| kata-irc-bot | <fupan> kata-shimv2 will do the same with cli to find the default config files from /etc and /usr/share. | 01:38 |
|---|---|---|
| kata-irc-bot | <fupan> yes, by now it doesn’t support passing --config option | 01:39 |
| kata-irc-bot | <fupan> @eric.ernst @xu | 01:42 |
| kata-irc-bot | <eric.ernst> I think adding support for --config will help a lot | 02:15 |
| kata-irc-bot | <eric.ernst> with this, the same binary can be used with QEMU and FC | 02:15 |
| kata-irc-bot | <eric.ernst> I think with CLI you can also set the path at compile time | 02:16 |
| kata-irc-bot | <eric.ernst> thx for clarifying @fupan | 02:16 |
| kata-irc-bot | <xu> I have two points: - shimv2 should try more path, such as /usr/local/share - shimv2 should be configured through the cri configuration fields of containerd I am not sure if the `--config` is essential | 02:18 |
| kata-irc-bot | <fupan> Since the shimv2's cmd option are parsed in the shim pkg located in containerd, if we want support it, we have to communicate with containerd guys. | 02:20 |
| kata-irc-bot | <xu> at least you may try more path and does parsing the environment vars work? | 02:22 |
| kata-irc-bot | <fupan> Hmm, it depends, if we want to fix the issue of https://github.com/kata-containers/runtime/issues/1082#issuecomment-451303997, I think it’s better to let k8s/containerd pass a parameter to shimv2. Thus the pod can be assigned to specific hypervisor. | 02:34 |
| *** lsm5 has joined #kata-dev | 05:18 | |
| *** lpetrut has joined #kata-dev | 08:15 | |
| *** jodh has joined #kata-dev | 08:35 | |
| *** davidgiluk has joined #kata-dev | 09:01 | |
| *** lpetrut has quit IRC | 09:18 | |
| *** stackedsax has quit IRC | 11:55 | |
| kata-irc-bot | <eric.ernst> @bergwolf - I have you listed as doing release next week - is that okay? And, @zhangwei555 you have interest in jumping in this rotation? | 15:53 |
| kata-irc-bot | <eric.ernst> Or if anyone else has any interest (@mvedovati, @sebastien.boeuf , others?). The wiki is located @ https://github.com/kata-containers/community/wiki/Release-Team-Rota. | 15:54 |
| *** LinuxMe has joined #kata-dev | 15:57 | |
| kata-irc-bot | <eric.ernst> @salvador.fuentes: when do you think it makes sense to switch default of testing runtimeClass? | 16:12 |
| kata-irc-bot | <eric.ernst> and/or, is there a job in place that tests it already? | 16:12 |
| kata-irc-bot | <xu> the two guys in China might be asleep right now :) | 16:13 |
| kata-irc-bot | <mnaser> does anyone know if nested virt works up until 3 layers? | 16:15 |
| kata-irc-bot | <mnaser> in order to test magnum support for kata, we run our CI inside VMS with nested virt | 16:15 |
| kata-irc-bot | <mnaser> so a VM deploys openstack with magnum, creates a cluster (which creates VMs inside the VM) and then if we want to add kata support, it will mean that there will be a third level of VMs... | 16:16 |
| kata-irc-bot | <eric.ernst> *should*? | 16:16 |
| kata-irc-bot | <eric.ernst> I mean, it should work up until you get too confused and fall over :slightly_smiling_face: | 16:17 |
| kata-irc-bot | <eric.ernst> 3, 4, 5... | 16:17 |
| kata-irc-bot | <eric.ernst> right? | 16:17 |
| kata-irc-bot | <eric.ernst> Are you seeing issues? | 16:17 |
| kata-irc-bot | <xu> sounds evil | 16:17 |
| kata-irc-bot | <eric.ernst> things like device pass-through won't | 16:17 |
| davidgiluk | well, 3 level passthrough will be emulation - so that will be slow as a slow thing | 16:18 |
| davidgiluk | sorry , 3 level (not 3 level passthrough) | 16:18 |
