| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2020-08-18 | |||
| 13:14:31 | stephenfin | setattr | |
| 13:14:35 | stephenfin | gfdi | |
| 13:15:02 | stephenfin | sean-k-mooney++ Thanks. Try as I may, I couldn't find that /o\ | |
| 13:15:03 | sean-k-mooney | three are only 4 fileds you could replace that with 4 lines | |
| 13:15:30 | stephenfin | I think I might. setattr kills discoverability | |
| 13:16:24 | sean-k-mooney | if there were more filed i might argue for settattr but its just vnic_type, vif_type, profile and details | |
| 13:17:04 | sean-k-mooney | thats not going to change without a neutron api change and a new microversion which will require a nova code change to use anyway | |
| 13:20:11 | sean-k-mooney | stephenfin: if your going to refactor it might be nice to drop this too https://github.com/openstack/nova/blob/d4c857dfcb1ccfa5410de55671e69c722bbc990e/nova/conductor/tasks/live_migrate.py#L372-L379 | |
| 13:20:25 | stephenfin | agreed | |
| 13:26:18 | sean-k-mooney | lyarwood: sorry you pinged me eairlier but i didnt have irc up on my monitor this morning. did you figure it out? | |
| 13:26:53 | sean-k-mooney | oh im not connected downstream either i must not have connect to the vpn | |
| 13:41:28 | sean-k-mooney | stephenfin: https://review.opendev.org/#/c/728037/6 is in merge conflict so ill adress your nit and ping you to re review once done. | |
| 13:42:07 | sean-k-mooney | if im rebasing i might as well fix them now | |
| 13:44:24 | stephenfin | sean-k-mooney: cool | |
| 13:45:19 | stephenfin | sean-k-mooney: another question for you. The NeutronFixture isn't returning a vif_type key in the response to 'POST /v2.0/ports/{port_id}/bindings' | |
| 13:45:48 | stephenfin | Is that neutron's responsibility? Feel free to tell me to RTFM (I don't have an SR-IOV deployment to hand to experiment :() | |
| 13:46:20 | sean-k-mooney | yes the vif type can technicaly change on a migration but the vif type is set by the ml2 driver when it binds the port | |
| 13:46:37 | sean-k-mooney | stephenfin: can you show me the line | |
| 13:47:00 | stephenfin | This piece you linked earlier https://github.com/openstack/nova/blob/d4c857dfcb1ccfa5410de55671e69c722bbc990e/nova/conductor/tasks/live_migrate.py#L433-L436 | |
| 13:47:13 | sean-k-mooney | i ment the line in the fixture | |
| 13:47:22 | sean-k-mooney | i wanted to see what you have acess too | |
| 13:47:28 | stephenfin | Ohh | |
| 13:47:42 | stephenfin | Sec while I figure that out | |
| 13:47:55 | sean-k-mooney | you could copy it form the soruce vif | |
| 13:48:01 | sean-k-mooney | in the migrate vif objects | |
| 13:48:25 | sean-k-mooney | it can change but only if your migrating from like linux bridge hosts to ovs | |
| 13:48:47 | sean-k-mooney | for sriov it should not change although its techincially possible | |
| 13:48:52 | stephenfin | sean-k-mooney: https://github.com/openstack/nova/blob/master/nova/tests/fixtures.py#L1659-L1676 | |
| 13:49:01 | stephenfin | That's what the fixture is doing. It's rather dumb | |
| 13:49:06 | sean-k-mooney | if you went from say a hardware offloaded ovs to sriov host | |
| 13:49:24 | stephenfin | Okay, I have the port_id so I can use the stored details | |
| 13:49:26 | sean-k-mooney | ya it does jus tthe minium it needed too | |
| 13:49:38 | stephenfin | with a big note saying that stub doesn't allow you to switch backends | |
| 13:49:47 | stephenfin | if that sounds reasonable | |
| 13:49:56 | sean-k-mooney | self._port_bindings[port_id][host] | |
| 13:50:22 | sean-k-mooney | so you you likely can just look it up assuming [host] also has another entry for the souce node | |
| 13:51:01 | sean-k-mooney | so loop over self._port_bindings[port_id] and fine the one where the key is not euqal to host | |
| 13:51:18 | sean-k-mooney | then grab the info from that | |
| 13:51:38 | stephenfin | Good call | |
| 13:53:22 | sean-k-mooney | that wont always work but it should work in the migration case | |
| 13:54:25 | sean-k-mooney | so you will need to guard it with an if and only do it if the vif_type is not in the binding data i guess | |
| 13:56:22 | sean-k-mooney | actully it might work because reasons. | |
| 13:56:40 | sean-k-mooney | https://github.com/openstack/nova/blob/master/nova/tests/fixtures.py#L1573 that is where the vif_type is currently coming form in the fixture | |
| 13:57:29 | sean-k-mooney | well i guess htere are other places too | |
| 13:57:43 | sean-k-mooney | like here https://github.com/openstack/nova/blob/master/nova/tests/fixtures.py#L1357 | |
| 13:58:21 | sean-k-mooney | we build new ports form port 2 ^ https://github.com/openstack/nova/blob/master/nova/tests/fixtures.py#L1836-L1837 | |
| 13:59:36 | sean-k-mooney | the current sriov_port does not have the vif_type set | |
| 13:59:39 | sean-k-mooney | https://github.com/openstack/nova/blob/master/nova/tests/fixtures.py#L1448-L1469 | |
| 14:02:02 | stephenfin | turns out we don't currently set that self._port_bindings field for the first binding | |
| 14:02:06 | stephenfin | big TODOs from gibi | |
