| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2017-07-04 | |||
| 13:55:02 | sean-k-mooney | jangutter: will the netronome solution work without modification to ovs. my understnading was that ye did not support the TC flower protocol | |
| 13:55:08 | alex_xu | bauzas: for testing the shared resource provider, jay's patch fake a mapping https://review.openstack.org/#/c/475448/24/nova/tests/functional/api/openstack/placement/fixtures.py@283 | |
| 13:55:33 | bauzas | alex_xu: sure, but that's just for the functional test, right?N | |
| 13:55:42 | alex_xu | bauzas: yea, just for functional test | |
| 13:55:58 | bauzas | tbc, just for a *fixture*, even not a test | |
| 13:56:07 | bauzas | meaning that we don't even assert something | |
| 13:56:30 | jangutter | sean-k-mooney: I believe TC-Flower support has just been submitted upstream. | |
| 13:56:59 | alex_xu | bauzas: emm... didn't get what means 'assert something' | |
| 13:57:16 | sean-k-mooney | jangutter: for the netronome smart driver in the kernel? | |
| 13:58:17 | jangutter | sean-k-mooney: correct, so if you use openvswitch.ko, kernel 4.13 and Open vSwitch 2.8, the card should just work... | |
| 13:58:42 | bauzas | alex_xu: the fact that jay pushed some fixture that is not exactly be in Nova | |
| 13:58:49 | bauzas | it's confusing | |
| 13:59:04 | bauzas | because for the moment, we don't really use that in our codetree | |
| 13:59:24 | sean-k-mooney | jangutter: ok then if we want to supprot that config in pike we need to merge the updates to os vif before the non client libariries frezze which is jul 17-21 | |
| 13:59:59 | alex_xu | bauzas: yea, but I think the 'MISC_SHARES_VIA_AGGREGATE' is something used by operator | |
| 14:00:06 | jangutter | sean-k-mooney: our PCI layout is a bit different from mellanox, and the sysfs interface isn't standardized, so that's why I'm trying to make sure we cooperate nicely. | |
| 14:00:38 | alex_xu | bauzas: do you mean 'MISC_SHARES_VIA_AGGREGATE' is not used in the our codetree? | |
| 14:00:43 | bauzas | alex_xu: even wit that, it's unclear | |
| 14:00:47 | bauzas | alex_xu: yeah | |
| 14:01:10 | sean-k-mooney | jangutter: yes from what i can tell this will be different for intel also be we wont have the required code in the kernel/ovs etc in time for pike. | |
| 14:01:11 | bauzas | alex_xu: Jay wrote some nice TDD, but it's not something we support it yet, right? | |
| 14:01:40 | bauzas | oh wait, I think about another point | |
| 14:02:02 | bauzas | if that's in os-traits and os-traits is synced, why should we need to create a new trait ? | |
| 14:02:17 | bauzas | the fixture probably forgets to sync the library, that's it | |
| 14:02:28 | alex_xu | bauzas: yes, we needn't create a new trait | |
| 14:02:28 | jangutter | sean-k-mooney: I should have clarity very soon about what we regard as stable. There's been some heated discussion on what the correct way is to resolve the VF's to the reprs. | |
| 14:02:45 | sean-k-mooney | jangutter: i had execpect to need to extend this in alter this in queens to supprot our nics. | |
| 14:03:23 | alex_xu | bauzas: but there is something strange in the gate http://logs.openstack.org/69/474869/5/check/gate-nova-tox-functional-ubuntu-xenial/308184e/testr_results.html.gz | |
| 14:03:37 | bauzas | alex_xu: so, 3.0.1 provides the trait https://github.com/openstack/os-traits/blob/0.3.1/os_traits/misc/__init__.py | |
| 14:03:39 | jangutter | sean-k-mooney: yep, it's always been in the back of my mind. I had thought of a support service that would persist VF settings across reboots. | |
| 14:04:09 | alex_xu | bauzas: edleafe have a fix https://review.openstack.org/478564, but looks like still have problem in the gate | |
| 14:04:09 | bauzas | and nova uses it https://github.com/openstack/nova/blob/master/requirements.txt#L56 | |
| 14:04:11 | jangutter | sean-k-mooney: maybe NetworkManager is the correct way to do it :-p | |
| 14:04:36 | sean-k-mooney | jangutter: except redhat dont use it... | |
| 14:05:09 | jangutter | sean-k-mooney: wha? RH wrote it! | |