| davidgiluk | and not very well tested | 16:19 |
| kata-irc-bot | <eric.ernst> maybe i'm lame - i'm not sure how big of a delta L2 v L3. (thinks through it...) | 16:19 |
| davidgiluk | I'd expect the L2 to be usable and the L3 to be painful as hell | 16:20 |
| kata-irc-bot | <eric.ernst> I'll wait to hear @mnaser's story first before commenting :slightly_smiling_face: | 16:20 |
| kata-irc-bot | <mnaser> oh yeah, i mean i haven't tried it yet | 16:21 |
| kata-irc-bot | <eric.ernst> Either way it has to drop down to L0 via n-1 anyway; i'm not sure how it'd be different. | 16:21 |
| kata-irc-bot | <eric.ernst> once you are on a pile of turtles... | 16:21 |
| kata-irc-bot | <mnaser> but i'm thinking if we want to get Magnum support with proper CI, it might be problematic | 16:21 |
| kata-irc-bot | <mnaser> so maybe i'm wondering if it will require baremetal | 16:21 |
| kata-irc-bot | <eric.ernst> maybe @manohar.r.castelino is around to comment? | 16:22 |
| kata-irc-bot | <eric.ernst> or @jonolson | 16:22 |
| kata-irc-bot | <eric.ernst> Also, I'm looking @ https://github.com/kata-containers/tests/blob/master/.ci/install_cri_containerd.sh#L44-L64, and wondering - we just grabbed this from one of the installs of containerd or from a repo, or? Just want to make sure we track if this changes. | 16:25 |
| kata-irc-bot | <manohar.r.castelino> @eric.ernst driving. Be online in about 20 | 16:25 |
| kata-irc-bot | <eric.ernst> i think that's more dangerous than a pile of turtles. | 16:25 |
| kata-irc-bot | <eric.ernst> also debating merits of building versus using package from containerd directly. | 16:26 |
| kata-irc-bot | <jonolson> Nested virt works to arbitrary depth, in theory | 16:26 |
| kata-irc-bot | <jonolson> You end up with a stack of shadow VMCS objects, so slowdown depends on how many layers your exit traverses before being handled | 16:27 |
| kata-irc-bot | <jonolson> (sorry, still coming up to speed on conversation here) | 16:28 |
| kata-irc-bot | <eric.ernst> I'm just trying to recall nested story. I thought there were enhancements such that you don't traverse each layer, that it'd drop back to L0 | 16:29 |
| kata-irc-bot | <eric.ernst> Now I wish I didn't go cross-eyed during the L7 debug example at kvm-forum :slightly_smiling_face: | 16:29 |
| kata-irc-bot | <eric.ernst> @gabriela.cervantes.te ^ you know? | 16:30 |
| kata-irc-bot | <mnaser> yeah so i don't know if it would be a reliable CI target | 16:32 |
| kata-irc-bot | <jonolson> Exits themselves always drop back to L0, but that may re-enter into an arbitrary level depending on how the nested guests implemented their nested VMX instruction sets -- I would not trust it as a stable CI target -- you'd be better off doing emulation (slow, but mostly reliable) | 16:35 |
| kata-irc-bot | <eric.ernst> @mnaser there was a good presentation by oracle, who uses L3 a decent amount, at kvm-forum. let me find the summary.. | 16:35 |
| kata-irc-bot | <jonolson> Why... Why do they do that? | 16:36 |
| kata-irc-bot | <eric.ernst> that's the summary i'm looking for :slightly_smiling_face: | 16:36 |
| kata-irc-bot | <jonolson> Monsters. | 16:36 |
| kata-irc-bot | <eric.ernst> ``` Therefore, setup is: ○ L0 = Public cloud provider hypervisor ○ L1 = Ravello’s hypervisor (KVM based) ○ L2 = Ravello guest which is a hypervisor (e.g. ESXi) ○ L3 = L2 guests ``` | 16:36 |
| kata-irc-bot | <mnaser> how slow is kata in qemu to run 1 or 2 tests | 16:37 |
| davidgiluk | huh ok, that's..... | 16:37 |
