| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-07-06 | |||
| 22:51:50 | mriedem | looks like it's going to be an all nighter | |
| 22:52:25 | ildikov | mriedem: yeah, I fixed a few added just a few more and went ahead to fix the attach patch... | |
| 22:52:35 | mriedem | np | |
| 22:52:37 | mriedem | i'll add the rest | |
| 22:52:52 | ildikov | mriedem: prolly waaaay faster if you add the rest | |
| 22:53:56 | ildikov | mriedem: I owe you | |
| 22:54:09 | ildikov | mriedem: thank you! | |
| 23:00:25 | mriedem | ildikov: confirmed that the new style volume swap was tested in the change at the top of the series http://logs.openstack.org/85/330285/86/check/gate-tempest-dsvm-neutron-full-ubuntu-xenial/15d8c0f/console.html#_2017-07-05_17_04_17_100023 | |
| 23:01:43 | mriedem | that's the nova-initiated swap, not the cinder one, | |
| 23:01:49 | mriedem | we have another job that tests that in the experimental queue i htink | |
| 23:02:21 | ildikov | mriedem: I saw it failing when I messed up something earlier so I knew it's supposed to run | |
| 23:02:29 | mriedem | the tempest test is TestVolumeMigrateRetypeAttached | |
| 23:02:57 | ildikov | Hmm I don't recall that one | |
| 23:02:59 | mriedem | i think it's tested from the lvm multibackend job | |
| 23:03:19 | mriedem | that's the one that scottda added for testing volume migration/retype | |
| 23:04:02 | mriedem | i think tempest-dsvm-neutron-scenario-multinode-lvm-multibackend | |
| 23:04:09 | ildikov | Yeah, I remember that I just don't remember having any issues with it so I didn't deal with it | |
| 23:04:27 | mriedem | there is also tempest-dsvm-lvm-multibackend but that doesn't run slow tests, and this is tagged as a slow test | |
| 23:04:33 | mriedem | experimental queue jobs won't vote | |
| 23:04:39 | mriedem | so you likely wouldn't notice | |
| 23:05:31 | ildikov | I checked the non-voting jobs too when they failed | |
| 23:05:55 | mriedem | yeah but check queue non-voting != experimental queue non-voting | |
| 23:06:06 | mriedem | experimental queue jobs are all run on-demand | |
| 23:07:02 | ildikov | ok, missed the experimental part for a sec | |
| 23:11:15 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Pre-load instance.device_metadata in InstanceMetadata https://review.openstack.org/481235 | |
| 23:11:16 | mriedem | dansmith: melwitt: ^ fixed | |
| 23:13:31 | mriedem | jaypipes: i might just update this one https://review.openstack.org/#/c/476631/ | |
| 23:57:44 | jaypipes | melwitt: go for it | |
| 23:58:24 | jaypipes | lol, that should have been mriedem. | |
| 23:58:33 | jaypipes | mriedem: but I can fix those issues. sorry, was out of the hosue. | |
| #openstack-nova - 2017-07-07 | |||
| 00:06:02 | gmann | mriedem, good morning. can you check this spec, i updated the quota class APi fix spec for network quota filter out - https://review.openstack.org/#/c/480815/ | |
| 00:16:35 | openstackgerrit | Jay Pipes proposed openstack/nova master: placement: scheduler uses allocation candidates https://review.openstack.org/476631 | |
| 00:16:36 | openstackgerrit | Jay Pipes proposed openstack/nova master: WIP - placement: claim resources in scheduler https://review.openstack.org/476632 | |
| 00:16:37 | jaypipes | mriedem: ^ k, fixed up. | |
| 00:30:27 | mriedem | cool | |
| 00:35:18 | jaypipes | mriedem: sorry for the delay. was out this afternoon. | |
| 00:35:30 | mriedem | np, one teeny tiny thing https://review.openstack.org/#/c/476631/17 | |
| 00:35:36 | mriedem | and then +2 to the extreme | |
| 00:41:15 | jaypipes | mriedem: done. | |
| 00:41:21 | openstackgerrit | Jay Pipes proposed openstack/nova master: placement: scheduler uses allocation candidates https://review.openstack.org/476631 | |
| 00:41:22 | openstackgerrit | Jay Pipes proposed openstack/nova master: WIP - placement: claim resources in scheduler https://review.openstack.org/476632 | |
| 00:42:03 | alex_xu | jaypipes: easy fix for AllocationCandidates https://review.openstack.org/480379 :) | |
| 00:42:23 | alex_xu | jaypipes: sorry, that is the hard one, need your feedback, the easy one is https://review.openstack.org/480130 | |
| 00:42:53 | jaypipes | alex_xu: cool, will review both later tonight. | |
| 00:42:59 | alex_xu | jaypipes: thanks! | |
| 00:46:42 | openstackgerrit | yuanyue proposed openstack/nova master: Add log info in scheduler to mark start of scheduling https://review.openstack.org/481340 | |
| 00:47:05 | jaypipes | alex_xu: approved the one. nice catch! :) | |
| 00:47:15 | alex_xu | jaypipes: thanks \o/ | |
| 00:48:07 | jaypipes | alex_xu: would be good to add a functional test that verified that (though, I understand that all of our fixtures and the way we currently do testing would *not* have shown that issue since the usages table was appropriately filtered by resource_class_id already. | |
| 00:49:57 | alex_xu | jaypipes: you mean a functional test for usage filtered by resource_class_id? | |
