Earlier  
Posted Nick Remark
#openstack-nova - 2020-01-17
11:47:31 admin0 well, hav been trying to do kolla many times .. but still on osa
11:47:40 yoctozepto no problem
11:48:08 sean-k-mooney ok so you libvirt is running in an lxc container or on the host?
11:48:24 admin0 thing is if i ask some hard questions on kolla related to config or troubleshooting, there is no help on kolla but on osa its more in depth and instant .. which probably is because of rackspace
11:48:29 admin0 its on metal
11:48:56 sean-k-mooney ok can you try virsh --connect qemu:///system list
11:48:58 yoctozepto admin0: :O no help on kolla? how is that?
11:49:30 sean-k-mooney well i guess with --all
11:50:24 sean-k-mooney the xmls should be in /etc/libvirt/qemu
11:50:27 yoctozepto admin0: finally try: virsh define /etc/libvirt/qemu/that_one_file_you_missed_so_much.xml - maybe there was an error in there
11:51:09 admin0 error: invalid argument: could not find capabilities for arch=x86_64 -- wow .same server ..
11:51:44 yoctozepto see, now you got help from kolla, your argument is no longer valid (pun intended)
11:51:55 yoctozepto ;D
11:52:13 yoctozepto kidding aside, this really looks lolz
11:52:45 sean-k-mooney it looks like libvirt cannot find the x86_64 emulator
11:52:57 admin0 i will try kolla again .. won't doubt that .. but this is #nova channel :D
11:53:01 yoctozepto mhm, at least its config/definitions
11:53:22 yoctozepto admin0: yeah, I'm just teasing
11:53:36 yoctozepto though reboot does not get rid of those
11:53:38 sean-k-mooney well its complaining that virsh capabilities is not working
11:54:11 sean-k-mooney no a reboot would not cause this and nither would nuking the contolers
11:54:26 sean-k-mooney this host has been broken by something else that is unrelated
11:54:57 yoctozepto +1, I know the drill
11:54:59 admin0 in virsh capabilities i do see <arch>x86_64</arch> reported with the flags
11:55:05 yoctozepto some issues popping up only after reboot
11:55:20 yoctozepto as in "cleanup gone wrong and discovered weeks later"
11:55:40 yoctozepto odd
11:56:00 yoctozepto uhm
11:56:10 yoctozepto is kvm loaded by any chance?
11:56:18 yoctozepto both main and cpu-specific?
11:56:22 yoctozepto lsmod | grep kvm
11:56:53 yoctozepto admin0: ^
11:56:55 admin0 but this is a simulated "oops something bad happened" .. i rebooted the box , rebooted the vms multiple times before nuking the controllers and restart to make sure no skeletons in the closet pop up
11:57:09 admin0 i see kvm_intel, and kvm and irqbypass
11:57:15 yoctozepto hmm, odd
11:57:27 yoctozepto and what about
11:57:29 yoctozepto dmesg | grep kvm
11:57:37 admin0 null on that
11:57:54 yoctozepto then fine
11:58:05 sean-k-mooney admin0: this behavior is not related to the host reboot or the contoler unless you are useing shard stroageve form the contolers on the compute nodes
11:58:05 admin0 modprobe kvm && dmesg | grep kvm = also nothing
11:58:06 yoctozepto everything's all right but it does not work
11:58:19 yoctozepto nah, that should be nothing
11:58:20 sean-k-mooney e.g. you put your instnace data dir on an nfs share
11:58:25 admin0 this has no ceph and no cinder .. its local storage
11:58:34 admin0 i have the disks file locally
11:58:37 yoctozepto yeah, the error is that x86_64 missing
11:58:38 yoctozepto somehow
11:58:51 yoctozepto what distro is this anyways?
