| Number8 | Hi everyone, I'm investigating a problem with cloud-init version 22.1. When I use it to configure almalinux 9.1, cloud-init configure the network with the files /etc/sysconfig/network-script/ifcfg-* | 07:59 |
|---|---|---|
| Number8 | I think cloud-init 22.1 is too old but I'm not sure | 08:00 |
| meena | Number8: yeah, Alma is pretty new, maybe upgrade to the latest cloud-init and see how that does for you | 08:11 |
| caribou | Hello, I've been using a custom built cloud-init from source to implement our DataSourceScaleway : cloud-init-23.1+70.g34637a49-1.el9.noarch | 11:46 |
| caribou | which works fine on rockylinux-9. | 11:46 |
| caribou | Since 23.2 came out, I rebuilt a new cloud-init-23.2-1.el9.noarch but now after installation I have no network. | 11:46 |
| caribou | The command `nmcli connection up filename /etc/NetworkManager/system-connections/cloud-init-eth0.nmconnection` fails with | 11:46 |
| caribou | Connection activation failed: No suitable device found for this connection (device lo not available because device is strictly unmanaged). | 11:46 |
| caribou | While I investigate what is added b/w commit 34637a49 and the 23.2 tag I thought I'd ask here | 11:47 |
| caribou | I also noticed the following message at the console during reboot : | 11:56 |
| caribou | modules.py[WARNING]: Could not find module named cc_refresh_rmc_and_interface (searched ['cc_refresh_rmc_and_interface', 'cloudinit.config.cc_refresh_rm | 11:56 |
| caribou | c_and_interface']) | 11:56 |
| caribou | ok, I identified the change : 009dbf85a72a9077b2267d377b2ff46639fb3def - net/sysconfig: enable sysconfig renderer if network manager has ifcfg-rh plugin* | 13:33 |
| caribou | For some reason it triggers the creation of a bad /etc/sysconfig/network-scripts/ifcfg-eth0 on RockyLinux 9 only. For the others the file is not touched by cloud-init | 13:35 |
| minimal | caribou: 23.2 removed the refresh_rmc_and_interface module but you must be using an old /etc/cloud/cloud.cfg (or file in /etc/cloud/cloud.d/*) that still references this module | 13:43 |
| caribou | yeah, I noticed that while investigating. I'll clean up the conf file | 13:44 |
| minimal | caribou: I'm not familiar with networkmanager so can't help there but if you have cloud-init debugging enabled then cloud-init.log should show how it handled network config and rendered it to the appropriate format | 14:05 |
| caribou | thanks for the tip. I'll try to dig in further tomorrow | 14:05 |
| falcojr | caribou: that PR added better sysconfig detection. My guess is that Rocky has support for sysconfig but cloud-init wasn't detecting it and using the NetworkManager renderer instead. Now it is detecting sysconfig instead and doing something weird. If you have pre and post logs I could confirm. | 15:03 |
| falcojr | the renders to use and their order is configurable in cloud.cfg.tmpl. If what I said is true, then we probably want to order NetworkManager detection before sysconfig on Rocky. | 15:05 |
| caribou | falcojr: I'll try to provide the logs sometimes tomorrow | 15:07 |
| caribou | falcojr: here they are : | 15:24 |
| caribou | Good (i.e. before 23.2) :https://pastebin.mozilla.org/Ke8tyOAR | 15:24 |
| caribou | Bad : https://pastebin.mozilla.org/KAOuRoXo | 15:24 |
| falcojr | yeah, so `2023-05-30 11:05:41,953 - __init__.py[DEBUG]: Selected renderer 'network-manager' from priority list: None` vs `2023-05-30 15:17:44,557 - __init__.py[DEBUG]: Selected renderer 'sysconfig' from priority list: None` | 15:25 |
| falcojr | changing the priority on Rocky would be a quick fix, but we should probably figure out why the sysconfig renderer isn't working correctly too | 15:26 |
| falcojr | huh... https://github.com/canonical/cloud-init/blob/main/config/cloud.cfg.tmpl/#L384 | 15:38 |
| falcojr | if this file is used/rendered when building for Rocky, the ordering should automatically be correct. If a different cloud.cfg used or generated in the downstream build process, then it would need to be updated accordingly. | 15:42 |
| caribou | Right now, I'm generating packages directly from the GH source tools (i.e. make srpm) | 15:47 |
| minimal | caribou: perhaps you have a file in place to override the default renderers list (which is what I assume also happened regarding the refresh_rmc_and_interface module) | 15:50 |
| caribou | I'm on my way out but I'll check that tomorrow when I'm back. Thanks for helping out | 15:50 |
| === djhankb4 is now known as djhankb | ||
| === m_ueberall is now known as ueberall | ||
| meena | @fal | 21:52 |
| meena | @falcojr added your patch to my pull request | 21:52 |
| falcojr | Thanks! | 22:28 |
| meena | @blackboxsw: remember how i said I plan to do some Kernel work? well, this thing here: https://github.com/lxc/lxd/pull/11761 triggers some unplanned kernel work. | 23:14 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!