| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-10-11 | |||
| 18:37:44 | openstackgerrit | Dan Smith proposed openstack/nova master: Refactor resource tracker to account for migration allocations https://review.openstack.org/506419 | |
| 18:37:44 | openstackgerrit | Dan Smith proposed openstack/nova master: Make migration uuid hold allocations for migrating instances https://review.openstack.org/506420 | |
| 18:37:45 | openstackgerrit | Dan Smith proposed openstack/nova master: Make live migration hold resources with a migration allocation https://review.openstack.org/507638 | |
| 18:39:02 | mriedem | ok | |
| 18:39:20 | mriedem | wonkification of the SchedulerReportClient 2.0 | |
| 18:40:59 | dansmith | heh | |
| 18:55:07 | openstackgerrit | Matt Riedemann proposed openstack/nova-specs master: Spec for flavor description https://review.openstack.org/501017 | |
| 18:55:11 | mriedem | sdague: made those changes on restricting the id and name fields ^ | |
| 18:55:49 | sdague | mriedem: will look | |
| 19:07:00 | openstackgerrit | Merged openstack/nova master: Update cinder in RequestContext service catalog https://review.openstack.org/510947 | |
| 19:12:51 | sdague | mriedem: that looks completely sane to me | |
| 19:13:32 | mriedem | not only sane, but completely sane? | |
| 19:13:33 | mriedem | nice | |
| 19:17:41 | mriedem | huh https://developer.openstack.org/api-ref/placement/#resource-provider-inventories | |
| 19:17:43 | mriedem | is missing POST | |
| 19:37:29 | openstackgerrit | sean mooney proposed openstack/nova master: WIP reorder destination host setup https://review.openstack.org/510658 | |
| 19:41:18 | openstackgerrit | melanie witt proposed openstack/nova stable/ocata: Target context for build notification in conductor https://review.openstack.org/511325 | |
| 19:47:45 | mriedem | easy fix here: https://review.openstack.org/#/c/511209/ | |
| 19:47:55 | mriedem | hot dogs and cracker jacks here! | |
| 19:49:07 | melwitt | +W | |
| 19:50:23 | mriedem | https://bugs.launchpad.net/nova/+bug/1722891 in case someone wants it | |
| 19:50:24 | openstack | Launchpad bug 1722891 in OpenStack Compute (nova) "api-ref: POST /resource_providers/{uuid}/inventories is missing" [Medium,Confirmed] - Assigned to hongbin (hongbin034) | |
| 19:50:57 | openstackgerrit | Hongbin Lu proposed openstack/nova master: Handle not found error on taking snapshot https://review.openstack.org/511074 | |
| 19:52:01 | mriedem | grenade jobs seem totally borked in zuulv3 | |
| 19:53:31 | melwitt | mriedem: looks like hongbin is already gonna work on the bug you linked, so that's cool | |
| 19:55:22 | mriedem | something really weird happens with our request id log handlin | |
| 19:55:27 | mriedem | seems to be not thread safe or something | |
| 19:55:40 | mriedem | i'm tracing a server create request through the n-cpu logs and see the normal stuff i'd expect while creating the instance, | |
| 19:55:52 | mriedem | but then i see the same request id showing up in log messages for periodic tasks that are running | |
| 19:56:52 | melwitt | hm, I can't say I've seen that before | |
| 19:57:25 | melwitt | which makes me think it's a relatively new issue | |
| 20:34:10 | openstackgerrit | Matt Riedemann proposed openstack/nova master: api-ref: fix server status values in GET /servers docs https://review.openstack.org/510696 | |
| 20:34:10 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Update "SHUTOFF" description in API guide https://review.openstack.org/510697 | |
| 20:43:08 | mriedem | uh, filter(~models.Migration.status.in_ means NOT IN right? | |
| 20:43:11 | mriedem | the ~ | |
| 20:47:20 | efried | sdague https://review.openstack.org/#/c/511006/ basically passed - the one failure seems unrelated and I'd rather not choke the already-seemingly-choking gate further. | |
| 20:54:57 | mriedem | dansmith: want to double check my thoughts about migration.status = 'done' in here? https://review.openstack.org/#/c/506419/ otherwise i think it's ok | |
| 21:01:23 | mriedem | this also seems like it could bite us https://github.com/openstack/nova/blob/64635ad4a5f60a79e1ec2d5369a8f84bfeccb7e4/nova/db/sqlalchemy/api.py#L4859-L4862 | |
| 21:01:47 | mriedem | when the RT pulls migration records, it's for all migrations where either the source or dest is our local node | |
| 21:08:13 | dansmith | mriedem: yep will in a sec | |
| 21:20:25 | dansmith | mriedem: replied | |
