Earlier  
Later
New Search
Posted Nick Remark
#openstack-nova - 2017-08-28
19:13:44 edleafe artom: probably just obj.foo to get the dict
19:14:47 cdent edleafe: any of this help https://github.com/openstack/nova/blob/master/nova/objects/resource_provider.py#L2338-L2359
19:14:56 artom edleafe, I would think it'd be just a Dict then
19:16:03 cdent edleafe: of https://github.com/openstack/nova/blob/master/nova/objects/resource_provider.py#L2437-L2455
19:16:11 cdent basically reuse the objects use to create the data in the first place
19:16:34 cdent If people want structured objects...
19:16:51 edleafe artom: but I got the feeling that fields.Dict was not to be used anymore
19:16:54 edleafe artom: https://github.com/openstack/nova/blob/master/nova/objects/fields.py#L72
19:17:27 openstackgerrit Merged openstack/nova master: VMware: Handle missing volume vmdk during detach https://review.openstack.org/484675
19:17:30 artom edleafe, ah, true, didn't see the comment
19:17:35 artom dansmith's the expert
19:17:46 edleafe cdent: The object needs to have the hostname and the corresponding allocation
19:17:51 edleafe that's it
19:17:55 artom cdent, I think the idea behind objects is a poor man's type system
19:17:55 openstackgerrit Merged openstack/nova master: libvirt: Fix getting a wrong guest object https://review.openstack.org/496515
19:18:16 artom So that we don't have random unversioned unknown undiscoverable dicts flying around
19:18:25 openstackgerrit Merged openstack/nova master: Tests: Add cleanup of 'instances' directory https://review.openstack.org/491589
19:18:28 cdent artom: I know. Me and types have never felt all that friendly.
19:18:51 artom cdent, typist ;)
19:20:05 edleafe cdent: I was thinking about using AllocationList, but that has DB stuff. This should just be a blob to send back
19:20:43 cdent edleafe: the other thing to keep in mind is that we really don’t want to be using placement objects on the nova side of the equation if we can help it
19:20:53 cdent so a fresh object might be the way to go
19:20:55 edleafe cdent: yeah, that too
19:21:08 mriedem edleafe: https://github.com/openstack/nova/blob/master/nova/objects/fields.py#L72 doesn't mean you can't use them,
19:21:14 mriedem it means use the fields from ovo
19:21:15 mriedem rather than nova
19:21:16 edleafe but this allocations dict doesn't seem to fit any of the ovoo types
19:21:19 mriedem as we're moving stuff from nova to ova
19:21:21 mriedem *ovo
19:21:52 mriedem so you have an Allocation object
19:22:00 mriedem which has a resource_provider and resources field
19:22:02 cdent edleafe: so you’re idea at this point is an object that is effectively; ‘target’: String, ‘allocation’: Dict
19:22:29 cdent could be moultiple providers
19:22:45 edleafe cdent: pretty much
19:22:46 cdent is the idea to maintain the opacity of the allocation request bits?
19:23:16 edleafe IMO, the allocation dict is already too much placement internals to be passed around in Nova
19:23:24 edleafe But everyone else seems fine with it, so...
19:24:04 edleafe cdent: and the example I pasted contains multiple providers. It will only have one hostname, though
19:24:10 edleafe We're still in nova-land
19:24:21 cdent I was responding to mriedem on the rp thing
19:24:47 cdent I’m ambivalent about the allocation requests being passed around. sailed ship
19:25:17 mriedem passing allocation requests dicts around within the scheduler, as we are today, seems fine
19:25:33 mriedem if you want to pass something structured back over rpc to conductor to use for alternatives, you can define an object for that
19:25:43 mriedem to avoid some list of 2-item tuples
19:25:46 edleafe mriedem: that's what I'm trying to do
19:25:59 edleafe mriedem: my question was what field type to use for the set of allocations
19:26:18 edleafe We don't seem to have a matching type for that kind of nested dict structure
19:26:33 mriedem well, you don't want to end up with a ListOfDictOfLists field
19:27:02 edleafe DictOfListOfDicts?
19:27:30 mriedem for http://paste.openstack.org/show/619670/ we're not going to just pass a top-level dict with a single "allocations" key are we?
