15 Jun 2021 |
colemickens | * I think I saw that preexecstart when skimming... but that's odd, again I'm surprised I hadn't noticed that when I was testing? I definitely had the guest/agent booting. | 07:29:09 |
colemickens | anyway, more good tips. | 07:29:17 |
Mic92 (Old) | The error message comes from qemu itself. I use virtio-fs | 07:29:56 |
Mic92 (Old) | I have not debugged this to the hypervisor itself. I only had a weekend to get it working. | 07:30:19 |
colemickens | ok | 07:30:59 |
colemickens | thanks for the input; one last one and then I have to run - I saw you forked vhive, any quick thoughts to share on it coming from the kata perspective? or just something you're keeping an eye on | 07:31:35 |
colemickens | * thanks for the input; one last one and then I have to run - I saw you forked vhive, any quick thoughts to share on it coming from the kata perspective? or just something you're keeping an eye on? | 07:31:36 |
Mic92 (Old) | It is based on a research project. It certainly shows what needs to be optimized to make lambda functions more scalable. But there is no big coorp like Intel working behind it. Just 4 PhD students working on their next publications. We use it in our research project to make a usecase for vmsh. | 07:33:45 |
Mic92 (Old) | Also quality and documentation is certainly above what you would see from a average research project. | 07:34:23 |
Mic92 (Old) | They got CI and end-to-end documentation. | 07:34:35 |
Mic92 (Old) | I think the vhive people also hangout in some firecracker slack or so. | 07:35:13 |
16 Jun 2021 |
| CRTified (old handle) joined the room. | 08:35:18 |
| Rosuavio joined the room. | 18:52:48 |
18 Jun 2021 |
| rager joined the room. | 17:02:26 |
rager | oh shoot... so it's a thing - cgroups v2 unified format is mucking up all us container users? | 17:05:49 |
rager | I'm running into lots of "failed to run Kubelet: mountpoint for cpu not found" when I run the nixos kubernetes packages in a nixos container | 17:07:40 |
colemickens | well that probably has to do with trying to run a container manager inside a container | 17:43:29 |
colemickens | somehow I didn't realize a patch was needed for this: https://www.phoronix.com/scan.php?page=news_item&px=Linux-Boot-Non-BD-File-Systems&utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+Phoronix+%28Phoronix%29 | 17:43:50 |
colemickens | which would be useful in my virtiofs + kata for Nix scenario | 17:44:03 |
andi- | Does virtiofsd still require a daemon running as root? | 17:53:31 |
colemickens | yes. " The program drops privileges where possible during startup although it must be able to create and access files with any uid/gid:" | 18:06:47 |
Mic92 (Old) | In reply to @andi:kack.it Does virtiofsd still require a daemon running as root? If you want to support multiple uids this would be needed. But in theory usernamespaces could help with that. | 18:31:50 |
Mic92 (Old) | I can check with my Kata containers | 18:32:18 |
andi- | I wanted to use it for the NixOS VM test a year ago when it was merged into the kernel but that daemon requirement was a show stopper :/ | 18:32:50 |
Mic92 (Old) | In reply to @rager:synapse.lickmy.app oh shoot... so it's a thing - cgroups v2 unified format is mucking up all us container users? K8s has an option to fix that | 18:33:15 |
Mic92 (Old) | In reply to @andi:kack.it I wanted to use it for the NixOS VM test a year ago when it was merged into the kernel but that daemon requirement was a show stopper :/ It should not be impossible support. Upstream probably would accept PRs for that. I put this on my to-do list. I know fuse quite a bit | 18:34:34 |
Arian | I have a Kubernetes module in my personal repo that uses cri-o and puts Kubelet in systemd Cgroup mode. It works fine with cgroups v2 | 18:35:43 |
Arian | But yeh by default Kubernetes still defaults to cgroupfs and docker and that will definitely not work | 18:35:59 |
Arian | The docker driver is practically deprecated so idk if it will ever support it. We should perhaps move the default CRI driver in nixos to cri-o | 18:36:39 |
Arian | I don't have enough time to turn this into a proper PR atm | 18:37:14 |