| kata-irc-bot | <mnaser> i dunno why L2 is there | 16:37 |
| kata-irc-bot | <mnaser> seems like you can just skip that part | 16:37 |
| kata-irc-bot | <mnaser> but hey | 16:37 |
| kata-irc-bot | <mnaser> i'm the one here suggesting the same idea so who am i to talk :slightly_smiling_face: | 16:37 |
| kata-irc-bot | <eric.ernst> The deck. Too long - couldn't find summary aside from "it works, perf suffers" | 16:39 |
| kata-irc-bot | <eric.ernst> https://events.linuxfoundation.org/wp-content/uploads/2017/12/Improving-KVM-x86-Nested-Virtualization-Liran-Alon-Oracle.pdf | 16:39 |
| kata-irc-bot | <krsna1729> Ravello was a company acquired by oracle. i think it used to offer a way to setup labs and testing openstack installations on public cloud | 16:45 |
| davidgiluk | yeh | 16:46 |
| kata-irc-bot | <gabriela.cervantes.te> @eric.ernst it is from the repo | 16:47 |
| kata-irc-bot | <eric.ernst> Was there discussion on whether to build from source or use the release before? | 16:49 |
| kata-irc-bot | <eric.ernst> wrt crio/containerd? | 16:49 |
| kata-irc-bot | <eric.ernst> If we just did the release, we'd get the service file, etc, setup 'out of the box' | 16:49 |
| kata-irc-bot | <manohar.r.castelino> folks we are trying to provide and easy to use crio + kata + firecracker + qemu + runc kubernetes setup... and wanted to setup device mapper in a way that worked in any enviornment using a sparse file backed loop device... If anyone has feedback on a better more production worthy setup the achieve the same (with devicemapper) I would appreciate the feedback https://github.com/clearlinux/cloud-native-setup/pull/25 | 17:22 |
| kata-irc-bot | <manohar.r.castelino> It is typically hard to setup devicemapper in a production worthy way without dedicated block devices/partitions | 17:22 |
| kata-irc-bot | <eric.ernst> ah, you're just working around CRIO forcing you to have it setup production-wise | 17:23 |
| kata-irc-bot | <manohar.r.castelino> @eric.ernst yes :( | 17:23 |
| *** LinuxMe has quit IRC | 17:24 | |
| kata-irc-bot | <eric.ernst> containerd doesn't force you | 17:25 |
| kata-irc-bot | <eric.ernst> ..but... it also doesn't have its own snapshotter :slightly_smiling_face: | 17:25 |
| kata-irc-bot | <gabriela.cervantes.te> @eric.ernst I did not remember if there was a discussion about building from source or use the release before | 17:29 |
| *** LinuxMe has joined #kata-dev | 17:34 | |
| *** jodh has quit IRC | 18:00 | |
| *** LinuxMe has quit IRC | 18:22 | |
| *** LinuxMe has joined #kata-dev | 18:34 | |
| *** LinuxMe has quit IRC | 18:49 | |
| kata-irc-bot | <raravena80> is devicemapper on CRIO even prod ready? I'm reading here that it's not yet. https://docs.openshift.com/container-platform/3.11/crio/crio_runtime.html. I suppose at one point it will be. | 19:31 |
| kata-irc-bot | <raravena80> | 19:31 |
| kata-irc-bot | <eric.ernst> I'll ask mrunal about it | 19:32 |
| kata-irc-bot | <eric.ernst> anyone want to spend their afternoon on a fun bug? https://docs.openshift.com/container-platform/3.11/crio/crio_runtime.html#configuring-cri-o-storage | 19:33 |
| *** LinuxMe has joined #kata-dev | 19:39 | |
| *** eernst has joined #kata-dev | 19:51 | |
| *** eernst has quit IRC | 19:53 | |
| *** eernst has joined #kata-dev | 20:06 | |
| *** davidgiluk has quit IRC | 20:28 | |
| *** LinuxMe has quit IRC | 21:19 | |
| *** eernst has quit IRC | 21:25 | |
| *** openstack has joined #kata-dev | 23:43 | |
| *** ChanServ sets mode: +o openstack | 23:43 | |
Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!