| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-02-04 | |||
| 14:34:37 | efried | bauzas: Yeah, here's the conversation with jaypipes (and cdent) earlier on: https://review.openstack.org/#/c/617042/14/nova/compute/manager.py@558 | |
| 14:35:06 | bauzas | okay, I don't want to hold for a nit if N = 1 | |
| 14:40:19 | efried | bauzas: Thank you sir. I added some more info to hopefully clarify why we get the rt's report client instead of creating a new one. | |
| 14:40:32 | bauzas | ack ok | |
| 14:41:11 | bauzas | efried: actually, I'm reviwing https://review.openstack.org/#/c/620711/18/nova/compute/manager.py | |
| 14:41:19 | bauzas | and all of that just disappears, so meh | |
| 14:41:51 | efried | more or less. We're still doing the same thing, getting the rc from the rt. We just don't need to be as careful about the getter anymore. | |
| 14:44:58 | bauzas | agreed, it's a femtonit | |
| 14:45:59 | bauzas | efried: except the fact that we uselessly need a dependency of the RT for calling the ReportClient but I guess we won't remove the RT dependency from the compute service soon :p | |
| 14:46:30 | efried | bauzas: Okay, but even if we did, we would then still have to do *something* to ensure we only had one reportclient. | |
| 14:46:38 | efried | else multiple caches getting out of sync, etc. | |
| 14:48:14 | efried | bauzas: I think I agree it would be prettier if we made the reportclient a global singleton. I can work on that in a fup if you think it's worthwhile. | |
| 14:49:29 | openstackgerrit | Rodolfo Alonso Hernandez proposed openstack/os-vif master: Add native implementation OVSDB API https://review.openstack.org/482226 | |
| 14:49:57 | bauzas | efried: nah, fine by me | |
| 14:50:09 | stephenfin | mriedem: Not sure if you saw this already, but I'd done the work for that BP before I saw your ping this morning https://review.openstack.org/634600 | |
| 14:50:15 | stephenfin | At least, I think that's most of it done | |
| 14:50:17 | efried | cdent, stephenfin: What's that sphinx syntax for referring to a doc in another repo? :doc-foo:`words </path/to>` or something? | |
| 14:51:26 | stephenfin | efried: https://docs.openstack.org/openstackdocstheme/latest/#external-link-helper | |
| 14:51:47 | efried | ah, the reverse, swhy I couldn't find it. Thanks stephenfin | |
| 14:51:52 | mriedem | stephenfin: ack i'll switch the blueprint to you | |
| 15:02:42 | efried | stephenfin: I thought there was a way to omit link text and it would automatically suck in the title of the target. Am I making that up? | |
| 15:17:58 | stephenfin | efried: Only if you're using referencing something from the same doc (i.e. the normal 'doc' role) | |
| 15:18:19 | mriedem | dansmith: on this per-instance serial number change, which now does what you asked for, https://review.openstack.org/#/c/619953/ - do you have any additional comments to what i already put in there? mine are mostly nits. kevin is out this week on holiday so if there are small changes i can make them. today is the last day for that one in a runway slot. | |
| 15:19:13 | dansmith | mriedem: let me get a refill and I'll have a look | |
| 15:19:54 | openstackgerrit | Rodolfo Alonso Hernandez proposed openstack/os-vif master: Add native implementation OVSDB API https://review.openstack.org/482226 | |
| 15:27:18 | efried | stephenfin: ack, thanks. | |
| 15:27:31 | efried | and, that makes sense; otherwise, how would it know? | |
| 15:33:31 | mriedem | dansmith: jaypipes: thanks on both, i'll work on follow ups this morning | |
| 15:34:00 | openstackgerrit | Jack Ding proposed openstack/nova master: Flavor extra spec and image properties validation https://review.openstack.org/620706 | |
| 15:46:10 | tinwood | dansmith, would it be possible, please, for you to take a quick look at: https://review.openstack.org/#/c/633279/ thanks. | |
| 15:52:48 | spatel | sean-k-mooney: morning! | |
| 15:53:36 | sean-k-mooney | spatel: o/ | |
| 15:53:53 | spatel | its been long time so thought "wave" | |
| 15:54:25 | sean-k-mooney | spatel: hopefully that means your cloud has been working to expectations | |
| 15:54:35 | spatel | i was busy in datacenter migration stuff.. but so far everything good | |
| 15:55:19 | spatel | My cloud is blooming...! +400 instance total and growing.. | |
| 15:55:20 | sean-k-mooney | spatel: see now you have doomed yourself by saying that. but glad to hear | |
| 15:56:18 | spatel | Soon planning to build new cloud.. not sure how much i can stuff in existing one. | |
| 15:56:23 | sean-k-mooney | it will be interesting to hear form you in 12-18 month as to if it proved more cost effective vs aws in the long run | |
| 15:57:31 | spatel | Believe me... its very cost effective... + my 50% datacenter is empty because i am stuffing application as much as possible in cloud | |
| 15:58:23 | sean-k-mooney | spatel: and you are achiving the perfomance KPI that are needed for your workload? | |
| 15:58:46 | sean-k-mooney | i assume you are till using sriov | |
| 15:58:53 | sean-k-mooney | *still | |
| 15:58:59 | spatel | interesting thing.. we are running real-time application in very latency sensitive environment :) got very good feedback from customer also :) | |
| 15:59:19 | spatel | Yes everything in sriov ( next cloud going to be on dpdk ) | |
| 16:00:04 | sean-k-mooney | well make sure to build a small test clust to verify its performacne before you make that switch | |
