| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-08-21 | |||
| 13:36:19 | openstackgerrit | Ed Leafe proposed openstack/nova master: WIP - add alternate hosts https://review.openstack.org/486215 | |
| 13:36:20 | openstackgerrit | Ed Leafe proposed openstack/nova master: WIP - Add allocations to the values returned from the scheduler https://review.openstack.org/495854 | |
| 13:36:20 | openstackgerrit | Ed Leafe proposed openstack/nova master: WIP - return alternates along with their allocations https://review.openstack.org/486253 | |
| 13:39:17 | gibi | alex_xu: no, it is not a bug. It is the way how the resouce claiming was implemented in the scheduler for the move operations | |
| 13:39:36 | gibi | alex_xu: there was couple of irc discussion about it in the past | |
| 13:40:16 | gibi | alex_xu: here is some irc history http://p.anticdent.org/logs/openstack-nova?dated=2017-08-08%2012:29:21.129291#15Hv | |
| 13:40:54 | stephenfin | cdent: Done | |
| 13:40:59 | stephenfin | https://review.openstack.org/#/c/491529/, that is | |
| 13:41:10 | cdent | thanks stephenfin welcome back | |
| 13:41:17 | stephenfin | cdent: Cheers :) | |
| 13:42:19 | cdent | I’m trying to sit outside in the sun but it may be too squinty | |
| 13:43:18 | stephenfin | The rain is bouncing off the pavement outside. I'm not at all jealous :D | |
| 13:43:31 | edleafe | cdent: I gotta make my eclipse viewing box this morning | |
| 13:43:32 | gibi | alex_xu, stephenfin: as you are already looking at functional tests and scheduling, there are two additional test patch that needs core eyes | |
| 13:43:38 | gibi | alex_xu, stephenfin https://review.openstack.org/#/c/495159/ | |
| 13:43:47 | gibi | alex_xu, stephenfin: https://review.openstack.org/#/c/493865/ | |
| 13:43:58 | cdent | edleafe: are you in the path? I thought you were too far south | |
| 13:44:08 | edleafe | Partial | |
| 13:45:09 | cdent | like your soul | |
| 13:45:33 | dansmith | I'm at 99.6% | |
| 13:45:42 | dansmith | don't expect me around much until after :) | |
| 13:46:08 | cdent | dansmith: are you overrun nearby by tourists and such? | |
| 13:46:38 | dansmith | cdent: no, I live in a heavily fortified mountaintop stronghold remember? | |
| 13:46:58 | cdent | dansmith: understood, but I wondered if there might be people at the walls | |
| 13:46:59 | edleafe | cdent: this is as good as it will get for me: http://imgur.com/a/z6FfF | |
| 13:47:01 | dansmith | but in reality, so far the traffic apocolypse hasn't been a thing | |
| 13:47:08 | dansmith | cdent: heh, not yet :) | |
| 13:47:33 | edleafe | dansmith: my sister-in-law lives in Albany. She says traffic has been hell there for the past week | |
| 13:48:16 | dansmith | edleafe: yeah, I think the places without road capacity are feeling it, but the bigger places haven't had the troubles they expected | |
| 13:48:38 | dansmith | we were over in eastern/central this weekend and it was a ghost town | |
| 13:48:53 | dansmith | gas stations have been stockpiling toilet paper and bullets for weeks and it was all for naught | |
| 13:49:11 | edleafe | heh | |
| 13:49:31 | edleafe | Well, for once Albany has something to attract tourists | |
| 13:49:43 | dansmith | heh | |
| 13:55:05 | edleafe | Scheduler subteam meeting in 5 minutes in #openstack-meeting-alt | |
| 14:07:37 | mriedem | hmm, so i heard stable gate sucked last week | |
| 14:07:39 | mriedem | and continues to suck | |
| 14:07:53 | mriedem | looking at a failure, the ComputeFilter kicked something out because the single-node host job has 4 VMs on it already | |
| 14:08:12 | mriedem | vcpus=8,vcpus_used=4 | |
| 14:08:29 | mriedem | ^ is the compute node stats before the NoValidHost | |
| 14:12:23 | mriedem | seems odd given there is capacity left on that node and it's also weird that we have CoreFilter enabled in devstack since it was disabled by default in code | |
| 14:14:22 | alex_xu | gibi: thanks, i will read the irc log | |
| 14:14:52 | mriedem | https://goo.gl/g7LGmM | |
| 14:15:50 | cdent | what manner of malware are you sending us to with that link mriedem ? | |
| 14:15:56 | mriedem | logstash | |
| 14:16:37 | cdent | mriedem: you’re being semi-ignored becasue I think anyone who cares is over in the scheduler meeting at the mo | |
| 14:16:45 | mriedem | i know | |
| 14:18:41 | alex_xu | mriedem: fyi, a regression in python-novaclient https://review.openstack.org/#/c/492003/, not sure whether it is worth a release | |
| 14:19:46 | mriedem | alex_xu: so it's been broken since 9.0.0 | |
| 14:19:52 | mriedem | if https://review.openstack.org/#/c/429512/ caused it | |
| 14:20:03 | alex_xu | yea, probably for a while | |
| 14:20:27 | mriedem | then we'll probably just fix in a patch release once stable/pike is ok for python-novaclient releases | |
| 14:20:54 | sean-k-mooney | mriedem: it is requsting 8 vcpus it should fit but it would be oversubscribing though not against itself | |
