike: Set regex flag on ostestr command for osprofiler tests https://review.openstack.org/510226
20:10:11 ronlund let's see
20:29:30 openstackgerrit Ildiko Vancsa proposed openstack/nova master: update live migration to use v3 cinder api https://review.openstack.org/463987
21:07:39 superdan ronlund: I think those two patches are good to go
21:08:06 superdan ronlund: cool if you want to wait for andreykurilin to run a test, but he confirmed on one of the earlier iterations, I have a repro test in there, etc
21:09:00 andreykurilin superdan: Does the new revision ready for testing? I can recheck the patch now
21:09:21 superdan andreykurilin: oh yeah, I poked you earlier but no response :)
21:09:32 andreykurilin oh.
21:09:38 superdan andreykurilin: would be awesome if you could, and maybe point me at the review so I can watch?
21:11:48 andreykurilin superdan: heh, sure. https://review.openstack.org/#/c/510144 gate-rally-dsvm-neutron-existing-users-rally job. grep tag-to-search to see debug messages which actually shows that everything works or not
21:13:23 superdan andreykurilin: doesn't it just fail if it fails?
21:15:27 superdan andreykurilin: I don't see that job in the first run of that patch...
21:16:14 ronlund ok, writing up a draft spec for discussion on this max_count rate limiting thing
21:16:21 andreykurilin superdan: no :( I tried to make this change as soon as possible, so i have not removed a hack to limit the loop. I'll do it now. the first results were not published, it looks like you posted a new revision before the whole jobs were finished
21:16:37 superdan ah okay
21:19:33 andreykurilin superdan: so originally the listing doesn't fail. it was just a inf loop. To add debug messages which will not flood the log file, I added a limit in the loop(10). That is why it stopped failing. Now I resubmitted a patch which fails in case of 10 iterations of the loop. so you do not need to check the logs
21:20:11 superdan andreykurilin: if len(log) == 10g -> fail? :)
21:20:12 superdan andreykurilin: awesome thanks
21:20:21 andreykurilin ha
21:29:00 openstackgerrit Matt Riedemann proposed openstack/nova-specs master: Limit instance create max_count (spec) https://review.openstack.org/510235
21:29:44 ronlund superdan: cfriesen: leakypipes: i know it's late so probably just ignore until next week, but ^ has some thoughts on the concurrent scheduling issue
21:30:10 superdan ack
21:31:11 mriedem now lemme take a looksie at these crazy patches
21:44:07 mriedem wow rally runs a lot of jobs
21:44:29 mriedem that's got to be a challenge to both tempest and trove
21:45:40 mtreinish mriedem: I think htat's more than tempest
21:56:10 cfriesen mriedem: did we ever fix the issues with multi-boot where scheduling more than min_count but less than max_count would result in instances in the "error" state?
21:58:40 mriedem cfriesen: i'd need more details than that
21:59:04 mriedem why would that put them in error state?
21:59:31 mriedem if i request min=5 max=10 but there are only 7 ports available in my port quota, the api changes max=7
21:59:50 mriedem you wrote that patch
22:02:44 jgwentworth zzzeek: are you around? I'm looking into something with oslo.db and was wondering, given a TransactionContextManager, how does it switch from reader to writer mode and vice versa between separate transactions?
22:07:37 mriedem superdan: done https://review.openstack.org/#/c/510203/
22:07:55 mriedem superdan: 2 issues, (1) the tests aren't using the args passed in and (2) apparently we have to handle boolean sort keys differently
22:08:14 mriedem superdan: presumably because not all db backends model booleans the same
22:08:17 mriedem some are bools, some are ints
22:08:31 cfriesen mriedem: I'm thinking about the case where we request min=5 and max=10 and the scheduler only finds hosts for 7 of them.
22:08:36 superdan mriedem: ah I didn't think we had any booleans, but I guess we do
22:09:14 superdan mriedem: hah, sorry, last minute cleanup on the tests and I didn't remember to plumb those through
22:09:38 melwitt zzzeek: it looks like it's controlled by the "independent" attribute, but I don't see that we use that
22:10:06 cfriesen mriedem: I think the ones without hosts might get left in the BUILD state
22:13:20 mriedem cfriesen: if you get fewer hosts than requested instances, it's NoValidHost https://github.com/openstack/nova/blob/master/nova/scheduler/filter_scheduler.py#L86
22:13:29 mriedem and conductor will put them all into ERROR state and into cell0 yes
22:13:30 cfriesen mreidem: yeah, was just looking at that
22:13:48 cfriesen mriedem: but that doesn't make sense if the scheduler was able to schedule at least min_count instances.
