| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-06-07 | |||
| 18:04:54 | rm_work | maybe zigo at least knows what channel it is then? :P | |
| 18:05:00 | mriedem | that's my hope | |
| 18:05:11 | mriedem | "If you prefer live discussions, some of us also hang out in #lxcontainers on irc.freenode.net." | |
| 18:05:14 | mriedem | https://github.com/openstack/nova-lxd | |
| 18:05:43 | rm_work | ah, missed that | |
| 18:05:44 | rm_work | thanks | |
| 18:30:58 | elady | looking for soft affinity filter code | |
| 18:36:08 | mriedem | elady: nova/scheduler/filters | |
| 18:40:06 | elady | The affinity_filter.py has only affinity + anti-affinity the soft-affinity and soft-anti-affinity are not there and missing in code | |
| 18:41:47 | elady | meriedm : The affinity_filter.py has only affinity + anti-affinity the soft-affinity and soft-anti-affinity are not there and missing in code | |
| 18:53:57 | openstackgerrit | Matt Riedemann proposed openstack/nova master: Handle uuid in HostAPI.compute_node_get https://review.openstack.org/461532 | |
| 18:53:58 | openstackgerrit | Matt Riedemann proposed openstack/nova master: WIP: Use uuid for id in Services API https://review.openstack.org/464280 | |
| 18:54:48 | mriedem | elady: the soft affinity stuff is in the weigher https://github.com/openstack/nova/blob/master/nova/scheduler/weights/affinity.py | |
| 18:55:19 | mriedem | https://specs.openstack.org/openstack/nova-specs/specs/mitaka/implemented/soft-affinity-for-server-group.html | |
| 18:56:34 | openstackgerrit | Merged openstack/nova master: libvirt: update LibvirtConfigGuestDeviceAddress to provide XML https://review.openstack.org/459737 | |
| 19:02:21 | elady | mriedem: I created server group with soft-affinity policy and set soft-affinity weight to 1.0 and still not working as expected | |
| 19:08:00 | dansmith | jaypipes: melwitt see this on that bottom patch? http://logs.openstack.org/45/470645/6/check/gate-nova-tox-functional-py35-ubuntu-xenial/c97c7cb/testr_results.html.gz | |
| 19:09:54 | dansmith | I guess that can't be related since the bottom patch is just db stuff | |
| 19:10:25 | melwitt | dansmith: yeah. I've seen that before several times in the past. some "authentication error" that's transient in functional tests. I don't get that | |
| 19:12:36 | dansmith | ack | |
| 19:13:33 | melwitt | hm, though I just noticed this: | |
| 19:13:35 | melwitt | KeyError: 'allocations' | |
| 19:13:35 | melwitt | return resp.json()['allocations'] | |
| 19:13:35 | melwitt | nova/scheduler/client/report.py", line 914, in get_allocations_for_resource_provider | |
| 19:14:33 | melwitt | it's not part of the patch in question but it's something that seems wrong? | |
| 19:14:43 | dansmith | that was my point | |
| 19:14:47 | dansmith | but this patch is only db stuff | |
| 19:14:59 | melwitt | yeah | |
| 19:15:22 | dansmith | recheck-and-chill, I say | |
| 19:15:57 | melwitt | hah | |
| 19:16:58 | openstackgerrit | Merged openstack/nova master: libvirt: set full description of the controller used by disk https://review.openstack.org/459738 | |
| 19:17:42 | openstackgerrit | Merged openstack/nova master: libvirt: rearange how scsi controller is defined https://review.openstack.org/459739 | |
| 19:19:18 | mriedem | dansmith: do you remember why we didn't do a ComputeHostNotUnique thing here https://github.com/openstack/nova/blob/master/nova/compute/api.py#L4568 like we did for services? https://github.com/openstack/nova/blob/master/nova/compute/api.py#L4371 | |
| 19:19:49 | mriedem | dansmith: i've seen that functional test fail before | |
| 19:19:51 | mriedem | it's random | |
| 19:19:53 | dansmith | mriedem: because theres no delete | |
| 19:20:02 | mriedem | dansmith: but there is show | |
| 19:20:04 | dansmith | mriedem: you can delete a service by id, and you might delete the wrong one | |
| 19:20:06 | dansmith | sure | |
| 19:20:36 | dansmith | you were going to fix it too, but just not worry about changing the behavior if you get the wrong compute node I think, because it's not destructive | |
| 19:20:44 | dansmith | fix it == make it use uuids | |
| 19:20:56 | mriedem | yeah that's what i'm working on now | |
| 19:21:14 | dansmith | doesn't hurt to make it behave the same if you want, it's just not critical I think | |
| 19:21:33 | mriedem | yeah it's definitely not as bad as delete | |
| 19:25:24 | openstackgerrit | Rafael Folco proposed openstack/nova master: Delete all inventory has its own method DELETE https://review.openstack.org/460147 | |
