| Posted | Nick | Remark | |
|---|---|---|---|
| #openstack-nova - 2019-05-17 | |||
| 13:39:12 | kashyap | sean-k-mooney: I'd guess so, from his test box | |
| 13:39:31 | sean-k-mooney | this is what i get on my desktop | |
| 13:39:32 | sean-k-mooney | http://paste.openstack.org/show/751512/ | |
| 13:39:34 | stephenfin | Hmm, Fedora doesn't come with libvirt out-of-the-box | |
| 13:39:55 | sean-k-mooney | running libvirt 5.2.0 | |
| 13:40:03 | stephenfin | sean-k-mooney: That's capabilities, not domain capabilities | |
| 13:40:16 | stephenfin | There's a difference in command and output but I'm trying to remember what it is | |
| 13:40:48 | sean-k-mooney | ah i have this for that http://paste.openstack.org/show/751514/ | |
| 13:41:17 | sean-k-mooney | can we ask them to add interface to that list | |
| 13:41:58 | sean-k-mooney | i get similar output on 4.0.0 | |
| 13:42:04 | stephenfin | Those are available under a subcommand, but I don't recall what that's called either | |
| 13:42:07 | sean-k-mooney | so i guess it been around for a while | |
| 13:42:40 | sean-k-mooney | its sudo virsh domcapabilities | |
| 13:42:55 | stephenfin | no, for the interface capabilities | |
| 13:43:12 | sean-k-mooney | oh well there are two things | |
| 13:43:23 | sean-k-mooney | there are the physical nic feature flags | |
| 13:43:37 | sean-k-mooney | and what virtual nics qemu can emulate | |
| 13:44:03 | sean-k-mooney | the nic feature flags are under the nodedev-dumpxml command | |
| 13:44:07 | kashyap | sean-k-mooney: It renders the same data without `sudo`, too. (With the ultra-minor diff being: with `sudo` it links to actual emulator path /usr/bin/qemu-system-x86_64; instead of the wrapper /usr/bin/qemu-kvm) | |
| 13:44:08 | stephenfin | ah, I was referring to the former in this case actually, my bad | |
| 13:44:20 | sean-k-mooney | the models that can be emulated are not available | |
| 13:44:22 | openstackgerrit | Guo Jingyu proposed openstack/nova master: Skip existing VMs when hosts apply force_config_drive https://review.opendev.org/659703 | |
| 13:44:46 | stephenfin | Remind me, what's the "is my system set up correctly for KVM" command? | |
| 13:45:02 | sean-k-mooney | kvm-ok | |
| 13:45:49 | stephenfin | ta | |
| 13:46:25 | sean-k-mooney | its part of the cpu-its part of the cpu-checker package if you dont have it installed | |
| 13:46:42 | stephenfin | ta x2 :) | |
| 13:47:26 | sean-k-mooney | its also only available on debian/ubuntu i think :) | |
| 13:47:48 | sean-k-mooney | so if you use somethin else all its doing is checking for /dev/kvm and checik lsmod for kvm | |
| 13:47:52 | stephenfin | So it would seem. Damn. I thought libvirt provided some tool for this but maybe I'm mistaking it for something | |
| 13:48:28 | sean-k-mooney | on and /proc/cpu-info for vmx or svm | |
| 13:48:49 | sean-k-mooney | you can tell form virsh capablityes | |
| 13:48:51 | stephenfin | Yeah, I went and did it manually instead. All good now | |
| 13:49:05 | sean-k-mooney | it list the emulator that can be run | |
| 13:49:21 | sean-k-mooney | if you dont se a hvm section then you cant use kvm | |
| 13:49:26 | stephenfin | I was seeing "error: invalid argument: unable to find any emulator to serve 'x86_64' architecture" because I hadn't installed qemu-kvm yet, assuming libvirt would do it by default | |
| 13:49:32 | stephenfin | *installing libvirt | |
| 13:50:10 | sean-k-mooney | or do sudo virsh capabilities | grep "<domain type='kvm'>" | |
| 13:50:41 | sean-k-mooney | stephenfin: libvirt does not install qemu by defualt | |
| 13:50:48 | stephenfin | In any case, I'm seeing something similar to you but I've got references to the OVMF stuff | |
| 13:50:53 | stephenfin | Yup, I know that now :) | |
| 13:51:03 | sean-k-mooney | and annoying installing qemu-kvm does not install qemu anymore... | |
| 13:51:50 | stephenfin | and I'm missing the virtio disk models, oddly | |
| 13:52:08 | sean-k-mooney | strange | |
| 13:52:35 | sean-k-mooney | you have installed qemu and qemu kvm packages and restarted libvirtd after | |
| 13:52:52 | sean-k-mooney | what os are you testing by the way | |
| 13:53:33 | stephenfin | Yeah. Not the end of the world though | |
| 13:53:35 | stephenfin | Fedora 30 | |
| 13:53:50 | stephenfin | Hold your "popOS!" comments now :P | |
| 13:53:51 | sean-k-mooney | stephenfin: they have been makign the qemu packages modular | |
| 13:54:06 | sean-k-mooney | have you tried gentoo :P | |
| 13:54:27 | stephenfin | No, I figured I should give Slackware a shot first | |
