| 16 Apr 2023 |
 @elvishjerricco:matrix.org | No? | 16:42:15 |
| @elvishjerricco:matrix.org | For stage 1, most people probably just want DHCP on one or all of the physical interfaces, as configured typically with networking.useDHCP or networking.interfaces.... Then stage 2 can take over network configuration and add any fancier things on top of that, but presumably most people will still want those same interfaces configured with DHCP | 16:44:28 |
| @elvishjerricco:matrix.org | * For stage 1, most people probably just want DHCP on one or all of the physical interfaces, as configured typically with networking.useDHCP or networking.interfaces... (often generated by nixos-generate-config). Then stage 2 can take over network configuration and add any fancier things on top of that, but presumably most people will still want those same interfaces configured with DHCP | 16:44:48 |
 @hexa:lossy.network | so your intent is to save one DHCP roundtrip in that scenario | 16:45:50 |
| @elvishjerricco:matrix.org | My intent was mainly to follow the suggestion in the systemd-networkd man page, and it made enough sense to me to make it default but still implement the flush behavior for those who want it | 16:46:41 |
| @elvishjerricco:matrix.org | but also | 16:46:42 |
| @elvishjerricco:matrix.org | there's things like root on NFS that would rely on having flushing disabled | 16:46:52 |
| @hexa:lossy.network | personally I configure a dedicated IPv6 address in stage1, and something else in stage2, so that I don't get conflicting hostkeys on SSH 🙂 | 16:47:03 |
| @hexa:lossy.network | no strong opinion, as you said I can enable flushing | 16:47:44 |
| @hexa:lossy.network | * no strong opinion, as you said I can enable flushing and I will | 16:47:50 |
| @hexa:lossy.network | since it is configurable I don't see it as a blocker fwiw | 16:48:34 |
| @hexa:lossy.network | the default also doesn't bother me too much | 16:48:44 |
| @elvishjerricco:matrix.org | Yea I think we'll leave flushing disabled by default when systemd initrd is enabled (and enabled by default with regular initrd, since that's the existing behavior). flokli does that sound ok to you? | 16:49:52 |
 flokli | Sorry, i was busy with family stuff all day and couldn't get to reply on messages | 16:50:45 |
| @elvishjerricco:matrix.org | no worries | 16:50:54 |
| flokli | I'll reply in a bit | 16:51:27 |
 Arian | Slight preference on opt in flushing from my side | 16:51:43 |
| Arian | But fine either way | 16:51:47 |
| flokli | I guess flushing is now opt-in. | 20:26:45 |
| flokli | ElvishJerricco: hexa I commented in https://github.com/NixOS/nixpkgs/pull/169116#discussion_r1168015358 | 20:26:58 |
| @hexa:lossy.network | If system-wide link configs are used in stage1 that would make my use case for flushing non-existant | 20:28:22 |
| @hexa:lossy.network | * If system-wide link configs are applied in stage1 that would make my use case for flushing non-existant | 20:28:29 |
| flokli | As far as other netdev settings (like controlling link speeds explicitly, or *{CheckSum,Segmentation}Offload, systemd also pretty much added all these as options | 20:32:53 |
| flokli | * As far as other netdev settings (like controlling link speeds explicitly, or *{CheckSum,Segmentation}Offload, systemd also pretty much added all these as options. I'm not sure how much of that is exposed to be accessible from the module system | 20:33:06 |
| flokli | and when it comes to "having a different IP address in initrd to avoid host key stuff", I'd probably go with a different port, and keep the IP. But in any case, that might be a usecase that could be mentioned in the description | 20:35:18 |
| 17 Apr 2023 |
| flokli | ElvishJerricco: sorry for being so pedantic on the networking bits in the PR, but i feel like we need to get the semantics sorted out before too many people end up switching to it, and possibly get angry if we switch semantics and break possibly remote machines. | 21:08:14 |
| @elvishjerricco:matrix.org | Yep, I totally understand | 21:08:29 |
| @elvishjerricco:matrix.org | I do really want to get it done before 23.05, but that shouldn't be too hard even with max pedantry :) | 21:08:50 |
| flokli | I'll try to check my notifications more frequently in the coming days :-) | 21:09:30 |
| @elvishjerricco:matrix.org | Rather than continuing to discuss this tangent on the PR,
As for the link units, the PR that added this comment also linked to an issue, but that issue was closed by this way before that PR was even made. I'm wondering why that systemd change isn't enough to resolve the problem. Surely you're not expected to do your interface renaming in stage 1. In fact, the comments in the issue suggest that the udev rules to rename the interface in stage 2 should happen before networkd is even notified the interface is ready.
| 21:18:09 |