| 14:05:53 | sean-k-mooney | jangutter: when you said networkmanager you were refering to /etc/network/interfeces and /etc/network/interfeces.d | |
| 14:06:10 | sean-k-mooney | jangutter: and the deamon that reads those files? | |
| 14:06:21 | bauzas | alex_xu: I can try to debut | |
| 14:06:23 | bauzas | debug | |
| 14:06:41 | bauzas | alex_xu: that's a transient bug, right? | |
| 14:07:15 | bauzas | ie. the gate is not blocked every time, but just having hits ? | |
| 14:07:29 | jangutter | sean-k-mooney: Yeah, the sysv initscripts and the thing that spawned from systemd. The problem with keeping state is proliferating yet another thing with another interface. | |
| 14:08:04 | sean-k-mooney | jangutter: but in either case no i think the configuration should be handled by os vif. eventurly this will get ported to the ovs windows datapath too so i would like to minimis external dependecies on linus | |
| 14:08:08 | sean-k-mooney | *linux | |
| 14:09:26 | alex_xu | bauzas: yes, looks like not very time | |
| 14:09:33 | jangutter | sean-k-mooney: aaah, yep. That's a very valid point. But my fever dream is that linux_net would call to nmcli or something if it exists, and that would persist VF settings, meaning that everything could come up asynchronously. | |
| 14:09:51 | sean-k-mooney | jangutter: for the intel intersect im actullly hoping to extend ovs to advertises the representor mapping in a table in the ovsdb that would be populated when you add the pf to the bridge | |
| 14:10:16 | jangutter | sean-k-mooney: yeah, that would be very useful. | |
| 14:10:19 | bauzas | alex_xu: I'm trying to reproduce the bug locally | |
| 14:10:37 | alex_xu | bauzas: if you lucky, you can see that locally :) | |
| 14:10:47 | bauzas | my guts is that this fixture doesn't correctly sync the lib | |
| 14:10:55 | openstackgerrit | Ghanshyam Mann proposed openstack/nova master: Correct the description of 'disable-log-reason' api-ref https://review.openstack.org/480209 | |
| 14:10:55 | sean-k-mooney | jangutter: i then want to have neutron read that table and populate a resouce provider for that in the placement api and have nova assign one of the representors as part of vm placement | |
| 14:11:13 | bauzas | and that, if you run other test before that functional test, you could be lucky because then you have that lib sync'd | |
| 14:11:33 | jangutter | sean-k-mooney: yep, that makes sense. Then ovs persists the VF settings. | |
| 14:11:35 | bauzas | alex_xu: so, that's probably just us being explicit in the fixture to load the lib | |
| 14:12:34 | jangutter | sean-k-mooney: on a completely tangential topic, is this the place to ask some really, really newbie questions regarding third-party os-vif plugins? | |
| 14:13:13 | sean-k-mooney | jangutter: sure as long as we are not boaring the rest of the nova guys too much | |
| 14:14:03 | sean-k-mooney | jangutter: what did you want to ask? | |
| 14:14:59 | jangutter | sean-k-mooney: so our legacy solution (non TC) will need a third-party OS-VIF and neutron plugin. Are there any docs to follow for publication and submission? | |
| 14:16:48 | sean-k-mooney | jangutter: well the simplest way to get started is ask infra to create a new repo, then clone clone the ovs plugin subtree and use that as the base for your plugin | |
| 14:17:44 | stephenfin | and it has. Wonderful. | |
| 14:18:13 | sean-k-mooney | stephenfin: is if permenet or a backup nick? | |
| 14:18:33 | stephenfin | sean-k-mooney: Probably permanent. I've been meaning to do it for ages | |
| 14:18:36 | jangutter | sean-k-mooney: so I've built some stuff in devstack using a local repo, that wouldn't impede our publishing? | |
| 14:19:02 | stephenfin | sean-k-mooney: Got an RedHat email alias set up, so I can move things to stephenfin@redhat.com too | |
| 14:19:49 | sean-k-mooney | jangutter: no it should not when you ask infra to create the repo you can use what you already have as a seed for the repo. alternitivly you just need to host is somewhare publicly yourself. | |
