Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2019-12-13
19:59:57 efried yeah, looking at the previous run, it's only pep8 and docs.
20:00:01 efried so it ought to work.
20:06:29 openstackgerrit Matt Riedemann proposed openstack/nova stable/rocky: Add --dry-run option to heal_allocations CLI https://review.opendev.org/698525
20:06:30 openstackgerrit Matt Riedemann proposed openstack/nova stable/rocky: Add BFV wrinkle to TestNovaManagePlacementHealAllocations https://review.opendev.org/698531
20:06:30 openstackgerrit Matt Riedemann proposed openstack/nova stable/rocky: Add --instance option to heal_allocations https://review.opendev.org/698529
20:17:27 mriedem efried: i replied for gibi on https://review.opendev.org/#/c/696574/ - see if that answers your question
20:19:22 efried thanks mriedem, +A
20:19:49 efried mriedem: wanna hit this trivial +A? https://review.opendev.org/698988
20:23:49 efried dansmith: bikeshedding opportunity:
20:23:56 mriedem done
20:24:59 efried RequestSpec needs a field (possibly two) to store root_required (which can take required or forbidden (with !) just like regular required). This is parallel with requested_resources.
20:25:31 efried It's what we always wanted to use from those request_filters rather than hacking into the flavor and doing obj_reset_changes(), which I'll fix up at the same time.
20:25:33 mriedem that can't be hacked into the RequestGroup?
20:25:50 efried No, specifically because it's not per request group.
20:25:51 dansmith right, was just going to say.. why isn't it inside requested_resources?
20:26:02 efried That's it's whole raison d'être
20:26:11 efried requested_resources is a list of RequestGroup.
20:26:14 mriedem the issue i always had with the RequestGroup for that traits-only stuff is the RequestGroup stuff in the scheduler made resources required
20:26:35 dansmith oh because requested_resources is a list of requestgroup objects
20:27:09 dansmith that field probably should have been an ObjectField('PlacementyShit')
20:27:11 efried well a) pretty sure that's not a thing anymore, I think we did away with that on the placement side; but b) that's exactly what root_required is for: so you can specify required/forbidden traits that you *for sure* want to look for on the root RP, without association with any request groups
20:27:12 mriedem can you have a root unnumbered group or something for modeling this?
20:27:26 mriedem DictOfStringsField(put all the crap)
20:27:58 mriedem so building on RequestGroup as an object would probably just be confusing for tihs right?
20:28:11 efried would make no sense
20:28:24 efried a request has N RequestGroup. It only has one root_required
20:28:39 efried RequestGroup contains resources, traits, aggregates. root_required only traits.
20:28:41 dansmith efried: what is your actual quesiton?
20:28:46 efried what should we name those fields
20:28:54 dansmith ?
20:28:54 dansmith field or fields
20:28:56 efried One for required, one for forbidden. SetOfStringsField.
20:29:03 efried root_required, root_forbidden?
20:29:14 efried root_required_traits, root_forbidden_traits?
20:29:21 mriedem feels like a field explosion
20:29:28 efried requested_root_traits_required, requested_root_traits_forbidden?
20:29:30 mriedem throw it into a RootProviderConstraints objects?
20:29:30 dansmith how about we add PlacementyShit and put them on that object, so that when we have the next thing, we don't keep adding placementy fields to reqspec?
20:29:42 dansmith mriedem: PlacementyShit
20:30:36 dansmith pretty sure the tc had to cancel their phone plan to pay the electric bill
20:30:53 efried yeah, we could make ResourceRequest an OVO and make RequestSpec.resource_request. That's probably something we want to consolidate eventually anyway.
20:31:04 mriedem ha
20:31:25 mriedem RequestSpec.resource_request and RequestSpec.requested_resources
20:31:26 efried that would mean we should drop requested_resources though. Deleting a field from an OVO sounds scary.
20:31:27 mriedem very confusing
20:31:33 mriedem it's a major version bump
20:31:39 mriedem the likes of which only dan-o can navigate
20:31:40 efried Course, everything OVO is scary to me.
