Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-12-12
19:46:18 mriedem alright then
19:46:35 dansmith as you should
19:47:03 dansmith let it be remembered for all time, the last thing mriedem tried to do was an api-level hack for vmware
19:47:14 dansmith can you self paddle?
19:47:21 mriedem can a bear shit in the woods?
19:47:26 dansmith heh
19:47:26 dansmith nevermind, don't answer
20:00:33 sean-k-mooney hehe if only this was archive publicly for al time
20:00:44 sean-k-mooney *all
20:11:17 KeithMnemonic mriedem do you have a few to help me with a different marker issue. this one is in master as well and easily reproducible
20:12:40 KeithMnemonic not sure if it is just how horizon handles it or is it something in nova. under instances, there is not a prev/next button that came from this patch https://review.opendev.org/#/c/498018/34
20:12:50 KeithMnemonic s/not/now
20:13:21 KeithMnemonic so that patch added a prev button, prior there was only next. but if you jump forward and back a few times you see it does not work as expected
20:14:09 KeithMnemonic seems to be due to the original list is not sorted or the sort is not maintained after doing 'next' than 'prev'
20:15:46 KeithMnemonic for example i have 5 instances test-1 through test-5. i set the "items per page" under admin settings to 2
20:17:25 KeithMnemonic if i jump fresh to instances i see test-5, test-4, if i click next i get test-2, test-1. if i click prev, i get test test-3 and test-1
20:17:44 KeithMnemonic so horizon is using marker= and prev_marker= and that seems to not work well
20:18:01 KeithMnemonic has anyone else seen this?
20:18:09 mriedem prev_marker isn't a thing in nova
20:18:20 KeithMnemonic so something horizon bonked?
20:18:38 mriedem maybe they aren't providing the full query params to maintain a consistent sort order, i.e. not passing the sort_key/sort_dir params when paging?
20:23:29 KeithMnemonic could be maybe somewhere in here https://opendev.org/openstack/horizon/src/branch/master/horizon/tables/base.py#L699
20:25:09 mriedem idk
20:25:14 mriedem don't really have time to dig into that
20:26:14 KeithMnemonic no problem let me go bother horizon, thanks for confirming
20:26:57 mriedem but it does look like they are losing the sort key/dir
20:29:05 efried sean-k-mooney: https://review.opendev.org/#/c/674072/ +A, nice work on those tests. (See inline though.)
20:31:41 mriedem a couple of low-hanging bugs if someone wants something to play with
20:31:43 openstack Launchpad bug 1856241 in OpenStack Compute (nova) "Compute API in nova - servers_links should link to paging doc" [Undecided,New]
20:31:43 mriedem https://bugs.launchpad.net/nova/+bug/1856241
20:31:46 mriedem https://bugs.launchpad.net/nova/+bug/1856240
20:31:47 openstack Launchpad bug 1856240 in OpenStack Compute (nova) "wait_for_versioned_notifications gives unhelpful error message "ValueError: Not a text type application/octet-stream" on timeout" [Undecided,Confirmed]
20:57:10 jroll melwitt: re https://bugzilla.redhat.com/show_bug.cgi?id=1688838 - I feel like cpu_arch used to be required, but not 100%. looks like we intentionally made it not required here, at least: https://review.opendev.org/#/c/620634/
20:57:11 openstack bugzilla.redhat.com bug 1688838 in openstack-nova "Ironic should not treat cpu_arch as mandatory" [Medium,New] - Assigned to mwitt
20:59:38 openstackgerrit Eric Fried proposed openstack/nova master: DRY: Build ImageMetaPropsPayload from ImageMetaProps https://review.opendev.org/698803
20:59:47 efried dansmith: question in here ^ for you
21:00:00 efried sean-k-mooney: that ^ passes tests :)
21:01:58 dansmith efried: if the question is whether they need to be instances or classes, they need to be the former
21:02:21 dansmith the instance is what contains things like nullable=, default=, etc
21:02:34 dansmith and just because those don't have any of those set (currently) doesn't affect what the rest of the code is expecting
21:02:51 efried The question is whether they need to be discrete instances for every field of every object, or whether I could just copy in the instances from the other class.
21:03:30 efried int1: f,
21:03:30 efried fields = {
21:03:30 efried class MyObj(...):
21:03:30 efried f = IntegerField()
21:03:30 efried like, could I theoretically do this:
21:03:31 efried }
21:03:31 efried int2: f,
21:03:52 dansmith oh, well, I'll have to go look,
21:03:59 dansmith but the convention is obviously for each to be new
21:04:11 dansmith I can't remember if we stash any object-specific state in the field
21:04:19 efried Okay. Then prolly leave the code as it is now
21:05:12 efried thanks
21:06:23 openstackgerrit Eric Fried proposed openstack/nova master: DRY: Build ImageMetaPropsPayload from ImageMetaProps https://review.opendev.org/698803
21:06:31 efried done, and changed the TODO to a NOTE
21:27:54 sean-k-mooney efried: oh i forgot the has. i just ran the test i added
21:27:58 sean-k-mooney *hash
21:28:16 efried sean-k-mooney: yahbut, see my fup
21:28:41 efried was going to say maybe you want to squash; but really they are quite different things, so let's keep them separate.