| 13:54:54 | sean-k-mooney | gentoo is the most painful disto i have used even more painful then linux form scratch | |
| 13:55:28 | sean-k-mooney | kashyap: is the virito support for qemu disk part of a sepereate package on fedroa 30? | |
| 13:55:47 | kashyap | sean-k-mooney: What do you mean? | |
| 13:55:50 | sean-k-mooney | kashyap: i woudl guess as that should be part of qemu by default | |
| 13:55:59 | kashyap | sean-k-mooney: Yeah | |
| 13:56:06 | kashyap | There has been a split of RPM packages for QEMU | |
| 13:56:10 | kashyap | Especially the block drivers | |
| 13:56:11 | sean-k-mooney | kashyap: stephenfin is not seing vitio in the domcapablities output on f30 | |
| 13:56:42 | stephenfin | I'm referring to lines 101-105 from here http://paste.openstack.org/show/751514/ | |
| 13:56:53 | stephenfin | It doesn't matter much. More curiosity than anything | |
| 13:56:57 | kashyap | stephenfin: Can you post your `rpm -qa | grep -i qemu` output? | |
| 13:57:21 | kashyap | sean-k-mooney: Check this one that i did recently: https://github.com/rdo-packages/nova-distgit/commit/4e490d2348cfe99b68d70c4b95408348d0b54492 (" Use granular libvirt and QEMU RPMs for EL-8 Nova") | |
| 13:57:27 | stephenfin | kashyap: http://paste.openstack.org/show/751515/ | |
| 13:57:31 | sean-k-mooney | stephenfin: well its novas default for qemu | |
| 13:57:36 | sean-k-mooney | and kvm | |
| 13:57:46 | sean-k-mooney | so if it really is not there your vms wont boot | |
| 13:58:02 | kashyap | stephenfin: Can you install 'qemu-kvm-core'? | |
| 13:58:05 | sean-k-mooney | unless you set hw_disk_bus=sata in the image or something | |
| 13:58:35 | sean-k-mooney | kashyap: waith how is qemu-kvm install without qemu-kvm-core | |
| 13:58:37 | kashyap | stephenfin: The domainCapabilities output you posted -- is that from F30? | |
| 13:58:42 | sean-k-mooney | that seam like a packageing bug | |
| 13:58:51 | stephenfin | kashyap: Done. No good. That paste is from sean-k-mooney | |
| 13:59:07 | kashyap | sean-k-mooney: Hang on -- I find it strange, too many things flying around :-) | |
| 13:59:40 | stephenfin | Here's mine, fwiw http://paste.openstack.org/show/751516/ | |
| 13:59:53 | stephenfin | 5.0.16-300.fc30.x86_64 | |
| 14:00:03 | sean-k-mooney | kashyap: :) that ws from manjor/arch with the latest libvirt available. | |
| 14:00:27 | sean-k-mooney | stephenfin: i see it on my popos system too with libvirt 4.0.0 so its not a libvirt issue i dont think | |
| 14:02:44 | roukoswarf | any recommendations on how to get nova scheduler to anti-affinity a bigger unit than a "host"? i have many "hosts" in a single chassis, or sometimes a single rack, and i need the ability to get something like octavia to schedule things to different chassis/racks for HA. | |
| 14:03:38 | roukoswarf | masses of cells seems like... the wrong solution? | |
| 14:06:46 | kashyap | sean-k-mooney: It is _not_ a packaging bug: in stephenfin's paste: you _do_ have: qemu-system-x86-core-3.1.0-8.fc30.x86_64 | |
| 14:06:56 | kashyap | It is 'qemu-system-$ARCH-core' | |
| 14:08:11 | kashyap | (Also note: 'qemu-kvm-core' is a _dummy_ package that pulls in 'qemu-system-x86' on Fedora) | |
| 14:11:20 | mriedem | efried: remind me, mock spec is a string right? https://review.opendev.org/#/c/659726/2/watcher/tests/datasources/test_ceilometer_helper.py | |
| 14:11:31 | kashyap | stephenfin: You need libvirt-5.2.0 or newer to get the virtio-* enum values reported. | |
| 14:11:44 | efried | mriedem: no, it is *not* a string. Unless you want to spec a string. | |
| 14:11:55 | mriedem | gdi | |
| 14:12:24 | efried | hold on, I have a commit that explains it clearly (I hope) | |
| 14:12:41 | kashyap | stephenfin: It's noted "v5.2.0 (2019-04-03) | |
| 14:12:49 | efried | mriedem: https://review.opendev.org/#/c/650370/ | |
| 14:13:02 | kashyap | stephenfin: Err. I meant: It's noted as part of the release notes of libvirt "v5.2.0 (2019-04-03)" here: https://libvirt.org/news.html | |
| 14:13:48 | mriedem | so i just changed to this (added autospec=True): | |
| 14:13:49 | mriedem | 'statistic_aggregation', autospec=True) | |
| 14:13:49 | mriedem | @mock.patch.object(ceilometer_helper.CeilometerHelper, | |
| 14:13:53 | mriedem | and it blew up as expected | |
| 14:13:58 | mriedem | so why not just autospec? | |
| 14:15:50 | mriedem | i think the answer is "sometimes it doesn't work how you'd expect" is the answer | |
| 14:15:55 | mriedem | like everything with specs and mock | |
| 14:25:29 | stephenfin | mriedem: claudiub has a series to convert everything to autospec. Would it be worth reviving that? | |