| 16:00:13 | spatel | we are providing close to HD audio quality with sr-iov | |
| 16:00:46 | spatel | ++ (will do lab test before go) | |
| 16:01:15 | sean-k-mooney | spatel: by the way you will be happy to know https://review.openstack.org/#/q/status:open+project:openstack/nova+branch:master+topic:bp/libvirt-neutron-sriov-livemigration is makeing good progress | |
| 16:01:57 | spatel | + boy... that would be my dream to do live-migration over SR-IOV | |
| 16:02:32 | openstackgerrit | Matt Riedemann proposed openstack/nova master: doc: link admin/configuration from admin home page https://review.openstack.org/634728 | |
| 16:02:34 | sean-k-mooney | spatel: be aware that for direct mode sriov there is still a network outage unless you create a bond in the guest | |
| 16:03:08 | spatel | bonding in guest is other complicated project :( | |
| 16:03:44 | sean-k-mooney | spatel: this is basically the first step in sriov migratoin. for macvtap sriov it will be complete. | |
| 16:04:25 | sean-k-mooney | for direct mode sriov there are some upcoming feature in the kernel/kvm and in vendor nics that may help | |
| 16:04:26 | spatel | macvtap isn't good, i did try and got packet drops during load-test | |
| 16:04:34 | sean-k-mooney | ya | |
| 16:05:17 | sean-k-mooney | the ida would be you would bound a direct mode and macvtap vf together and fall back to the macvtap during migration and then swithc back to direct mode vf | |
| 16:05:40 | spatel | oh!!! that is sweet! | |
| 16:06:24 | spatel | i don't mind small downtime during live-migration as far as it meet SLA | |
| 16:07:06 | spatel | Do you think it will be ready in stain release ? | |
| 16:07:42 | sean-k-mooney | openstack wont do the bonding for you but i hope we will merge the migration support in stein yes | |
| 16:08:01 | spatel | cool! | |
| 16:08:22 | sean-k-mooney | spatel: currently the code is complete so we are just waiting for a runway slot and reviewer feedback at this point. | |
| 16:09:02 | spatel | I don't have any lab at present otherwise i love to do some testing.. | |
| 16:09:58 | openstackgerrit | Matt Riedemann proposed openstack/nova master: doc: link Kashyap's cpu model talk to the libvirt driver config docs https://review.openstack.org/634731 | |
| 16:09:59 | mriedem | kashyap: ^ nice talk | |
| 16:10:39 | kashyap | mriedem: Hi, thanks! /me practised it 4 times to a rubber duck | |
| 16:10:51 | kashyap | On a call; meetings-- | |
| 16:11:19 | mriedem | kashyap: it showed that you practiced it, very smooth | |
| 16:11:25 | kashyap | BTW, on Sunday ran into The Tuba Guy at FOSDEM. He happened to be sitting right behind me | |
| 16:11:25 | mriedem | but you were soooo thirsty | |
| 16:11:42 | kashyap | mriedem: Darn, there is a story behind that water bottle trick | |
| 16:12:19 | kashyap | Back when I was in university, some 12 years ago, I had to give my first public talk on USB 3.0 or something... | |
| 16:12:35 | kashyap | And my throat completely dried out, and the teacher insulted me in front of the whole class: | |
| 16:12:43 | kashyap | Saying ... "What kind of a talk is that?!" | |
| 16:12:53 | kashyap | Never forget that experience | |
| 16:13:11 | kashyap | But more to the bottle: I use it as a trick to forcefully "pause" myself | |
| 16:13:15 | kashyap | And not drone on-and-on. | |
| 16:13:31 | kashyap | s/bottle/water bottle/ | |
| 16:26:23 | bauzas | ok, bailing out from upstream after 3 hours working on it \o | |
| 16:26:36 | bauzas | https://twitter.com/sylvainbauza/status/1092426500434022402 | |
| 16:27:01 | bauzas | at least I dedicated 3 hours for nova since 2 weeks... | |
| 16:28:01 | sean-k-mooney | bauzas: the vpn outage proably helped somewhat today | |
| 16:28:20 | bauzas | sean-k-mooney: not really, I planned to work for 3 hours anyway | |
| 16:29:07 | bauzas | I thought about this from the last weekend | |
| 16:29:19 | sean-k-mooney | i planned to start backpoting something and did some code review upstream instead | |
| 16:29:28 | sean-k-mooney | bauzas: its a good idea | |
| 16:29:33 | bauzas | and finally I think I very need to say "sorry but I need to work upstream *now*' | |
| 16:29:43 | artom | sean-k-mooney, to answer your downstream channel question, yeah, more work is required to actually generate the new XML | |
| 16:29:49 | bauzas | because if not, there will be some other priorities again and again | |
| 16:29:56 | artom | The claims are used to get the new instance NUMA topology | |
| 16:30:49 | sean-k-mooney | artom: are they actully used currently. i was not sure they were actully being used yet in the code you have pused. | |
| 16:31:06 | artom | sean-k-mooney, they're not used yet, no | |
| 16:31:32 | sean-k-mooney | artom: ok because the titile of https://review.openstack.org/#/c/634606/ consufed me a little | |
| 16:31:49 | artom | I mean, when creating the claim, the new instance NUMA topology is calculated, so it should throw a resources unavailable exception if it can't fit | |
| 16:32:03 | artom | But then that new topology isn't used anywhere (yet) | |
| 16:32:20 | sean-k-mooney | artom: ya that makes sense | |
| 16:32:35 | artom | sean-k-mooney, hrmm, yeah, maybe I'll rephrase it as "create claims" or something | |
| 16:32:44 | artom | Since as you correctly pointed out, they're not actually used | |