| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2018-02-23 | |||
| 02:20:50 | openstackgerrit | Takashi NATSUME proposed openstack/nova stable/queens: [placement] Add sending global request ID in get https://review.openstack.org/543116 | |
| 02:34:53 | openstackgerrit | Merged openstack/nova stable/pike: Drop extra loop which modifies Cinder volume status https://review.openstack.org/546218 | |
| 03:36:33 | mnaser | i'm seeing more and more cinder snapshot failures in stable/pike | |
| 04:43:31 | alex_xu | mriedem, dansmith, working on eric patch now | |
| 05:21:56 | openstackgerrit | Tetsuro Nakamura proposed openstack/nova-specs master: Support shared/dedicated vCPUs in one instance https://review.openstack.org/545734 | |
| 05:26:26 | openstackgerrit | Nakanishi Tomotaka proposed openstack/nova master: Test Compute API in multiple cells https://review.openstack.org/547273 | |
| 05:34:00 | openstackgerrit | Tetsuro Nakamura proposed openstack/nova-specs master: Support shared/dedicated vCPUs in one instance https://review.openstack.org/545734 | |
| 06:11:18 | Spazmotic | Morning Novaers | |
| 06:16:54 | tetsuro | is cfriesen around? | |
| 06:17:05 | itlinux | hello nova team.. i have a question will it be better to set the hypervisor to kvm or just leave them to qemu? | |
| 06:17:14 | itlinux | thanks for the advice.. | |
| 06:19:39 | cfriesen | tetsuro: I'm here. | |
| 06:20:02 | tetsuro | itlinux: KVM has more functionality. Let me quote from the document https://docs.openstack.org/ocata/config-reference/compute/hypervisors.html “QEMU - Quick EMUlator, generally only used for development purposes.” | |
| 06:20:25 | itlinux | ok. | |
| 06:20:30 | tetsuro | cfriesen thanks for reply | |
| 06:20:35 | cfriesen | tetsuro: and I just added a bunch of comments to the review | |
| 06:20:40 | itlinux | thanks.. | |
| 06:20:51 | tetsuro | Yup, I didn’t understand your comment… | |
| 06:21:03 | tetsuro | so what is it from the placement view? | |
| 06:21:40 | tetsuro | alway 9 VCPUs in the inventory, right? | |
| 06:21:53 | cfriesen | I'm not an expert on placement, but I think so | |
| 06:23:27 | cfriesen | I'm not sure how it tracks allocated cpus though, whether it factors in cpu_allocation_ratio or not | |
| 06:23:58 | tetsuro | okay, if you want (8*1+1*(1/2))=9, who will exclude the host if you claim the third shared vcpu ? | |
| 06:24:26 | tetsuro | I mean (1/2)*2+(1/2)>1 | |
| 06:25:09 | tetsuro | where is the logic of “(1/2)*2+(1/2)>1”? | |
| 06:26:50 | cfriesen | in our internal patch we allocate "dedicated" pCPUs dynamically, so we would start with 9, allocate 8 dedicated ones (reporting a usage of 8), then allocate a "shared" one (and report a usage of 8.5) | |
| 06:26:51 | tetsuro | Maybe corefilter is extended and he is aware of shared cpus and shared dedicated cpus? | |
| 06:26:54 | tetsuro | oops | |
| 06:27:35 | cfriesen | then at that point CoreFilter would prevent you from allocating another dedicated pCPU since that would put usage at 9.5 which is larger than 9 | |
| 06:28:36 | cfriesen | but you could still allocate another shared pCPU since that would put usage right at 9 | |
| 06:29:54 | cfriesen | If we statically determine which pCPUs are shared/dedicated, then I think CoreFilter or the placement equivalent would need to be modified. | |
| 06:31:21 | tetsuro | Ah, so you mean your approach doesn’t need to pre-set something like “cpu_dedicate_set” ? | |
| 06:32:26 | itlinux | ok thanks tetsuro: my nova.conf says virt_type = kvm but my openstack hypervisor list show qemu type.. confused | |
| 06:32:30 | cfriesen | yes. anything in NUMACell.cpuset that is not in NUMACell.pinned_cpus is available to run "shared" vCPUs | |
| 06:32:46 | tetsuro | You are taking dynamically allocation approach that I put into alternative. | |
| 06:32:52 | tetsuro | Okay. | |
| 06:33:00 | cfriesen | yes, that's why I mentioned it originally | |
| 06:33:09 | tetsuro | cleared | |
| 06:34:34 | openstack | Launchpad bug 1195361 in OpenStack Compute (nova) "QEMU hypervisor type returned when libvirt_type = kvm" [Low,In progress] - Assigned to Tetsuro Nakamura (tetsuro0907) | |
| 06:34:34 | tetsuro | itlinux: that’s the default behavior https://bugs.launchpad.net/nova/+bug/1195361 | |
| 06:35:26 | cfriesen | the way we avoid needing to re-affine all the VMs when we allocate a new "dedicated" pCPU is that we put all the vCPU tasks into linux cpusets on the host. that way we can just reaffine the linux cpuset and all the VMs are affected | |
| 06:35:59 | cfriesen | this does mean we don't use the libvirt cpusets though, so I'm not sure that can be used as a generic solution for upstream | |
| 06:36:57 | cfriesen | with statically-allocated divisions between "shared" and "dedicated" pCPUs we might want to report two separate "cpus" and "cpus_used" numbers in the hypervisor details, one for "shared" and one for "dedicated" | |
| 06:39:29 | tetsuro | itlinux: because KVM only works with QEMU hypervisor, but agree this is confusing. We will fix this at least in Rocky realease. | |