| 21:22:22 | mriedem | ok didn't think about not having migration allocations for evacs | |
| 21:22:41 | mriedem | but makes sense to not put the source node allocations on the migration record for an evacuation since the source node should be dead | |
| 21:22:53 | dansmith | and you're never going back there.. no rollback | |
| 21:22:58 | mriedem | but...are we doubling today? | |
| 21:23:07 | mriedem | in the scheduler when we evacuate we must be doubling | |
| 21:23:12 | dansmith | if we are it's not for any useful reason I think | |
| 21:23:25 | mriedem | i think it's just because we do it generically | |
| 21:23:28 | dansmith | I think we delete the allocation for the old node first anyway right? | |
| 21:25:25 | dansmith | allocate _for_evacuate_dest_host | |
| 21:26:27 | mriedem | that's when we're forcing the host during evacuate and bypassing the scheduler | |
| 21:26:33 | dansmith | yeah, we end up calling claim_resources and doubling | |
| 21:26:45 | dansmith | no, that method calls the scheduler | |
| 21:26:57 | mriedem | _allocate_for_evacuate_dest_host? | |
| 21:26:59 | mriedem | it doesn't | |
| 21:27:08 | mriedem | it calls scheduler utils | |
| 21:27:11 | dansmith | https://github.com/openstack/nova/blob/master/nova/conductor/manager.py#L782-L782 | |
| 21:27:16 | mriedem | b/c we do the same util thing for live migration with a forced host | |
| 21:27:17 | dansmith | which calls claim_resources | |
| 21:27:29 | dansmith | sorry, maybe it does't call scheduler, but it does the doubling, which is what I meant | |
| 21:27:30 | mriedem | right, that's not select_destinations | |
| 21:27:33 | mriedem | yeah | |
| 21:27:39 | dansmith | sure, I was focused on the allocs | |
| 21:27:44 | mriedem | this is the thing i want to remove with a skip_filters flag to select_destinations | |
| 21:27:56 | mriedem | so getting back to my original question i guess, | |
| 21:27:59 | dansmith | regardless, I don't think this affects the new-world path | |
| 21:28:05 | mriedem | we never remove the allocation from the source node for an evacuation | |
| 21:28:15 | mriedem | except when the source node comes back, if it does | |
| 21:28:18 | openstackgerrit | Hongbin Lu proposed openstack/nova master: placement: add API reference for create inventory https://review.openstack.org/511342 | |
| 21:28:29 | dansmith | you think we never do it now? | |
| 21:28:47 | dansmith | or you think we never will with the new-world way? | |
| 21:28:51 | mriedem | https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L680 | |
| 21:28:55 | dansmith | ah, maybe your point is, | |
| 21:29:07 | mriedem | it's going to be a weird side thing | |
| 21:29:08 | dansmith | in the new world way we'll still end up calling the doubler? | |
| 21:29:15 | mriedem | it's a move operation where the source allocation isn't on the migration uuid | |
| 21:29:43 | mriedem | unlike live migrate, cold migrate/resize | |
| 21:29:43 | dansmith | can't we just stop doing the doubling across the board? | |
| 21:29:50 | dansmith | just delete it before we call claim_resources | |
| 21:29:57 | mriedem | for evac? | |
| 21:30:01 | dansmith | that should work for both old and new paths | |
| 21:30:02 | dansmith | yeah | |
| 21:30:08 | mriedem | well, unless evac fails on the dest | |
| 21:30:13 | mriedem | and the instance never moved | |
| 21:30:15 | dansmith | what does it matter? you're not going back | |
| 21:30:34 | dansmith | we've created the migration, the source node is going to nuke it when it wakes up, per the rules | |
| 21:30:42 | mriedem | not necessarily | |
| 21:30:43 | mriedem | https://github.com/openstack/nova/blob/master/nova/compute/manager.py#L643 | |
| 21:30:56 | mriedem | the source only cleans up if the move completed, or is in progress | |
| 21:32:03 | dansmith | um | |
| 21:32:24 | dansmith | the whole point of that robustification was to not do that, | |
| 21:32:32 | dansmith | else we'll race with the operation finishing | |
| 21:32:46 | dansmith | I don't see where we're setting =accepted anymore actually | |
| 21:32:53 | mriedem | conductor | |
| 21:32:56 | mriedem | er api | |
| 21:33:00 | dansmith | I don't see it | |
| 21:33:11 | dansmith | ah I see | |
| 21:33:20 | mriedem | yeah so the api creates the migration record in 'accepted' status, | |
| 21:33:22 | dansmith | right, so that happens synchronously with the api call | |
| 21:33:28 | dansmith | right? | |