| 30 Jun 2025 |
 emily | like NixOS does not have a second source policy :) | 15:26:44 |
 hexa (clat on linux when) | it isn't, but it makes it easier | 15:26:45 |
 Sandro 🐧 | but Marcel is in the chat here | 15:26:49 |
| hexa (clat on linux when) | to point to alternatives | 15:26:53 |
| emily | we have tons of things we have only one in-tree interface for | 15:26:52 |
| emily | and if we deprecate networking.* then there's even less reason | 15:27:02 |
| emily | since we are not blessing anything at that point | 15:27:06 |
| Sandro 🐧 | for the services, not core routing AFAIK | 15:27:45 |
 Marcel | In reply to @hexa:lossy.network
would be great to offer people an alternative to networkd so we can drop scripted eventually I plan to upstream it eventually but I want to wait till after V2 which will introduce braking change in the configuration file. If you don't care about these I can also upstream the v1 | 15:48:03 |
| Marcel | In reply to @sandro:supersandro.de
for the services, not core routing AFAIK Also IBH (local isp) uses it. There also for the routing infrastructure. | 15:49:00 |
|  dan_nrw changed their profile picture. | 17:18:13 |
 Zhaofeng Li | For those with some kind of overlay mesh network, what kind of optimization do you do? Like tuning OSPF metrics based on some periodic measurement, etc | 22:06:08 |
|  Etherbloom [they/sie] {you&} 🐲🦊👿 joined the room. | 22:35:42 |
| 1 Jul 2025 |
|  Pratham Patel changed their display name from Pratham Patel (you can mention me) to Pratham Patel. | 05:10:13 |
|  xanderio joined the room. | 12:12:49 |
| xanderio | Hey, i'm currently hitting this issue (https://github.com/NixOS/nixpkgs/issues/380987) where systems using systemd-resolved resolve there FQDN to 127.0.0.2 which break services on that system trying to contact other services on the same host. This bug is included in 25.05.
I'm currently in the process on rebasing and trying to fix the outstanding issues with https://github.com/NixOS/nixpkgs/issues/132646. | 12:15:59 |
| hexa (clat on linux when) | I run babeld with tunnel type, which uses an RTT based metric https://bird.network.cz/?get_doc&v=20&f=bird.html#toc6.2 | 12:19:04 |
| hexa (clat on linux when) | as defined in https://datatracker.ietf.org/doc/rfc9616/ | 12:19:24 |
| hexa (clat on linux when) | congestion causes delay, which will deprioritize a path | 12:19:38 |
| hexa (clat on linux when) | i would suggest staying away from something as static as OSPF for private use | 12:19:50 |
 magic_rb | Related note, say i ran bird on top of wireguard. Bird adjusts routing tables so it will still end up using the inkernel wireguard implementation right | 12:23:57 |
| magic_rb | Yeah both babeld and bird just create a mesh on top of the existing wireguard network by adjusting routes | 12:26:25 |
| magic_rb | Interesting | 12:26:26 |
| hexa (clat on linux when) | wireguard just provides a transfer network, just make sure AllowedIPs= does not block the traffic you intend to send over it | 12:27:50 |
| magic_rb | Mhm yeah thats how i understood things too, cool, maybe one day | 12:28:58 |
| Zhaofeng Li | Yeah, AllowedIPs is a pain for dynamic routing... each edge in my mesh is on a separate interface with AllowedIPs=::/0 | 12:40:54 |
| hexa (clat on linux when) | yeah, same here | 12:41:08 |
| xanderio | Ok after looking deeper into this issue, I'm not convinced that this is the initial networking issue i've assumed. The more I look into this the more confused I get. | 12:58:52 |
| Zhaofeng Li | actually I was thinking of modifying the wireguard client to have a fake ethernet header that's solely used to select the peer 🙃 | 13:01:16 |
| Zhaofeng Li | but I never got to do it, and being able to see the peer name in bird (since it's the interface name) is a nice side-effect of having separate interfaces | 13:02:58 |