| 14:20:36 | sean-k-mooney | jangutter: the process to add a plugin is simply pip install it | |
| 14:20:58 | sean-k-mooney | jangutter: you nova patch has already merged correct? | |
| 14:21:52 | sean-k-mooney | stephenfin: i thaugh you were planning to move to stephen@that.guru instead? | |
| 14:22:01 | jangutter | sean-k-mooney: nope, but the Mellanox patch does 99% of what we need for the legacy ml2. | |
| 14:22:20 | stephenfin | sean-k-mooney: Possibly, but I think I've got my Gmail filters working now so it might not be necessary | |
| 14:22:23 | stephenfin | We'll see, sure | |
| 14:23:33 | jangutter | sean-k-mooney: we just need the VNIC_VIRTIO_FORWARDER added to neutron-lib | |
| 14:23:49 | sean-k-mooney | jangutter: currently because of how the nova ovs_utils wrapper works you can use external plugins in os-vif without modifying it. | |
| 14:24:35 | sean-k-mooney | jangutter: in your case you would be better off createing a fork of os-vif with your patch applied and pip installing that for your legacy ml2 dirver | |
| 14:25:02 | sean-k-mooney | jangutter: or if we can make it compatible and land it upstream even better | |
| 14:25:34 | gmann_ | stephenfin: its you :) | |
| 14:25:57 | openstackgerrit | Merged openstack/nova master: trivial: Use valid UUIDs in test_admin_password https://review.openstack.org/393711 | |
| 14:25:58 | stephenfin | gmann_: Yeah, I decided to start using the nick I use everywhere else | |
| 14:26:02 | jangutter | sean-k-mooney: The kinky blocker is that switchdev support is needed for the upstream ovs driver to work out-of-the-box with our legacy. | |
| 14:26:13 | gmann_ | stephenfin: with ref we can only show complete api file, but i want to provide the exact API link - https://review.openstack.org/#/c/467938/16/api-ref/source/os-quota-class-sets.inc | |
| 14:26:16 | stephenfin | I should probably keep the other one running with an /away message, but eh :) | |
| 14:26:35 | jangutter | sean-k-mooney: So for older kernels we need a fallback mechanism to do the VF-to-repr mapping. | |
| 14:26:35 | gmann_ | stephenfin: nice | |
| 14:27:39 | sean-k-mooney | jangutter: you would not happen to be useing vlan for that would you? | |
| 14:28:24 | openstackgerrit | Moshe Levi proposed openstack/nova master: hardware offload support for openvswitch https://review.openstack.org/398265 | |
| 14:28:27 | sean-k-mooney | jangutter: https://review.openstack.org/#/c/456481/1 this is the nove patch that is out standing for the legacy version correct | |
| 14:28:40 | stephenfin | gmann_: Could you just put an anchor into https://raw.githubusercontent.com/openstack/nova/master/api-ref/source/os-quota-sets.inc ? | |
| 14:28:48 | stephenfin | Assuming one isn't automatically inserted for you | |
| 14:29:26 | jangutter | sean-k-mooney: yep, that's the old stuff I'm refactoring now. | |
| 14:29:32 | openstackgerrit | Alex Szarka proposed openstack/nova master: Transform instance-live_migration_abort notification https://review.openstack.org/420453 | |
| 14:29:39 | stephenfin | e.g. above the 'List Default Quotas For Tenant' title, add '.. _list-default-quotas-for-tenant:' | |
| 14:29:56 | stephenfin | again, assuming ':ref:`list-default-quotas-for-tenant`' doesn't work already | |
| 14:29:59 | jangutter | sean-k-mooney: the external os-vif plugin will make that much cleaner. | |
| 14:30:25 | stephenfin | I know Sphinx can automatically insert references for headers, but I'm not sure if it's enabled for that build | |
| 14:30:56 | gmann_ | stephenfin: humm, ok. let me check otherwise ll add anchor | |
| 14:31:26 | stephenfin | gmann_: Do please. The advantage of doing this is that it always works while the link is valid, and the build fails if not | |
| 14:32:09 | jangutter | sean-k-mooney: https://review.openstack.org/#/c/456481/1/nova/network/neutronv2/api.py <--- this bit is obsoleted by https://review.openstack.org/#/c/398265/14/nova/network/neutronv2/api.py | |
| 14:32:40 | jangutter | sean-k-mooney: I think... | |