| 14:21:17 | mriedem | sean-k-mooney: it's requesting 1 vcpu | |
| 14:21:19 | alex_xu | ok | |
| 14:21:19 | sean-k-mooney | mriedem: i assume that the 4 cores already in use are not dedicated? | |
| 14:21:35 | mriedem | sean-k-mooney: we don't have cpu pinning in the upstream infra gate jobs no | |
| 14:21:53 | sean-k-mooney | mriedem: ah sorry missread http://logs.openstack.org/28/468528/2/check/gate-tempest-dsvm-neutron-full-ubuntu-xenial/337af98/logs/screen-n-sch.txt#_2017-08-21_04_15_14_765 | |
| 14:22:07 | sean-k-mooney | {"cores": 1, "cells": 1, "threads": 1, "sockets": 8} | |
| 14:22:11 | sean-k-mooney | that is for the host | |
| 14:22:16 | sean-k-mooney | not the vm right | |
| 14:22:20 | mriedem | correct | |
| 14:25:19 | mriedem | looking at the CoreFilter code this doesn't make any sense | |
| 14:26:03 | mriedem | we have 8 total, and cpu_allocation_ratio is 16.0, so we end up with vcpus_total = 24, 24-vcpus_used (4) = 20, so we have 20 free vcpus and we're requesting 1 | |
| 14:27:32 | cdent | mriedem: is that a new vm or move/migration/whatever? | |
| 14:27:37 | mriedem | new vm | |
| 14:28:03 | sean-k-mooney | mriedem: well you should be able to use 128 vcps with 8 cpus and an allocation ratios of 16. anyway did you notice this error 'disabled_reason': u'AUTO: Connection to libvirt lost: | |
| 14:28:42 | sean-k-mooney | at http://logs.openstack.org/28/468528/2/check/gate-tempest-dsvm-neutron-full-ubuntu-xenial/337af98/logs/screen-n-sch.txt#_2017-08-21_04_15_14_766 | |
| 14:29:31 | mriedem | hmm seeing that now yeah | |
| 14:30:40 | mriedem | oh duh, | |
| 14:30:48 | mriedem | it's not the CoreFilter, it's the ComputeFilter | |
| 14:30:50 | mriedem | god | |
| 14:31:19 | sean-k-mooney | there are a load of tracebacks in the n-cpu log | |
| 14:31:46 | sean-k-mooney | NeutronAdminCredentialConfigurationInvalid: Networking client is experiencing an unauthorized exception. | |
| 14:31:52 | openstackgerrit | Balazs Gibizer proposed openstack/nova master: WIP: test allocation handling during scheduler retry https://review.openstack.org/495891 | |
| 14:32:11 | mriedem | sean-k-mooney: that one is a known issue, | |
| 14:32:16 | mriedem | the real problem is libvirt crashed | |
| 14:32:25 | mriedem | so the libvirt driver auto-disables the compute service | |
| 14:35:16 | sean-k-mooney | mriedem: there does not seam to be any usfull info in the libvirtd log other then a bunch of virNetlinkEventCallback spam | |
| 14:35:33 | guimaluf | Could you guys help me to debug this error? http://paste.openstack.org/show/618918/ I'm trying to launch an instance within my testing environment, which consist of 4 vms, keystone/api-controllers, mysql/ceph/rabbitmq, neutron and nova. | |
| 14:36:04 | guimaluf | my concern is mainly focused on HTTPMultipleChoices: HTTPMultipleChoices (HTTP 300) Requested version of OpenStack Images API is not available. I expect nova to contact glance v2 api, but it is trying to reach v1 even with only v2 available | |
| 14:37:46 | mriedem | sean-k-mooney: i won't pursue any further - libvirt crashing is not a new thing in the gate | |
| 14:41:01 | sean-k-mooney | mriedem: well strangly the libvirtd log seams to still be active when the connection drops on the nova side but ya this does not really look like it s novas problem | |
| 14:41:28 | sean-k-mooney | mriedem: the compute agent will reconnect and reactivate itself anyway correct | |
| 14:41:33 | mriedem | yes | |
| 14:42:03 | mriedem | in a prod cloud with hundreds/thousands of nodes, that's probably good enough | |
| 14:42:09 | mriedem | for single node gate jobs it's a killer | |
| 14:43:32 | sean-k-mooney | ya tempest could maybe retry the failed jobs automatically to a certenlimit at the end fo the run in the future but honestly that will likey just slow down the gate when code is actully broken | |
| 14:49:57 | maciejjozefczyk | guimaluf: check nova-scheduler and rabbitmq | |
| 14:50:32 | maciejjozefczyk | guimaluf: s/nova-scheduler/nova-conductor/ | |
| 14:51:17 | mriedem | maciejjozefczyk: anything glance related is likely blowing up in the api | |
| 14:51:21 | mriedem | sounds like misconfiguration | |
| 14:51:32 | mriedem | [glance]api_servers is where i'd look | |
| 14:51:38 | mriedem | in nova.conf | |
| 14:51:56 | mriedem | and make sure that matches what's in the service catalog for the image service | |
| 14:57:34 | maciejjozefczyk | mriedem: Could you check this cherry-pick https://review.openstack.org/#/c/494974/ please :) ? | |
| 14:59:03 | mriedem | maciejjozefczyk: has to go through pike first, | |
| 14:59:19 | mriedem | and i see the cherry pick for pike, but this isn't a regression in pike, so i'm not sure we should include it in the rc2 for pike | |
| 15:01:09 | maciejjozefczyk | mriedem: So it should wait for stable pike? | |
| 15:01:25 | mriedem | maciejjozefczyk: it has to be in stable/pike before we can merge it in stable/ocata, yes | |