| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-07-16 | |||
| 17:21:06 | sean-k-mooney | im wait on food to be delivered in 20 mins or so | |
| 17:21:18 | efried | I think it was your suggestion anyway from the subsequent patch | |
| 17:21:54 | sean-k-mooney | oh right. in the non ovs case we added a noop pluging to os vif to handel generic vhost-user plugging | |
| 17:22:07 | sean-k-mooney | and for the ovs case its already handeled by os-vif | |
| 17:23:09 | efried | yeah, see, I can read those words, but I've never been able to follow the code around that area | |
| 17:23:11 | sean-k-mooney | so those functions should never get called. but ill trace through the code and confim | |
| 17:23:16 | efried | thanks | |
| 17:24:27 | sean-k-mooney | well it used to do an indrect function call by appending the vif type to a prefix and using gettr to look up the function | |
| 17:24:39 | sean-k-mooney | so it was not exactly the simplest code | |
| 17:25:27 | sean-k-mooney | that actully predates me working on openstack but i never really understood why we did it that way | |
| 17:25:30 | efried | yeah, the getattr removal is in the subsequent patch, and I was able to follow that. But it's also dependent on this being right. | |
| 17:26:20 | sean-k-mooney | this code path is for netwroking-vpp or snabb switch | |
| 17:26:41 | sean-k-mooney | its very very rarly used but we still dont want to break people that use it | |
| 17:37:07 | openstackgerrit | Merged openstack/nova master: Defaults missing group_policy to 'none' https://review.opendev.org/657796 | |
| 17:53:27 | sean-k-mooney | efried: done. i have put a walk through of how the dead code would have been called in the past but it is never used after https://review.opendev.org/#/q/Ifab3006454708ab290b93f02d82b794c334c3946 was merged last cycle | |
| 17:53:44 | efried | thanks sean-k-mooney | |
| 17:54:31 | sean-k-mooney | no worries dinner just arrived so ill be away for a while. i might be back in an hour or two | |
| 18:01:06 | spatel | sean-k-mooney: yt? | |
| 18:01:49 | spatel | virsh list command showing instance-<uuid> is there anyway it can show hostname of instance | |
| 18:03:29 | spatel | In google people saying "virsh domhostname <domain>" but what is domain going to be? | |
| 18:03:51 | spatel | also i don't think kvm support that | |
| 18:20:18 | donnyd | Was there still interest in the NUMA for fortnebula? | |
| 18:20:49 | donnyd | LOL... interest in NUMA topologies | |
| 18:21:35 | donnyd | I have things to a more stable state with decent performance, so now I can move forward | |
| 18:25:31 | openstackgerrit | Mauricio Tavares proposed openstack/nova master: Increases the level of detail provided by the debug messages in pci_passthrough_filter.py. https://review.opendev.org/671106 | |
| 18:30:41 | efried | artom, sean-k-mooney: ---^ ? | |
| 18:39:58 | artom | donnyd, yes please, I would like 1 the NUMA ;) | |
| 18:40:24 | donnyd | LOL +2 | |
| 18:40:27 | artom | For serious though, yeah, flavors with >= 2 NUMA nodes would be awesome | |
| 18:40:34 | donnyd | you get 2 numas artom | |
| 18:41:26 | donnyd | I am not deep in the zuul or nodepool code, so how does the test work. Can you ask for a specific flavor? | |
| 18:41:53 | donnyd | I need a way to stay compatible with the other jobs | |
| 18:41:59 | artom | efried, I don't even know where to start with that. I guess look at the logic before drowning them with the formatting and whitespace nits? | |
| 18:42:24 | artom | donnyd, that's not my strong suite, I'm afraid | |
| 18:42:55 | efried | artom: I was pointing at donnyd, not at Mauricio's patch | |
| 18:43:06 | artom | efried, oh, hah | |
| 18:43:38 | efried | my ---^ was an attempt to go around the corner | |
| 18:44:16 | artom | donnyd, sean-k-mooney would most likely know, though I hate to pile even more things on this plate | |
| 18:44:31 | artom | Or ask in -infra? Is that good advice? | |
| 18:46:39 | donnyd | 2:44 PM <clarkb> we can do provider specific flavors but generally avoid that to handle cloud outages | |
| 18:46:39 | donnyd | 2:44 PM <clarkb> there isnt | |
| 19:04:21 | donnyd | artom: I will have to learn how to setup NUMA aware flavors, (unless someone knows off the top of their head) and it looks like we have a path forward | |
| 19:05:59 | artom | donnyd, it's kinda tedious, but not super hard: https://docs.openstack.org/nova/latest/user/flavors.html | |
| 19:06:05 | artom | Scroll down to CPU topology | |
| 19:06:24 | artom | Sorry, NUMA topology | |
| 19:06:34 | donnyd | Ok, I will take a look | |
| 19:07:08 | artom | Actually both | |
| 19:07:58 | artom | hw:cpu_ controls the guest topology, hw:numa_ controls how it maps onto host hardware | |
| 19:11:30 | openstackgerrit | Merged openstack/nova master: Use Adapter global_request_id kwarg https://review.opendev.org/670907 | |
| 19:58:44 | prometheanfire | is one of the use cases for nova to run 1-3 compute nodes in 3-4 DCs across the world? (via host aggregates and AZs) | |
| 19:59:02 | openstackgerrit | Eric Fried proposed openstack/nova master: DB API changes to get non-matching aggregates from metadata https://review.opendev.org/671074 | |
