| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-07-13 | |||
| 12:06:52 | sean-k-mooney | the vif type shoud just be VIF_TYPE_VHOSTUSER | |
| 12:07:14 | jangutter | sean-k-mooney: I am using VIF_TYPE_VHOSTUSER for the actual plugging. | |
| 12:07:28 | sean-k-mooney | jangutter: you should also be using it for port binding | |
| 12:08:11 | sean-k-mooney | we are trying to stop using vif types as a flag for how to plug interfaces and retrun them to being a statemant about what type if interface you want | |
| 12:09:03 | sean-k-mooney | or in otherwords the vif_type is an abstion for the hyperviser interface type and is not related to how that interface is pluged to the network backend at all | |
| 12:10:06 | jangutter | sean-k-mooney: yeah. so one thing that kinda breaks the abstraction is that the "forwarder" is actually a kind of translation layer on top of the hypervisor. | |
| 12:10:29 | jangutter | sean-k-mooney: and introducing another layer just leads to madness. | |
| 12:10:29 | sean-k-mooney | jangutter: that is captured by the vnic_type | |
| 12:11:00 | jangutter | sean-k-mooney: yep. let me see if I can figure out if I'm doing the wrong thing in my ML2 driver. | |
| 12:12:50 | jangutter | sean-k-mooney: (facepalm) network_model.VIF_DETAILS_VHOSTUSER_OVS_PLUG | |
| 12:13:43 | sean-k-mooney | jangutter: ya you might need to set that | |
| 12:14:14 | sean-k-mooney | that is a legacy thing that should go away when we can specify the plugin to use in the vif details | |
| 12:17:19 | foutatoro | hi all, I would like to know how to deploy hypervisors VMware vSphere and Hyper-V on compute nodes ? | |
| 12:18:14 | openstackgerrit | OpenStack Proposal Bot proposed openstack/nova master: Imported Translations from Zanata https://review.openstack.org/477091 | |
| 12:18:41 | openstackgerrit | Béla Vancsics proposed openstack/nova master: Transform aggregate.update_metadata notification https://review.openstack.org/460625 | |
| 12:19:28 | openstackgerrit | Balazs Gibizer proposed openstack/nova master: use already loaded BDM in instance. |
|
| 12:30:54 | cdent | gibi: glad it was of some use, looking forward to seeing more of what you're doing | |
| 12:32:05 | stephenfin | sean-k-mooney, jaypipes: Fancy taking a look at this? Easy +1/2, imo https://review.openstack.org/#/c/483030/ | |
| 12:34:09 | toabctl | is there a way to list hosts which are not attached to a cell (v2) ? | |
| 12:38:48 | sean-k-mooney | stephenfin: sure, looking at it now | |
| 12:42:57 | stephenfin | sean-k-mooney: Cheers. Out of curiosity, is the expectation that all the old, non-os-vif plugging/unplugging code would eventually be replaced by code that harnesses vendor-developed os-vif plugins? | |
| 12:43:04 | stephenfin | e.g. All of these https://github.com/openstack/nova/blob/master/nova/virt/libvirt/vif.py#L599-L764 would be replaced by these https://github.com/openstack/nova/blob/master/nova/network/os_vif_util.py#L346-L403 | |
| 12:44:09 | openstackgerrit | Merged openstack/nova master: doc: Enable pep8 on doc generation code https://review.openstack.org/480469 | |
| 12:44:33 | openstackgerrit | OpenStack Proposal Bot proposed openstack/nova master: Updated from global requirements https://review.openstack.org/483337 | |
| 12:47:03 | sean-k-mooney | stephenfin: yes that was the original intention was to add an os-vif pluging for all the old types. modify neutron ml2 driver to pass plugin name in vif details and then eventurlly remove os_vif_util when we finally have neutron retrunin serialised os-vif objects from its api | |
| 12:50:54 | stephenfin | sean-k-mooney: OK, so regarding the comment here https://review.openstack.org/#/c/483030/2/nova/virt/libvirt/vif.py ... | |
| 12:52:13 | stephenfin | would we be submitting a bug against os-vif, or would that be a separate plugin maintained in its own repo by the floodlight team? | |
| 12:52:15 | sean-k-mooney | stephenfin: yep i was suggesting opening a bug against https://github.com/openstack/networking-bigswitch | |
| 12:52:31 | stephenfin | ahhh | |