11:59:02 yoctozepto (right time to ask that heh)
11:59:04 admin0 ubuntu 16.04
11:59:06 sean-k-mooney admin0: i think you need to set up your experiment again clenly
11:59:11 yoctozepto quite an oldie there
11:59:46 admin0 ok.. then for the purpose of this, how do do virsh capabilities and edit hte nova file manually so that it just boots up, gets the IP and rsync, recovery is possible
11:59:47 sean-k-mooney the behavior you are seeing is not consitent with how nova operates
12:00:02 admin0 i undestand it may not boot in a very performant or optimal state, but the scope is recovery here
12:00:20 sean-k-mooney normally you would just do virsh start on the domains
12:00:25 sean-k-mooney no need to edit them
12:00:33 admin0 same would be if the virsh/os crashed and i manually had to copy the disk and xml to another (non-similar) compute node
12:00:41 yoctozepto admin0: is qemu-kvm installed?
12:01:01 admin0 qemu-kvm is already the newest version (1:2.10+dfsg-0ubuntu3.8~cloud1).
12:01:33 admin0 <type arch='x86_64' machine='pc-i440fx-artful'>hvm</type>
12:01:37 yoctozepto then I'm out of cards with this issue, really what comes to mind is some bug in ubuntu's libvirt that you have discovered
12:01:41 admin0 i guess that pc-* is the issue ?
12:01:56 yoctozepto well, you could try modifying that
12:02:31 yoctozepto pc-i440fx-xenial should be there
12:02:42 yoctozepto (at least that's what common sense suggests)
12:02:56 admin0 i see https://gist.github.com/a1git/f3696c126a732cda88757c09263bc2de in virsh capabilites related to x86_64
12:03:16 yoctozepto nothing suspicious
12:03:37 sean-k-mooney the machine type is normal
12:03:45 gibi efried: I got pulled downstream this morning so my review on the provider config progressing slowly. sorry
12:03:53 sean-k-mooney it will now show up in the cpu feature list
12:04:00 sean-k-mooney *not
12:04:15 sean-k-mooney its basicaly the "motherboard" for lack of a better name the vm uses
12:09:34 admin0 i edited pc-i440fx-artful to pc-i440fx-xenial .. same error as invalid argument: could not find capabilities for arch=x86_64 . i have the virsh capabilities in full , lsmod and the vm definition here => https://gist.github.com/a1git/4528c53090f8934b9218586817c51911
12:10:42 sean-k-mooney admin0: its not related to the model
12:11:10 admin0 ok .. but based on the output, there is no way this instance can boot up ?
12:11:12 sean-k-mooney admin0: you could set it to "pc" which is an aliase for the latest model supproted that is no q35
12:11:55 sean-k-mooney admin0: the machine types are not listed in that output
12:12:15 admin0 how do i list them ?
12:13:08 sean-k-mooney its likely because the x68 emulator is also not listed
12:16:15 sean-k-mooney you shoudl have a section like this
12:16:19 sean-k-mooney <guest>
12:16:20 sean-k-mooney <os_type>hvm</os_type>
12:16:22 sean-k-mooney <arch name='x86_64'>
12:16:24 sean-k-mooney <wordsize>64</wordsize>
12:16:26 sean-k-mooney <emulator>/usr/local/bin/qemu-system-x86_64</emulator>
12:16:28 sean-k-mooney <machine maxCpus='255'>pc-i440fx-4.1</machine>
12:16:30 sean-k-mooney <machine canonical='pc-i440fx-4.1' maxCpus='255'>pc</machine>
12:16:34 sean-k-mooney in my case it extends quite far
12:17:09 admin0 i will try to remove the libvirt-bin package and reinstall it
12:17:19 sean-k-mooney its not libvirt that is the issue
12:17:31 sean-k-mooney it cant find the x86 qemu bindary
12:17:36 stephenfin sean-k-mooney: Any decisions on what we want to do about https://github.com/openstack/nova/blob/master/nova/virt/libvirt/vif.py#L299-L301 ?
12:17:42 stephenfin That function doesn't exist
12:17:44 sean-k-mooney so you need to reinstall qemu
12:18:28 sean-k-mooney set_vif_host_backedn_brdige_config?
12:18:34 sean-k-mooney stephenfin: what is the context?
12:18:43 stephenfin yes
12:18:53 stephenfin it hasn't existed since commit 92323586b5d03b31c18657ad64646a8ce3b8a742

Earlier   Later