19:27:48 mriedem as the outermost object
19:27:55 mriedem pass a list of allocation objects
19:27:57 edleafe mriedem: yes. This is the body that is needed to claim/unclaim a host
19:27:58 artom Remove the top-level 'allocation' and just use a ListOfObjectsField?
19:28:07 mriedem artom: +1
19:28:22 artom I still think it depends on what you'll be doing with that
19:28:26 edleafe So the object will have to have a method for creating the POST body
19:28:36 artom As in, if the provider uuids are important at all, it could be a dict keyed on those
19:28:55 edleafe artom: this should be opaque. It is the body needed to claim/unclaim a selected host and its resources
19:29:20 mriedem creating an actual allocation request is pretty trivial https://review.openstack.org/#/c/496031/3/nova/conductor/tasks/live_migrate.py@122
19:29:22 artom Completely opaque? Just a string then?
19:29:34 mriedem we're doing that when bypassing the scheduler and forcing a host for live migration ^
19:29:35 cdent edleafe: should we fix this first: https://bugs.launchpad.net/nova/+bug/1708204
19:29:36 openstack Launchpad bug 1708204 in OpenStack Compute (nova) "placement allocation representation asymetric on PUT and GET" [Wishlist,Confirmed]
19:29:39 edleafe sure, we can chop this up any way that fits ovo, but then it has be able to put back together
19:30:18 edleafe mriedem: that won't be as easy with shared/nested providers
19:30:41 mriedem yeah i understand
19:31:00 mriedem and per the todo in that patch, i'd like to still call the scheduler even if we're forcing the host during live migration
19:31:06 edleafe the whole point of returning these allocation dicts was to handle those more complex situations
19:31:46 mriedem we already have an AllocationCandidate object today right?
19:32:12 mriedem sorry, it's called AllocationRequest
19:32:53 mriedem which has a list of AllocationRequestResource
19:33:09 edleafe mriedem: I guess I'll try following that pattern
19:33:20 mriedem why not just use the same exact objects?
19:33:30 mriedem if that's exactly what you want for the structure?
19:33:36 edleafe it's not
19:34:05 mriedem ok
19:34:11 edleafe I need a thing that is a hostname, and the blob needed to claim/unclaim the resources for the particular request
19:34:49 mriedem "(2:22:02 PM) cdent: edleafe: so you’re idea at this point is an object that is effectively; ‘target’: String, ‘allocation’: Dict"
19:35:04 edleafe mriedem: yep
19:35:08 mriedem and then return a list of those
19:35:19 mriedem AlternativeHost, and AlternativeHostList
19:35:21 mriedem or something
19:35:35 edleafe we can bikeshed on the name :)
19:35:56 mriedem and you're wondering if you need to break down the 'allocation' field into a more specific type?
19:36:05 edleafe what I was asking is: what ovo field type do I use for the allocation part of that?
19:36:21 edleafe I don't want to break it down if at all possible
19:36:27 edleafe It's a unit
19:36:39 edleafe (for a particular claim)
19:37:01 artom edleafe, so a single allocation (ie, resource_provider and resources)
19:37:02 artom ?
19:37:24 edleafe artom: well, it would be a set of allocations, but yes
19:37:28 mriedem edleafe: then it's just Dict
19:37:53 mriedem edleafe: and this is an allocation, yes?
19:37:54 mriedem { "resource_provider": { "uuid": "9cf544dd-f0d7-4152-a9b8-02a65804df09" }, "resources": { "VCPU": 2, "MEMORY_MB": 8096 } }
19:37:59 edleafe mriedem: even with dansmith's note at https://github.com/openstack/nova/blob/master/nova/objects/fields.py#L72 ?
19:38:12 mriedem edleafe: yes, as i said, that's about using the nova.fields vs ovo fields
19:38:24 mriedem versioned objects started in nova,
19:38:33 mriedem that note is saying, stuff is moving to ovo, so use the fields from there directly
19:38:56 mriedem https://github.com/openstack/oslo.versionedobjects/blob/master/oslo_versionedobjects/fields.py#L657

Earlier   Later