20:31:58 efried so what's the lesser evil here?
20:32:13 efried placement_shit = DictOfStringsField?
20:32:14 dansmith yep, don't do that right now, I'd just add a peer object next to requested_resources, leave the latter the same, put your flags and stuff in the new object
20:32:16 dansmith no
20:32:21 dansmith god no.
20:32:56 efried new PlacementShit OVO and placement_shit = PlacementShitField()?
20:33:03 dansmith no
20:33:29 dansmith reqspec.PlacementShit = ObjectField('PlacementShit'),
20:33:48 dansmith then in the PlacementShit object, whatever fields you need
20:33:52 dansmith root_required is what, a boolean?
20:33:58 efried set of traits
20:33:59 dansmith or a list of traits
20:34:08 efried prefer a set if poss, to squash dups
20:34:10 dansmith so root_required = ListOfStringsField()
20:34:26 efried We have a SetOfStringsField, any reason to avoid it?
20:34:28 mriedem we do have a set of strings field, RequestGroup uses it
20:34:34 dansmith not sure we have one of those, but seriously we don't have to use sets everywhere, just run it through a set somewhere
20:34:43 dansmith no, if we have one that's fine
20:34:43 mriedem sets or bust!
20:34:48 efried cool
20:34:52 dansmith it's silly, IMHO, but... :)
20:34:53 efried thanks.
20:34:56 mriedem BagODicts
20:41:05 openstackgerrit Balazs Gibizer proposed openstack/nova master: Func test for qos live migration reschedule https://review.opendev.org/699015
20:43:42 efried dansmith: and what's the accepted way to use default on a field?
20:44:15 efried e.g. default=set(), does that help me not have to check if present => check if None => check if empty?
20:46:07 openstackgerrit Merged openstack/nova-specs master: [ussuri][goal] Drop python 2.7 support https://review.opendev.org/698988
20:48:07 openstackgerrit Merged openstack/nova stable/pike: Add functional regression test for bug 1849409 https://review.opendev.org/690734
20:48:08 openstack bug 1849409 in OpenStack Compute (nova) pike "openstack server list --deleted --limit -1 hangs" [High,In progress] https://launchpad.net/bugs/1849409 - Assigned to Matt Riedemann (mriedem)
20:50:40 dansmith efried: default=set() isn't going to work anyway
20:50:55 efried it's what RequestGroup is doing
20:51:03 efried so, in what way not work?
20:51:31 dansmith efried: IIRC, that will result in every object getting the same set(), potentially cross-polluting between requests
20:51:41 dansmith because the field is once per definition, and done at import time
20:51:47 efried oh, neat.
20:52:19 dansmith I don't see request group doing that
20:52:46 dansmith er, maybe I'm looking in the wrong place
20:53:22 dansmith ah yeah I see
20:54:09 dansmith oh, hmm, maybe we copy the field's definition in obj_set_defaults, hang on
20:54:17 dansmith ISTR that coming up and we hacked around it that way
20:54:44 dansmith yeah https://github.com/openstack/oslo.versionedobjects/blob/master/oslo_versionedobjects/base.py#L588
20:54:57 dansmith alright, so ignore me I guess
20:55:31 dansmith a little dangerous but probably not in practice
20:55:47 efried then back to the original question, how should I manage this field?
20:55:48 dansmith but no, it doesn't help you not have to check for setted-ness
20:56:01 dansmith efried: you should always set it, or call obj_set_defaults() after you create it
20:56:11 efried but not via __init__
20:56:22 dansmith I think you know the answer to that :)
20:56:35 efried would this be a thing for __new__ somehow?
20:56:55 dansmith it's less of a problem for a transient rpc-only object, but we really shouldn't create more examples for people to copy, not knowing the problems
20:56:56 dansmith efried: no
20:57:18 mriedem RequestGroup just sets defaults if it lazy-loads a field and i think all fields have defaults

Earlier   Later