| *** gmmaha has quit IRC | 02:36 | |
| *** gmmaha has joined #kata-dev | 03:57 | |
| *** david-lyle has joined #kata-dev | 04:13 | |
| *** gmmaharaj has joined #kata-dev | 04:14 | |
| *** gmmaha has quit IRC | 04:15 | |
| *** gmmaharaj is now known as gmmaha | 04:15 | |
| *** dklyle has quit IRC | 04:16 | |
| *** sameo has joined #kata-dev | 04:52 | |
| *** kgz has quit IRC | 05:51 | |
| *** kgz has joined #kata-dev | 05:55 | |
| *** sgarzare has joined #kata-dev | 06:37 | |
| *** tmhoang has joined #kata-dev | 06:44 | |
| *** jodh has joined #kata-dev | 07:05 | |
| *** davidgiluk has joined #kata-dev | 08:01 | |
| *** Wimpress has quit IRC | 08:04 | |
| *** gwhaley has joined #kata-dev | 08:05 | |
| *** Wimpress has joined #kata-dev | 08:07 | |
| *** lpetrut has joined #kata-dev | 09:40 | |
| *** tmhoang has quit IRC | 11:17 | |
| *** tmhoang has joined #kata-dev | 11:30 | |
| kata-irc-bot | <graham.whaley> Quote of the day so far for me - from the Linux kernel Kconfig file for the virtio device drivers..... `So look out into your driveway. Do you have a flying car? If so, you can happily disable this option and virtio will not break.` | 12:52 |
|---|---|---|
| kata-irc-bot | <graham.whaley> funnily enough, I think we need to enable it, but only for certain circumstances (but we will enable it always) | 12:52 |
| gwhaley | hi stefanha (/cc davidgiluk) - I have a virtio question, wonder if you might know. It seems, from some debug etc., that the kata kernel needs CONFIG_VIRTIO_PCI_LEGACY=y set, but maybe only in the nested-VM case. Does that make any sense to you? My best guess, that the host (L1) VM sets up or passes something into the guest (L2), that does not happen for a bare metal (L0) deploy. https://github.com/kata-containers/packaging/pull/314#iss | 13:00 |
| gwhaley | uecomment-484069938 | 13:00 |
| davidgiluk | hmm that's ringing a bell, but why.... | 13:02 |
| davidgiluk | gwhaley: No, that's the opposite way from what I was thinking of; I don't think I know why that would happen | 13:04 |
| gwhaley | davidgiluk: heh, oh! Yeah, I never noticed as I always test on bare metal, but somebody found my kernel fragment config constructor did not work for them - and the key was a nested VM... but, it is hard to figure out why your kernel/agent are not chatting in kata when things don't boot :-( | 13:07 |
| davidgiluk | gwhaley: We do hit nesting bugs from time to time, but if anything the pci-legacy is the weirder case I think - so I don't know what's going on | 13:09 |
| stefanha | gwhaley: CONFIG_VIRTIO_PCI_LEGACY means that L1 QEMU is exposing a legacy VIRTIO device instead of VIRTIO 1.0. | 13:31 |
| stefanha | gwhaley: Are the QEMU command-lines in L1 different from L0? Is the set of virtio devices added to guests different? | 13:32 |
| stefanha | gwhaley: virtio-pci devices have disable-legacy=on|off|auto and disable-modern=on|off properties. | 13:33 |
| gwhaley | stefanha: that would indeed have had been my sort of guess. I'll have to check - the L1 is, in this instance, a 'ccloudvm' machine, so a qemu/kvm L1. Let me ask the ccloudvm man if he happens to know. I would have thought it would have just used a default host-like machine. | 13:34 |
| stefanha | gwhaley: If you have a QEMU HMP monitor, you can use 'info qtree' to see the device properties. That will show the value of disable-legacy and disable-modern. | 13:35 |
| gwhaley | ooh, I can grab the qemu cmdline - let me post it from my slack... | 13:35 |
| kata-irc-bot | <graham.whaley> This is the cmdline running the L1 I believe:qemu-system-x86_64 -qmp unix:/home/gwhaley/.ccloudvm/instances/considerate-lancelot/socket,server,nowait -m 6144M -smp cpus=4 -drive file=/home/gwhaley/.ccloudvm/instances/considerate-lancelot/image.qcow2,if=virtio,aio=threads,format=qcow2 -drive file=/home/gwhaley/.ccloudvm/instances/considerate-lancelot/config.iso,if=virtio,media=cdrom -daemonize -enable-kvm -cpu host -net | 13:36 |
| kata-irc-bot | nic,model=virtio -device virtio-rng-pci -net user,hostfwd=tcp:127.3.232.1:10022-:22,dnssearch=Home,hostname=considerate-lancelot -display none -vga none``` | 13:36 |
| stefanha | gwhaley: This command-line runs L1? That shouldn't affect the Kata kernel though, which runs in L2. | 13:39 |
