=== Kuraokami is now known as IchabodCrane | ||
ajd | has anyone here ever tried getting cloud-init working with buildroot? (i realise that buildroot is normally something one uses for the exact opposite of the cloud, but i've got a very niche use case where i want to use buildroot to generate images for VMs i'm running in openstack, so i might try wiring up cloud-init...) | 02:28 |
---|---|---|
Akila | hi, autoinstall + cloud-init handles OS installation and OS parameter customization in Ubuntu..Do we have such autoinstall kind of option with RHEL & SLES that works with cloud-init | 05:58 |
caribou | Hello, long time no see :) | 14:49 |
caribou | We may have identified an issue with the Network-Manager renderer (tested on Jammy's 23.3.1-0ubuntu1~22.04.1). | 14:49 |
caribou | When rendering a v2 yaml config, with mixed IPv6 and IPv6, the IPv6 route ends up in the IPv6 section of the *.nmconnection file. | 14:49 |
caribou | When using other renderers the routes are in their correct section. | 14:49 |
caribou | You can find an example here : https://pastebin.mozilla.org/s20gvnuE | 14:49 |
caribou | TL;DR : The cloud-init-eth0.nmconnection has : | 14:49 |
caribou | [ipv4] | 14:49 |
caribou | ... | 14:49 |
caribou | route1=169.254.42.42/32,62.210.0.1 | 14:49 |
caribou | route2=::/0,fe80::dc00:ff:fe20:186 | 14:49 |
minimal | caribou: I've seen the same behaviour in the past with the ENI renderer | 15:10 |
minimal | I started to investigate it but other things got in the way and so I never tracked down the cause | 15:10 |
caribou | well, just tested it and it's working fine here | 15:11 |
caribou | (i.e. ENI) | 15:11 |
caribou | I'll try to take some time to look at it | 15:12 |
minimal | hmm, let me retest. It has been a few months since I last looked at this | 15:12 |
caribou | btw I'll be more than happy to file a bug on this one if needed | 15:16 |
minimal | caribou: yupe, still behaves that way with eni | 15:23 |
caribou | I don't see it with the config.yaml that I posted in the pastebin | 15:23 |
caribou | but doesn't mean it's not failing | 15:23 |
minimal | have "pre-up route add" and "pre-down route del" entries in the "iface eth0 inet static" section for IPv6 routes rather than in the "iface eth0 inet6 static" section | 15:24 |
minimal | this is with my own pre-existing v2 network config | 15:24 |
minimal | caribou: testing with your YAML also behaves the same with ENI | 15:32 |
minimal | caribou: https://tpaste.us/D7PO | 15:36 |
caribou | Doing a bit of online debugging we found that here : https://github.com/canonical/cloud-init/blob/c84369ac4fb91c9a32af28d47771d07631e2f3bc/cloudinit/net/network_manager.py#L257 | 15:38 |
caribou | Looks like the renderer is expecting a v1 structure and it gets a v2, so the test never returns an 'ipv6' family | 15:38 |
minimal | I should have raised a bug for this when I first noticed it but I thought I'd be able to track down the problem and raise a PR with the fix.....but it took some time to debug and (from memory) another unrelated network rendering bug appeared around the same time that I got distracted by | 15:39 |
minimal | caribou: it's been a while since I looked at this code | 15:42 |
caribou | yeah, I suppose | 15:42 |
caribou | It's getting near EOD here, I'll give it a closer look tomorrow morning | 15:43 |
caribou | & create a bug for it | 15:43 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!