| 12:54:12 | sean-k-mooney | stephenfin: IVS is a bigswitch networks competitor to ovs . it is the only thing that uses the IVS vif type https://github.com/openstack/networking-bigswitch/blob/master/networking_bigswitch/plugins/ml2/drivers/mech_bigswitch/driver.py#L422-L431 | |
| 12:54:57 | stephenfin | sean-k-mooney: Yeah, so they'd "own" the repo. It would be merged into 'openstack/os-vif'? | |
| 12:56:01 | sean-k-mooney | stephenfin: no its an non reference backend so im proposing that we wright an os-vif plugin for them and proposes it to there repo then deprecate the nova code and remove it once it is merged | |
| 12:56:32 | sean-k-mooney | stephenfin: alternitively we can create our only os-vif-ivs repo and that we own for this backend | |
| 12:57:18 | sean-k-mooney | it depend on whetere you want to tie the packaging of the plugin to the packaging of the ml2 driver which arguably you might want to keep seprate | |
| 13:00:35 | stephenfin | We'll let them make that decision, I guess | |
| 13:00:57 | stephenfin | sean-k-mooney: ...though IMO they're separate products so they should be. We'll see | |
| 13:01:00 | stephenfin | I'll file that bug now | |
| 13:02:18 | sean-k-mooney | am well if we have a seperate repo we dont need a bug on networking-bigswtich though we could use that bug to disucss this with them i guess? | |
| 13:05:33 | stephenfin | sean-k-mooney: Yeah, that's what I think. They will probably have their own ideas too | |
| 13:05:44 | stephenfin | If nothing else, they'll need to give us a repo to populate | |
| 13:06:03 | openstackgerrit | OpenStack Proposal Bot proposed openstack/nova master: Updated from global requirements https://review.openstack.org/483337 | |
| 13:16:43 | stephenfin | sean-k-mooney: https://bugs.launchpad.net/networking-bigswitch/+bug/1704129 | |
| 13:16:45 | openstack | Launchpad bug 1704129 in networking-bigswitch "Add an IVS os-vif plugin" [Undecided,New] | |
| 13:19:24 | openstackgerrit | Balazs Gibizer proposed openstack/nova master: use already loaded BDM in instance. |
|
| 13:22:22 | sean-k-mooney | stephenfin: cool i have subscribed to the bug so i should get pinged if they respond | |
| 13:24:32 | sean-k-mooney | stephenfin: the only thing i hate about the contrail plugin is that they do not use the setup.cfg to declare there entrypoitns and instead to it all in setup.py https://github.com/Juniper/contrail-nova-vif-driver/blob/master/setup.py | |
| 13:25:02 | stephenfin | sean-k-mooney: That's because they're not using pbr | |
| 13:25:08 | sean-k-mooney | stephenfin: but its a good example of how to do in correctly otherwise | |
| 13:25:37 | stephenfin | pbr had setuptools plugins that allow reading of things like that from 'setup.cfg'. By default though, setuptools doesn't support that | |
| 13:25:41 | stephenfin | *has | |
| 13:25:54 | stephenfin | sean-k-mooney: We could always draft a cookiecutter template if it would help people? | |
| 13:26:06 | sean-k-mooney | stephenfin: oh really . there is one | |
| 13:26:16 | sean-k-mooney | oh you mean for os-vif plugins? | |
| 13:26:40 | sean-k-mooney | stephenfin: the cookicutter template for new openstack project i think has not been updated for pbr yet | |
| 13:26:42 | stephenfin | Yup, a cookiecutter template for plugins | |
| 13:26:45 | sean-k-mooney | stephenfin: unless you did it | |
| 13:26:51 | stephenfin | I think I did | |
| 13:27:22 | sean-k-mooney | oh i have been meaning to aks you how to fix one of our repos to use pbr correctly | |
| 13:28:24 | stephenfin | sean-k-mooney: link? | |
| 13:28:49 | sean-k-mooney | stephenfin: https://github.com/openstack/devstack-plugin-libvirt-qemu | |
| 13:28:52 | gmann_ | mriedem: i replied on https://review.openstack.org/#/c/482994/1 | |
| 13:28:57 | openstackgerrit | Hironori Shiina proposed openstack/nova master: Ironic: Support boot from Cinder volume https://review.openstack.org/215385 | |
| 13:29:06 | sean-k-mooney | stephenfin: hum need to fix the readme too... | |
| 13:29:33 | gmann_ | mriedem: 'events' can be empty fro coding we have on controller side :) but very rare to rare timing things | |
| 13:29:42 | gmann_ | *from | |