22:14:02 mriedem cfriesen: then don't request max?
22:14:08 mriedem but yeah that's weird
22:15:24 mriedem cfriesen: see check_num_instances_quota
22:15:34 mriedem if you have quota for max, then that's what we say to build
22:16:01 mriedem we don't pass anything to the scheduler about what minimum number should be built if possible
22:16:08 mriedem as far as i can tell
22:16:30 mriedem wonder how this behaved in ocata
22:16:36 cfriesen mriedem: if I remember right the quota part is okay, but the scheduler part is a bit wonky. we added a local patch to pass "min_num_instances" in the spec_obj
22:16:44 cfriesen mriedem: it's been broken since forever
22:17:20 mriedem did you try upstreaming that patch ever?
22:17:37 cfriesen mriedem: I think so, and it was suggested to just get rid of multiboot, but that got objections
22:17:52 cfriesen I can try reviving it. :)
22:18:01 cfriesen the patch, I mean
22:18:27 mriedem yeah so multicreate isn't going away
22:18:30 mriedem just like rebuild isn't going away
22:18:46 mriedem so yeah would probably be good to revive that
22:19:02 cfriesen would that count as an API change and need a spec?
22:19:27 cfriesen or is it incorrectly returning an error now and we can just correct it
22:23:09 zzzeek jgwentworth:. Can you email me at mike@zzzcomputing
22:23:31 zzzeek .com and I'll get back to you tomorrow? Not at a computer right now
22:24:10 melwitt zzzeek: okay, will do that if I can't figure this out. thanks
22:26:53 fried_rice Be there existing docs describing the resources:<resource_class>=<count> syntax in flavor extra specs?
22:30:01 cfriesen mriedem: hah, I thought I had a bug open: https://bugs.launchpad.net/nova/+bug/1458122
22:30:03 openstack Launchpad bug 1458122 in OpenStack Compute (nova) "nova shouldn't error if we can't schedule all of max_count instances at boot time" [Wishlist,Opinion] - Assigned to Chris Friesen (cbf123)
22:30:18 cfriesen there's also this one: https://bugs.launchpad.net/nova/+bug/1623809
22:30:19 openstack Launchpad bug 1623809 in OpenStack Compute (nova) "Quota exceeded when spawning instances in server group" [Wishlist,Opinion]
22:30:45 cfriesen that latter one is really about min_count/max_count and quota_server_group_members
22:31:06 openstackgerrit Dan Smith proposed openstack/nova master: Fix instance_get_by_sort_filters() for multiple sort keys https://review.openstack.org/510203
22:31:06 openstackgerrit Dan Smith proposed openstack/nova master: Always put 'uuid' into sort_keys for stable instance lists https://review.openstack.org/510140
22:31:10 mtreinish mriedem, jgwentworth: http://logs.openstack.org/26/510226/1/check/gate-nova-python27-ubuntu-xenial/939736e/testr_results.html.gz
22:31:39 superdan mriedem: good catches, thanks
22:39:49 mriedem YES!
22:40:04 mriedem i need to get out of here and get some pizza
22:40:08 mriedem and later some cookies maybe
22:40:27 mriedem i won't be 160 by christmas with pizza and cookies though
22:51:21 mriedem it's all that poutin you eat
22:51:34 mriedem i heard canadian babies are bottle fed poutin
22:51:41 mriedem aye
22:53:12 cfriesen mmmm...poutine.
22:53:30 cfriesen hard to find the good stuff though....lots of crappy versions
22:54:17 cfriesen hot crispy fries, squeeky cheese curds, and boiling hot gravy
22:55:15 cfriesen great...now I want poutine and I'm pretty sure there's going to be something healthy for supper.
22:55:45 mriedem superdan: ok comments in those 2 changes
22:56:04 mriedem but really leaving this time
22:56:44 superdan mriedem: just spiteful on that bug fix eh?
22:57:18 superdan I think doing pagination without a requested sort order is kinda reckless, but I'll capitulate
22:58:48 openstackgerrit Merged openstack/nova master: Note TrustedFilter deprecation in docs https://review.openstack.org/509931
22:59:58 openstackgerrit Dan Smith proposed openstack/nova master: Always put 'uuid' into sort_keys for stable instance lists https://review.openstack.org/510140

Earlier   Later