| 14:03:27 | sean-k-mooney | actully looking at the neutron code the VIF_TYPE for sriov depned on the vnic_type | |
| 14:03:28 | sean-k-mooney | https://github.com/openstack/neutron/blob/master/neutron/plugins/ml2/drivers/mech_sriov/mech_driver/mech_driver.py#L81-L85 | |
| 14:03:51 | luyao | stephenfin: I split the vpmem enhencement patch to three and polish them with Alex's help, I'll appreciate that you review it again if you have time, thanks. https://review.opendev.org/#/c/714653/ | |
| 14:04:14 | stephenfin | luyao: Thanks for the reminder. I'll try get to that this evening or tomorrow :) | |
| 14:04:34 | luyao | stephenfin: Thank you in advance. :) | |
| 14:04:44 | sean-k-mooney | stephenfin: so its "hostdev_physical" for PFs and "hw_veb" for everyting else | |
| 14:05:15 | sean-k-mooney | VIF_TYPE_MACVTAP = 'macvtap' is not used for sriov its for the macvtap backend | |
| 14:05:34 | sean-k-mooney | so you can ignore that | |
| 14:13:33 | openstackgerrit | sean mooney proposed openstack/os-vif master: update tox envs and support pdf docs https://review.opendev.org/728037 | |
| 14:35:00 | iurygregory | hey nova folks =) in ironic we are trying to create specific flavors (according to the boot_option that is set on capabilities) and I'm facing some problems with No valid host since nova filters doesn't match https://zuul.opendev.org/t/openstack/build/65e4cb538b7149d48f0942e8639b9bb3/log/controller/logs/screen-n-sch.txt#864 any ideas on how I can solve this? =) | |
| 14:35:44 | iurygregory | I have the feeling I need to do something on tempest to make it work, https://review.opendev.org/741445 this is the patch I have | |
| 14:36:06 | iurygregory | and this one https://review.opendev.org/741886 tests the changes from tempest | |
| 14:39:13 | stephenfin | sean-k-mooney: that's correct XML for an SR-IOV VIF, right? http://paste.openstack.org/show/796916/ | |
| 14:46:34 | sean-k-mooney | stephenfin: for vnic_type=direct yes that is the minimal xml although for a flat network | |
| 14:47:21 | sean-k-mooney | stephenfin: you can just create an instance of the config object and get it to render the xml if you want | |
| 14:48:56 | stephenfin | So how is this supposed to work? https://github.com/openstack/nova/blob/master/nova/virt/libvirt/migration.py#L378 | |
| 14:49:21 | stephenfin | interface_dev is the lxml.etree representation of http://paste.openstack.org/show/796916/ | |
| 14:50:10 | stephenfin | so '.find('address') should read '.find('source/address')' (or whatever the correct incantation is), no? | |
| 14:50:44 | sean-k-mooney | stephenfin: let me check the libvirt docs | |
| 14:51:34 | sean-k-mooney | <interface type='bridge'> | |
| 14:51:36 | sean-k-mooney | <source bridge='br1'/> | |
| 14:51:38 | sean-k-mooney | <target dev='vnet7'/> | |
| 14:51:40 | sean-k-mooney | <mac address="00:11:22:33:44:55"/> | |
| 14:51:42 | sean-k-mooney | </interface> | |
| 14:51:44 | sean-k-mooney | addrss is in the mac eleemnt for linux bridge too | |
| 14:53:31 | sean-k-mooney | oh that is proably looking up the device by pci adress? | |
| 14:53:40 | stephenfin | yeah, this is PCI address | |
| 14:53:49 | sean-k-mooney | the target pci adress? | |
| 14:54:00 | sean-k-mooney | something like this <address type='pci' domain='0' bus='0' slot='4' function='0' multifunction='on'/> | |
| 14:54:06 | stephenfin | yup | |
| 14:54:25 | stephenfin | The comment says "Save off the hw address and MTU presented to the guest since that can't change during live migration." | |
| 14:54:35 | stephenfin | but the place it's getting the hw address from seems wrong :-\ | |
| 14:54:54 | sean-k-mooney | the hardware adress normaly means mac address | |
| 14:55:17 | sean-k-mooney | interface_nodes = xml_doc.findall('./devices/interface') | |
| 14:55:28 | sean-k-mooney | so ya that looking for an address element in the interface | |
| 14:55:39 | sean-k-mooney | but they could be saving the guest pci address | |
| 14:55:53 | sean-k-mooney | this is populated by libvirt | |
| 14:56:00 | sean-k-mooney | let me check a vm quickly | |
| 14:56:13 | stephenfin | ah, so there are multiple address. Okay, that makes sense. | |
| 14:56:19 | stephenfin | *address elements | |
| 14:56:35 | sean-k-mooney | yes there is a top level one for the target adress to present the device in the guest | |
| 14:56:47 | sean-k-mooney | and then a sub elemen inside the source element which is the host address | |
| 14:57:02 | stephenfin | and parsing code for it here https://github.com/openstack/nova/blob/master/nova/virt/libvirt/config.py#L1891-L1893 | |
| 14:57:08 | stephenfin | so bad mocks again | |
| 14:57:39 | sean-k-mooney | http://paste.openstack.org/show/796917/ | |
| 14:57:45 | sean-k-mooney | that is what an ovs interface looks like | |
| 14:58:05 | sean-k-mooney | if needed i can stack my sriov host but it will looke more or less the same | |
| 14:58:36 | sean-k-mooney | its getting the element as line 11 | |
| 14:59:30 | stephenfin | sean-k-mooney++ ta | |
| 14:59:36 | stephenfin | definitely bad mocks so | |