| stefanha | gwhaley: The Kata kernel shouldn't be affected by any of this because its devices are emulated by the QEMU in L1. | 13:40 |
| gwhaley | stefanha: ah, you want me to check if a kata-on-baremetal qemu is commandline different from a kata-under-L1 bare metal - yeah, I can do that. Will take me a little bit, gotta check I have the bare metal one still set up the same... | 13:41 |
| stefanha | gwhaley: Also, do you trust the L1 environment - is it using an unpatched kata and QEMU/qemu-lite? | 13:43 |
| stefanha | This issue could occur if someone has tweaked the settings or used an ancient QEMU. | 13:44 |
| gwhaley | The two kata environments (bare and in-L1) are the same - I built them both and they are afaict SHA identical. | 13:44 |
| gwhaley | The L1 itself (ccloudvm) should be using afaik the sytem default qemu (F29 in my case). | 13:45 |
| gwhaley | having said that, one of my kata qemus was from packages and the other from the kata qemu source - but, we change that qemu so infrequently I suspect they are identical. If we find a cmdline difference, I can double check that. | 13:46 |
| stefanha | By default QEMU enables virtio modern and legacy. If the device is on an emulated PCIe bus, then only modern. | 13:47 |
| stefanha | The case you're hitting seems to be that only legacy works. | 13:47 |
| stefanha | Either disable-modern=on was set or some environmental issue is affecting it (I'm not sure). | 13:48 |
| stefanha | gwhaley: One thing that implicitly sets disable-modern=on is if you use an old QEMU machine type. | 13:48 |
| gwhaley | hmm, the only thing that rings a bell to me is we (at least used to) disable modern for one of the virtio devices iirc, as there was a bug in the kernel we were working around... | 13:49 |
| gwhaley | sameo: do you remember the virtion modern disable details from last year (I know, a lot of water flowed past since then for you). | 13:49 |
| stefanha | gwhaley: QEMU 2.6 and older didn't have modern. So pc-2.6 and earlier machine types disable it by default. | 13:50 |
| stefanha | gwhaley: Ah, that's probably the issue then. | 13:50 |
| gwhaley | stefanha: does this look relevant: https://github.com/kata-containers/runtime/blob/e15f3e4938a877b8e609366004703d76c2ed6261/virtcontainers/qemu_arch_base.go#L289 | 13:50 |
| gwhaley | reads to me as 'disable modern if we are nested' ! | 13:51 |
| stefanha | gwhaley: :-) | 13:51 |
| * gwhaley goes to 'blame'.. | 13:51 | |
| gwhaley | hah - we inherited that from the base virtcontainers import - it was something added in by sameo and mcastelino some time back to work around a bug... | 13:52 |
| gwhaley | OK, at least now I am happy I understand why I need to set the config for nested case :-) | 13:52 |
| gwhaley | I will nudge to see if that bug case is still valid | 13:52 |
| gwhaley | stefanha, many thanks :-) | 13:54 |
| stefanha | gwhaley: you're welcome | 13:55 |
| davidgiluk | oh well that's why I didn't think of it :-) | 14:17 |
| gwhaley | heh, tricky damned users eh davidgiluk :-) Yeah, once legacy/modern was mentioned then my archival brain started ringing little bells.... | 14:18 |
| *** lpetrut has quit IRC | 14:28 | |
| *** david-lyle is now known as dklyle | 14:35 | |
| *** marcov has joined #kata-dev | 14:49 | |
| *** marcov has quit IRC | 14:49 | |
| *** lpetrut has joined #kata-dev | 15:01 | |
| *** lpetrut has quit IRC | 15:42 | |
| *** tmhoang has quit IRC | 15:59 | |
| *** dims has quit IRC | 16:07 | |
| *** lpetrut has joined #kata-dev | 16:09 | |
| *** david-lyle has joined #kata-dev | 16:24 | |
| *** igordc has joined #kata-dev | 16:25 | |
| *** gmmaha has quit IRC | 16:27 | |
| *** dklyle has quit IRC | 16:27 | |
| *** gmmaha has joined #kata-dev | 16:27 | |
| *** EricRen has joined #kata-dev | 16:31 | |
| kata-irc-bot | <jose.carlos.venegas.m> release/CI question | 16:33 |
| kata-irc-bot | <jose.carlos.venegas.m> on 1.5 branch , I see a unit test failing on ARM CI | 16:33 |
| kata-irc-bot | <jose.carlos.venegas.m> ARM CI: ``` 09:25:17 --- FAIL: TestEnvGetHostInfo (0.00s) 09:25:17 assertions.go:239: Error Trace:kata-env_test.go:448 09:25:17 Error: Not equal: 09:25:17 expected: main.HostInfo{Kernel:"99.1", Architecture:"arm64", Distro:main.DistroInfo{Name:"Foo", Version:"42"}, CPU:main.CPUInfo{Vendor:"ARM Limited", Model:"v8"}, VMContainerCapable:true, | 16:33 |