| 19:59:02 | openstackgerrit | Eric Fried proposed openstack/nova master: Nova object changes for forbidden aggregates request filter https://review.opendev.org/671072 | |
| 19:59:03 | openstackgerrit | Eric Fried proposed openstack/nova master: Enable request filter forbidden_aggregates https://review.opendev.org/667952 | |
| 19:59:03 | openstackgerrit | Eric Fried proposed openstack/nova master: Add a new request filter to forbidden aggregates https://review.opendev.org/671075 | |
| 20:00:10 | efried | prometheanfire: What's this in reference to? | |
| 20:00:53 | prometheanfire | sometihng I may be doing soon | |
| 20:01:09 | prometheanfire | cross continental rabbit could be fun | |
| 20:01:34 | prometheanfire | just not sure if it's an intended (supported) use case for nova | |
| 20:03:34 | openstackgerrit | Merged openstack/nova master: vif: Remove 'plug_vhostuser', 'unplug_vhostuser' https://review.opendev.org/670588 | |
| 20:06:09 | efried | ohh, okay. So you're asking if you can make a setup like that? | |
| 20:06:22 | efried | prometheanfire: in a single cell? | |
| 20:06:39 | efried | Cause I would imagine you can surely do a multi-cell that way | |
| 20:06:50 | efried | donnyd: do you know about this? | |
| 20:07:19 | prometheanfire | ya, could do it multi-cell I suppose, single cell may be nice management wise though | |
| 20:07:48 | efried | prometheanfire: isn't that how zuul is set up? | |
| 20:07:59 | prometheanfire | the main question I have is if I'll need to run non-compute stuff at each location or not (I'd rather it just be nova-compute and neutron agent type stuff) | |
| 20:08:00 | donnyd | well if you look at any of the big clouds out there, it would appear they try to stay away from this type of thing | |
| 20:08:01 | efried | running a nodepool across a dozen different locations | |
| 20:08:46 | donnyd | Control plane / internal stuff usually stays on the same l2//3 networks | |
| 20:08:54 | efried | You might have to run a cell conductor at each site. I would think your db performance would be crap if you didn't. | |
| 20:09:03 | donnyd | But I am not entirely sure what the ask is | |
| 20:09:56 | prometheanfire | how to best build a cloud that spans multiple DCs, but it still small beans (a couple hosts per location) | |
| 20:10:32 | donnyd | What is the plan for N/S E/W traffic... that has been my big hurdle in the past... | |
| 20:10:44 | artom | prometheanfire, ooooo, are you being edgy? It's all the rage these days ; | |
| 20:10:45 | artom | ;) | |
| 20:10:51 | donnyd | wanted to host my control plane in someone elses cloud, and then just have hypervisors and storage here | |
| 20:11:00 | prometheanfire | artom: kinda, but not quite that scale | |
| 20:12:11 | prometheanfire | donnyd: as far as traffic, we'll have site to sites, non-dedicated. the E/W should be minimal traffic wise (between DCs at least) | |
| 20:15:52 | donnyd | I don't see any reason why it cannot be done if you already have S2S connections to secure traffic.. Gonna spend quite a lot of time having fun with rabbit... | |
| 20:16:10 | prometheanfire | yep, I dread rabbit | |
| 20:17:46 | donnyd | and def be using DVR | |
| 20:18:41 | prometheanfire | I'm not sure we'd use more than linuxbridge and maybe gre (or just vlan) | |
| 20:18:53 | prometheanfire | kiss | |
| 20:20:19 | donnyd | The way its supposed to work is the same l2 between l3 agent and the vm | |
| 20:20:46 | prometheanfire | yep | |
| 20:20:51 | prometheanfire | that's possible | |
| 20:21:45 | donnyd | you would want to use DVR because the L3 agent will be over a wan connection, and all your broadcast traffic would also be doing the same | |
| 20:22:12 | prometheanfire | ya, that'd be rough | |
| 20:22:31 | prometheanfire | at least this isn't 4 years ago, dvr is better now I hear | |
| 20:23:16 | prometheanfire | https://wiki.openstack.org/wiki/Neutron/DVR still mentions the road to juno :D | |
| 20:23:30 | prometheanfire | first result when googling for 'neutron dvr' | |
| 20:25:59 | donnyd | I am not surprised | |
| 20:26:22 | donnyd | juno was only a couple releases back | |
| 20:28:20 | prometheanfire | feels like | |
| 20:28:31 | prometheanfire | 9 releases? | |
| 20:47:09 | openstackgerrit | Merged openstack/nova master: vif: Stop using getattr for VIF lookups https://review.opendev.org/666630 | |
| 21:47:56 | openstackgerrit | melanie witt proposed openstack/nova master: nova-manage db archive_deleted_rows is not multi-cell aware https://review.opendev.org/507486 | |
| 22:26:26 | Nick_A | we have snapshots set to qcow2 in nova.conf. I noticed they take 2x disk usage until uploaded to glance. Is it trying to convert qcow2 > qcow2 or something? | |
| #openstack-nova - 2019-07-17 | |||
| 02:23:10 | openstackgerrit | Yongli He proposed openstack/nova master: Clean up orphan instances virt driver https://review.opendev.org/648912 | |
| 02:23:11 | openstackgerrit | Yongli He proposed openstack/nova master: clean up orphan instances https://review.opendev.org/627765 | |