| 06:39:56 | itlinux | thanks tesuro: I am in OOO | |
| 06:41:22 | tetsuro | your welcome | |
| 06:44:48 | tetsuro | cfriesen: Ah, I see. I think I got you. So you are not re-configure xml files but re-pinning them in linux layer. | |
| 06:47:26 | cfriesen | itlinux: the complication arises because both kvm and qemu end up calling the "qemu" binary as the hypervisor, but in the kvm case libvirt passes something like "-machine pc-i440fx-rhel7.3.0,accel=kvm" to tell it to use the hardware acceleration | |
| 06:47:39 | cfriesen | tetsuro: yes, that's correct | |
| 06:48:19 | itlinux | cfriesen: how can I verify that? | |
| 06:49:15 | cfriesen | from the compute node you can just look at the "qemu" commandline using "ps -ef" or similar | |
| 06:49:47 | tetsuro | cfriesen: Okay, I’ll add some words in the alternative and discuss with cores how they think. BTW, are you coming to PTG? | |
| 06:50:00 | itlinux | is that in the /etc/libvirt/qemu has a file I will chk that.. | |
| 06:50:06 | cfriesen | yes, I'll be there starting monday afternoon | |
| 06:50:28 | cfriesen | itlinux: no, when you have a guest up and running, then look at the commandline for the running process | |
| 06:50:41 | itlinux | looking now | |
| 06:51:27 | itlinux | http://paste.openstack.org/show/682875/ | |
| 06:51:39 | itlinux | looks correct to me but could you confirm it cfriesen: | |
| 06:51:42 | itlinux | thanks | |
| 06:52:05 | cfriesen | tetsuro: my concern about not reporting usage for "dedicated" pCPUs is that it'll make it tricky to debug scheduler failures if we don't have any way to see how "full" the compute nodes are | |
| 06:52:27 | cfriesen | itlinux: looks good to me | |
| 06:52:33 | itlinux | thanks | |
| 06:53:42 | tetsuro | I’m coming to Dublin, too. looking forward to see ya. :) | |
| 06:53:54 | cfriesen | likewise. I should get to bed, it's 1am | |
| 06:54:06 | tetsuro | OMG | |
| 06:54:19 | cfriesen | :) | |
| 06:54:20 | tetsuro | I got your concern, too. | |
| 06:54:34 | cfriesen | cool. should be a fun discussion | |
| 06:54:42 | cfriesen | later | |
| 06:55:00 | tetsuro | have a good night! I gotta go now, too. | |
| 07:40:50 | openstackgerrit | Andreas Jaeger proposed openstack/osc-placement master: DNM: Testing py35 https://review.openstack.org/547313 | |
| 08:23:08 | sahid | jaypipes: i saw you interest on some libvirt related specs, can you have look at these specs too: https://review.openstack.org/#/c/511188/ | |
| 08:23:12 | sahid | https://review.openstack.org/#/c/539605/ | |
| 08:24:00 | sahid | they address important use cases related to NFV | |
| 08:57:07 | bauzas | good morning Nova | |
| 08:57:16 | Spazmotic | Morning bauzas | |
| 08:57:38 | Spazmotic | Any cores that could pu this on their radar for this week? https://review.openstack.org/#/c/538415/ Been sitting for about 3 weeks now. | |
| 08:59:16 | Spazmotic | I am back from Korea btw, good to see you all again. | |
| 08:59:55 | bauzas | Spazmotic: most of the folks are preparing to travel for the PTG | |
| 09:00:08 | bauzas | maybe you could ask for reviewing after the next week ? | |
| 09:00:12 | Spazmotic | Yeah I know it, just hoping to get some eyes on it I guess | |
| 09:00:17 | bauzas | today, I'll look at specs | |
| 09:03:44 | Spazmotic | But asking is about all that I can do for now, and would remiss if I didn't do it :) | |
| 09:07:51 | bauzas | sure, just explaining | |
| 09:08:54 | Spazmotic | Yeah I understand of course :D | |
| 09:39:09 | openstackgerrit | Lajos Katona proposed openstack/nova master: WIP: ServerMovingTests with custom resources https://review.openstack.org/497399 | |
| 09:44:56 | ttsiouts | johnthetubaguy: Goodmorning, are you around? | |
| 09:48:35 | johnthetubaguy | ttsiouts: I am | |
| 09:48:52 | openstackgerrit | Arvind Nadendla proposed openstack/nova-specs master: Support traits in Glance https://review.openstack.org/541507 | |
| 09:51:12 | ttsiouts | johnthetubaguy: Hello! :) we were thinking about the second call to placement | |
| 09:51:38 | johnthetubaguy | ttsiouts: what was the plan before, just returning what you had previously fetched from placement? | |
| 09:52:40 | ttsiouts | johnthetubaguy: hmmm if the freeing of the resources is successful then we can call placement for a second time. | |
| 09:53:00 | ttsiouts | we chose to form the response in the service just to save time | |
| 09:53:15 | ttsiouts | and not having to trigger placement again.. | |
| 09:53:31 | johnthetubaguy | ttsiouts: I think that proxy is bad really, you don't know what microversion nova wants to ask for, etc | |
| 09:53:48 | johnthetubaguy | ttsiouts: simpler to tell Nova if you were successful or not | |
| 09:54:39 | ttsiouts | johnthetubaguy: Great. I'll do that. | |
| 09:55:14 | johnthetubaguy | ttsiouts: I think its worth doing a cheaky import of your client in there for now, or a link to the code at least, just so its possible to follow the breadcrumbs | |
| 09:56:04 | ttsiouts | johnthetubaguy: yes, seems better | |
| 09:56:44 | ttsiouts | johnthetubaguy: what do you think is better? an api call or importing the service? | |