| 13:29:57 | sean-k-mooney | stephenfin: we can gret rid of all the mail map stuff correct and there is some other boiler plate stuff that pbr will do for us too i think | |
| 13:29:58 | gmann_ | https://github.com/openstack/nova/blob/d955eb1db753f191acb02eb9758080329da5b866/nova/api/openstack/compute/instance_actions.py#L106 | |
| 13:31:11 | mriedem | gmann_: ok, but an empty events list is different from an event with no 'event.start_time' key | |
| 13:33:41 | stephenfin | sean-k-mooney: What's wrong with that project exactly? Looks a-ok to me? | |
| 13:33:56 | gmann_ | mriedem: is it. 'events' has direct 'start_time' so 'events' will be with no sub key | |
| 13:34:34 | gmann_ | mriedem: i commented the event record on compute service side and then we will have like this - http://paste.openstack.org/show/615266/ | |
| 13:35:01 | sean-k-mooney | stephenfin: it may be fine now. i know orginally after teh cookie cotter template it had the old change log and a few other things but rodolfo may have allready cleaned up everythin | |
| 13:35:40 | mriedem | gmann_: consider it like the os-extended-volumes:volumes_attached field in the server details response | |
| 13:35:58 | mriedem | that can be an empty list, which doesn't mean os-extended-volumes:volumes_attached.id is marked optional in the api ref | |
| 13:36:16 | mriedem | it means os-extended-volumes:volumes_attached will be in the response, but can be an empty list | |
| 13:36:25 | mriedem | And the description reflects that, " The attached volumes, if any." | |
| 13:36:49 | mriedem | now if you have volumes attached, the os-extended-volumes:volumes_attached.id and os-extended-volumes:volumes_attached.delete_on_termination keys will be in there, | |
| 13:36:51 | mriedem | and they are required | |
| 13:37:01 | mriedem | i'm saying it's the same thing for instance action events and the start_time field per event | |
| 13:37:10 | mriedem | *Except* the 'events' list is optional | |
| 13:37:16 | mriedem | which was what i changed in my patch | |
| 13:37:25 | mriedem | https://review.openstack.org/#/c/480792/ | |
| 13:41:27 | gibi | mriedem: hi! I proposed the first patch to use already in scope BDM for the notification payload as you suggested yesterday. Could take a brief look to see if it is a good directions? https://review.openstack.org/#/c/483324/ | |
| 13:42:42 | gmann_ | mriedem: i see your point. and from api-ref yes your point makes more sense to understand the api response. but from tempest response schema those should not be marked required | |
| 13:42:53 | mriedem | gibi: yeah maybe later | |
| 13:43:29 | mriedem | gmann_: but why? if there are events, and there is an event within the list of events, tempest should care that start_time is there | |
| 13:44:26 | mriedem | the keys for the event will be in the response, the values may be null if the event is not done, like finish_time and traceback | |
| 13:44:27 | mriedem | and result | |
| 13:44:38 | mriedem | required == you expect a key in the body dict, right? | |
| 13:44:49 | mriedem | required != that key must map to a non-None value | |
| 13:45:31 | gmann_ | mriedem: i am saying it can eb emtpy list means no start_time inside 'events' key | |
| 13:46:02 | gmann_ | mriedem: if api controller always insert those subkey even as null then i agree but that is not case | |
| 13:46:40 | gmann_ | because we insert 'events' key in response based on policy and insterts its subkey based on event we have in db or not | |
| 13:47:30 | gmann_ | so if later is not there we end up with 'events' with no sub key | |
| 13:48:36 | mriedem | gmann_: is jsonschema going to validate against an entry that isn't in the response? | |
| 13:48:52 | mriedem | we should have 2 schemas, | |
| 13:49:03 | mriedem | 1 for the instanceAction response, which has the 'events' key which is optional and is an array | |
| 13:49:27 | stephenfin | sean-k-mooney, ralonsoh: https://review.openstack.org/483398 | |
| 13:49:29 | mriedem | and 1 for the instanceAction.events, which is an object type, which has fields like start_time, finish_time, traceback, etc - those are required | |
| 13:49:34 | stephenfin | Ye silly gooses :) | |