21:28:57 efried plus I don't want to lose the ability to +2 yours :P
21:29:22 sean-k-mooney yep i did i was wondering about that
21:29:34 sean-k-mooney if we do the follow up then we might as well nuke the object
21:29:56 sean-k-mooney well
21:31:23 sean-k-mooney we could replace the ImageMetaPayload 'properties': fields.ObjectField('ImageMetaPropsPayload') definition
21:32:02 sean-k-mooney with another filed that just reference the other object
21:32:24 melwitt jroll: thanks. I'm trying to figure out what's the correct thing to do in the nova ironic driver if ironic is not filling in a cpu_arch in a node. should we choose a default arch to report for ironic scheduling? if so, how should we choose it? a new config option? those are the kind of questions I'm trying to find answers about
21:35:06 sean-k-mooney efried: ok just read your comments on the patch below. ill create a fup to add the last combination and move the comments to doc strings afat i fix the Sync patch
21:35:51 sean-k-mooney and ya the ddt module is really nice to use. ill try to use it more often when it makes sense
21:37:13 efried sean-k-mooney: ++
21:38:18 efried sean-k-mooney: Reasons not to nuke the object: it would make version boundaries weird; and I'm not sure what happens with SCHEMA.
21:38:32 efried It may be possible. But I'd leave it up to gibi
21:39:34 sean-k-mooney well with your follow up patch every time we add a filed to the normal nova object the notificaiton object fileds and schema will auto update chaning the hash
21:40:14 sean-k-mooney so if we apply that patch we need to rememebr to bump the notificion object version and document the new filed even though we are not modifying the class
21:40:33 efried ohhh, good point, the version. That needs a -1.
21:40:57 sean-k-mooney that is still less work then today however. ya i can put that in a review comment
21:41:28 efried Yeah, we at least need to make a note in the class with that instruction.
21:41:41 efried the test will fail, but you could get around it by just updating the hash without bumping the version.
21:41:52 sean-k-mooney yep
21:41:54 efried an astute reviewer ought to notice, but...
21:42:11 jroll melwitt: it feels like it should be a required field for an ironic node
21:42:20 jroll there's no way to guess at it
21:56:24 sean-k-mooney jroll: the cpu architecture
21:57:07 sean-k-mooney ya i mean some hyperisors can emulate other archs but that is kind of true for all virt drivers
22:00:32 mriedem ironic != hypervisor
22:00:54 sean-k-mooney also true
22:01:55 sean-k-mooney i was makeing the point that most of the virt driver only support one archtecture at a time so you amost always want to know the cpu architecture if you care about check the image for compatablity
22:03:30 sean-k-mooney there was that project denvor chip that nviad wanted to make but then never closed the license agreement with intel for x86
22:04:41 sean-k-mooney iwas going to be an arm architecture chip with native traslation of x86 to arm opcode in hardware allowing them to run code comiled for either achtecture
22:05:21 sean-k-mooney im sure that would have been a pain to debug but it would have been cool if it ever worked
22:11:28 openstackgerrit sean mooney proposed openstack/nova master: Sync ImageMetaPropsPayload fields https://review.opendev.org/698613
22:15:36 openstackgerrit Matt Riedemann proposed openstack/nova master: Use COMPUTE_SAME_HOST_COLD_MIGRATE trait during migrate https://review.opendev.org/695220
22:23:39 sean-k-mooney efried: actully my brain has already switched off today so ill do the FUP tomorrow the other FUP i promised is https://review.opendev.org/#/c/698260/ but there is no really rush with reviewing that
22:24:01 efried ack
22:24:42 openstackgerrit Merged openstack/nova master: trivial: Resolve (most) flake8 3.x issues https://review.opendev.org/695732
22:41:53 openstackgerrit Matt Riedemann proposed openstack/nova master: Allow overriding self.api in _build_minimal_create_server_request https://review.opendev.org/698813
22:41:53 openstackgerrit Matt Riedemann proposed openstack/nova master: WIP: Add NodeOwnerFilter https://review.opendev.org/697331
22:51:58 openstackgerrit Merged openstack/nova master: Don't hardcode Python versions in test https://review.opendev.org/698732
23:39:21 openstackgerrit Merged openstack/nova master: Keep pre-commit inline with hacking and fix whitespace https://review.opendev.org/674057
23:42:35 openstack bug 1849409 in OpenStack Compute (nova) queens "openstack server list --deleted --limit -1 hangs" [High,In progress] https://launchpad.net/bugs/1849409 - Assigned to Matt Riedemann (mriedem)
23:42:35 openstackgerrit Merged openstack/nova stable/queens: Add functional regression test for bug 1849409 https://review.opendev.org/690724

Earlier   Later