| kata-irc-bot | SupportVSocks:false} 09:25:17 actual: main.HostInfo{Kernel:"99.1", Architecture:"arm64", Distro:main.DistroInfo{Name:"Foo", Version:"42"}, CPU:main.CPUInfo{Vendor:"ARM Limited", Model:"v8"}, VMContainerCapable:true, SupportVSocks:true} 09:25:17 09:25:17 Diff: 09:25:17 --- Expected 09:25:17 +++ Actual 09:25:17 | 16:33 |
| kata-irc-bot | @@ -12,3 +12,3 @@ 09:25:17 VMContainerCapable: (bool) true, 09:25:17 - SupportVSocks: (bool) false 09:25:17 + SupportVSocks: (bool) true 09:25:17 } ``` | 16:33 |
| kata-irc-bot | <jose.carlos.venegas.m> As it is not faling in master, could you point me to the fix to add it to the branch? | 16:34 |
| kata-irc-bot | <eric.ernst> just to be sure... did this ever pass on 1.5? | 16:35 |
| kata-irc-bot | <eric.ernst> do we know? | 16:35 |
| kata-irc-bot | <jose.carlos.venegas.m> Last release 1.5 for arm CI failed | 16:36 |
| kata-irc-bot | <jose.carlos.venegas.m> Oh intial 1.5 had not ARM CI | 16:38 |
| kata-irc-bot | <jose.carlos.venegas.m> https://github.com/kata-containers/runtime/pull/1156 | 16:38 |
| kata-irc-bot | <jose.carlos.venegas.m> ok safe to not try to fix it | 16:40 |
| kata-irc-bot | <jose.carlos.venegas.m> wont fix | 16:40 |
| kata-irc-bot | <eric.ernst> ok, sounds good :slightly_smiling_face: | 16:42 |
| *** eernst has joined #kata-dev | 16:44 | |
| *** eernst has quit IRC | 16:48 | |
| *** dims has joined #kata-dev | 16:54 | |
| *** dims has quit IRC | 16:59 | |
| *** jodh has quit IRC | 17:01 | |
| *** dims has joined #kata-dev | 17:02 | |
| kata-irc-bot | <graham.whaley> hi @fungi @cboylan @thierry - looking for a little input on some zuul setup if you can? Over on the kata zuul, I need to switch our PATH to include our now newly installed golang, which is in /usr/local/go/bin. Much though I want to just stick it in the role EXPORT PATH line at https://git.openstack.org/cgit/kata-containers/zuul-config/tree/roles/CI-setup/tasks/main.yaml#n21 - I have a feeling that will only be for that setup | 17:07 |
| kata-irc-bot | task, and will not persist back out to the main job that runs the actual CI tests. Any suggestions - should I set an `environment` setting somewhere 'higher up the tree' - out at the 'projects' level would maybe feel high enough up. thx! | 17:07 |
| *** gwhaley has quit IRC | 17:08 | |
| fungi | you can include environment variables in playbooks on a given host and it should apply to all tasks run for that host | 17:10 |
| fungi | use the "environment" key in a hosts declaration and then put an associative array of envvar names and values in there | 17:11 |
| fungi | i've found random examples of doing that, but having trouble hunting down the exact bit of ansible's documentation which shows it | 17:16 |
| *** igordc has quit IRC | 17:26 | |
| *** jugs has quit IRC | 17:36 | |
| *** jugs has joined #kata-dev | 17:37 | |
| *** igordc has joined #kata-dev | 17:44 | |
| *** sgarzare has quit IRC | 17:45 | |
| *** david-lyle is now known as dklyle | 17:55 | |
| *** lpetrut has quit IRC | 18:41 | |
| *** lpetrut has joined #kata-dev | 18:42 | |
| *** davidgiluk has quit IRC | 19:16 | |
| *** EricRen has quit IRC | 19:19 | |
| *** lpetrut has quit IRC | 19:40 | |
| kata-irc-bot | <eric.ernst> interesting : https://github.com/kubernetes/enhancements/blob/master/keps/sig-node/20190212-ephemeral-containers.md | 20:39 |
| kata-irc-bot | <raravena80> yeah, looks like runtimes will have specifically support them ^ | 20:42 |
| kata-irc-bot | <eric.ernst> I *think* it's just yet another 'startContainer' call | 20:42 |
| kata-irc-bot | <eric.ernst> but, if it is supported, it'd open up the potential lifecycle of a pod. | 20:43 |
| kata-irc-bot | <eric.ernst> ie, anyone could 'later on' add another container. | 20:43 |
| kata-irc-bot | <raravena80> yes, one use case would be if you'd need to sync some data into a pod local fs without restarting the pod (for example) There are probably several other use cases. | 20:45 |
| kata-irc-bot | <raravena80> oh there are some use cases in https://github.com/kubernetes/enhancements/blob/master/keps/sig-node/20190212-ephemeral-containers.md#user-stories :slightly_smiling_face: | 20:47 |
| *** sameo has quit IRC | 23:00 | |
| *** igordc has quit IRC | 23:21 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!