| 00:51:33 | jaypipes | alex_xu: no, I mean add a functional test that showed the missing WHERE condition that was fixed in your patch. | |
| 00:52:30 | alex_xu | jaypipes: emm..I changed the function tests to expose that bug https://review.openstack.org/#/c/480130/3/nova/tests/functional/db/test_resource_provider.py | |
| 00:52:54 | mriedem | jaypipes: +2 | |
| 00:53:05 | mriedem | dansmith: for your morning, +2 on the allocation candidates in scheduler change https://review.openstack.org/#/c/476631/ | |
| 00:55:06 | jaypipes | alex_xu: understood. but those tests, while exposing the bug, didn't directly call _get_usages_by_provider_and_rc(). that's all I was saying, sorry, I wasn't being clear! | |
| 00:56:05 | alex_xu | jaypipes: ah, I see now | |
| 00:56:55 | alex_xu | I will check that | |
| 01:03:31 | openstackgerrit | Merged openstack/nova master: Updated from global requirements https://review.openstack.org/481189 | |
| 01:15:08 | openstackgerrit | Matt Riedemann proposed openstack/nova master: DNM: test new style volume attach with live migration https://review.openstack.org/481290 | |
| 01:15:10 | openstackgerrit | yuanyue proposed openstack/nova master: Add log info in scheduler to mark start of scheduling https://review.openstack.org/481340 | |
| 01:16:47 | openstackgerrit | Matt Riedemann proposed openstack/nova master: DNM: test new style volume attach with live migration https://review.openstack.org/481290 | |
| 01:21:17 | alex_xu | jaypipes: https://review.openstack.org/#/c/480379/2/nova/objects/resource_provider.py@2422...i'm doing stupid thing...why I join the table! | |
| 01:21:30 | jaypipes | alex_xu: :) done that many times myself. | |
| 01:22:15 | alex_xu | jaypipes: for 2637, it should be English problem, I shouldn't use 'filter out', I will reparse | |
| 01:22:26 | jaypipes | kk | |
| 01:22:51 | alex_xu | jaypipes: bauzas expect one SQL to done all the things, do you expect that, or you are happy the current way. | |
| 01:23:16 | jaypipes | alex_xu: I don't mind if there's >1 SQL statement, no. | |
| 01:23:20 | mriedem | are you asking jaypipes if he thinks sql can do everything? | |
| 01:23:25 | jaypipes | heh | |
| 01:23:27 | alex_xu | mriedem: :) | |
| 01:24:08 | mriedem | jay's quest in life is to find the one single sql statement that can rule them all | |
| 01:24:23 | jaypipes | SELECT 42 FROM DUAL. | |
| 01:24:40 | alex_xu | jaypipes: can you rewrite the whole nova api with one sql? please | |
| 01:24:51 | jaypipes | lol. dansmith would kill me. | |
| 01:27:35 | openstackgerrit | yuanyue proposed openstack/nova master: Add log info in scheduler to mark start of scheduling https://review.openstack.org/481340 | |
| 01:53:50 | Kevin_Zheng | Hi, when we boot instance, in cells v2, we call schedule_and_build_instances, but if we have to reschedule, we called build_instances, why is that? | |
| 01:54:38 | Kevin_Zheng | shouldn't we call schedule_and_build_instances again? | |
| 01:54:39 | mriedem | because nova-compute has always done a reschedule by doing an rpc cast to build_instances | |
| 01:54:51 | mriedem | Kevin_Zheng: no, build_instances is in the cell-level conductor, | |
| 01:54:54 | mriedem | so the compute can reach it | |
| 01:55:04 | mriedem | but schedule_and_build_instances talks to the scheduler, which is all at the top | |
| 01:55:04 | Kevin_Zheng | AH, ok | |
| 01:55:09 | mriedem | and we don't want the computes to be able to reach the top | |
| 01:55:16 | mriedem | "super" conductor | |
| 01:55:21 | Kevin_Zheng | I understood now | |
| 01:55:28 | mriedem | although, build_instances calls the scheduler too... | |
| 01:55:49 | mriedem | build_instances isn't cell aware either, | |
| 01:56:06 | mriedem | schedule_and_build_instances is working under the assumption that the instance does not exist in a cell yet | |
| 01:56:11 | mriedem | that's the better reason why the compute doesn't call it | |
| 01:57:44 | Kevin_Zheng | error isntances should stay in cell0 or cellN？ | |
| 01:57:54 | mriedem | cell0 | |
| 01:58:04 | mriedem | if you're talking about an instance that fails to build at the top and doesn't reach a cell | |
| 01:58:15 | mriedem | if it reaches a cell and eventually goes to error, then it's in the cell it was scheduled in | |
| 01:58:51 | Kevin_Zheng | OK | |
| 01:59:03 | Kevin_Zheng | got it | |
| 01:59:34 | mriedem | we should probably have something in https://docs.openstack.org/nova/latest/cells.html#faqs about cells0 | |
| 01:59:40 | mriedem | since that's a common question | |
| 02:00:15 | mriedem | although higher up it describes cell0, | |
| 02:00:16 | mriedem | "Instances that are never scheduled are relegated to the cell0 database, which is effectively a graveyard of instances that failed to start. All successful/running instances are stored in “cell1”." | |
| 02:00:48 | mriedem | an instance can reach a cell and still fail to start, | |