| 19:36:23 | mriedem | uuuuuuuuuuuuuuggggggggggggggghhhhhhhhhhhhhhhh | |
| 19:36:27 | mriedem | so, | |
| 19:36:41 | mriedem | returning compute nodes from the api with uuids means we're also going to need to change the paging links | |
| 19:36:54 | mriedem | the paging stuff plumbed all the way down to the database is using id | |
| 19:37:51 | dansmith | gross | |
| 19:37:54 | mriedem | dansmith: so if the marker in https://github.com/openstack/nova/blob/master/nova/objects/compute_node.py#L402 can be an id or uuid, i assume that means i'll need a new remotable method? | |
| 19:37:57 | dansmith | same with services? | |
| 19:38:02 | mriedem | we don't have paging on services | |
| 19:38:04 | mriedem | thank god | |
| 19:38:36 | dansmith | mriedem: well, yeah, you should have a new remotable | |
| 19:38:40 | mriedem | also, | |
| 19:38:52 | mriedem | this likely means we aren't handling paging correctly right now with multiple cells | |
| 19:38:56 | dansmith | because otherwise we have no way to know that the other end doesn't support it | |
| 19:38:57 | dansmith | also, | |
| 19:39:06 | dansmith | you could do a get by uuid, then use the id and make the paged call | |
| 19:39:10 | dansmith | which is dumb, but.. | |
| 19:39:14 | dansmith | or | |
| 19:39:21 | dansmith | you could cheat and do that only if you pass a uuid and it barfs | |
| 19:39:49 | dansmith | well, if the marker is integer id, you *can't* do paging right | |
| 19:41:53 | mriedem | so get the compute by uuid, which we can then look up the host mapping and know which cell it's in, and then do a cell-targeted context to page by the int id for that compute node | |
| 19:41:55 | mriedem | right? | |
| 19:42:16 | mriedem | i was just thinking about how we page across cells for instances | |
| 19:42:17 | mriedem | by uuid | |
| 19:42:39 | dansmith | um | |
| 19:42:58 | dansmith | you can't get compute node by id or uuid without a host mapping already in your hand | |
| 19:43:09 | mriedem | if we have 10 compute nodes in 2 cells, and our marker is the 10th CN in cell1 and we have a limit of 11, then wouldn't you expect to continue paging to the next cell? | |
| 19:43:34 | dansmith | yes/ | |
| 19:44:23 | mriedem | "you can't get compute node by id or uuid without a host mapping already in your hand" | |
| 19:44:32 | mriedem | we don't get a host mapping when doing a show on os-hypervisors | |
| 19:44:41 | mriedem | that's the thing where we have an int id today and iterate the cells looking for one | |
| 19:44:49 | mriedem | which is https://github.com/openstack/nova/blob/master/nova/compute/api.py#L4568 | |
| 19:45:35 | mriedem | seems we should have stored the compute_nodes.uuid in the host_mappings table, | |
| 19:45:46 | mriedem | then if we're given a uuid, we could have looked up the host mapping by uuid to find the cell rather than iterate | |
| 19:46:15 | dansmith | mriedem: you can't, because a hostmapping is for nova-compute not for the ComputeNode right? | |
| 19:46:20 | dansmith | actually, | |
| 19:46:36 | dansmith | maybe that's wrong. I guess ironic needs discovery when adding new ironic nodes | |
| 19:46:54 | mriedem | the host mapping is the the compute node | |
| 19:47:00 | mriedem | mapped by compute_node.host | |
| 19:47:04 | mriedem | rather than uuid | |
| 19:47:07 | mriedem | for whatever reason... | |
| 19:47:13 | dansmith | my "you can't" statement was in response to your "get by compute uuid, then look up host mapping" comment | |
| 19:47:22 | dansmith | if you've found it, you need not lookup the hostmapping at that point | |
| 19:47:28 | mriedem | we can look up the host mapping via compute_node.host | |
| 19:47:41 | mriedem | sure, i was thinking about the paging across cells thing | |
| 19:47:44 | dansmith | you can, but you've already found it at that point | |
| 19:47:55 | dansmith | you don't need the hostmapping if you already know what cell it's in, | |
| 19:47:59 | dansmith | and if you found it, you know | |
| 19:48:06 | dansmith | right? | |
| 19:48:54 | mriedem | idk | |
| 19:48:57 | mriedem | been up since 4 | |
| 19:48:58 | mriedem | this sucks | |
| 19:49:19 | mriedem | i'm just going to TODO bomb this change for now | |
| 20:05:13 | zigo | rm_work: You should ask jamespage about lxd. | |
| 20:05:46 | rm_work | will do, thanks | |
| 20:08:17 | melwitt | merge conflict extravaganza! | |