| 13 Aug 2022 |
 colemickens | * it's just initrd -> and then... it does mount a rootfs (over nfs), but it could trivially be a tmpfs | 22:48:57 |
| colemickens | * I serve the nix-store for the entire netbootserver up, but I also serve a specialized nix-db export of the netboot-client's store export to the client | 22:49:09 |
| colemickens | now that my rpipkg branch is rebased, I can say that visionfive, rpi02w, rpi3b+, rpi4b all netboot with systemd-init with my sbin patch (that I think I might even be able to eliminate) | 23:31:51 |
 @elvishjerricco:matrix.org | wow that's excellent | 23:32:14 |
| 14 Aug 2022 |
| colemickens | I would like to maybe try using this on one of my rpi02w systems: https://github.com/mhx/dwarfs | 00:23:26 |
| colemickens | it seems rather slow with this nfs setup compared to one with an sd card in it | 00:24:04 |
| colemickens | I suspect even just tmpfs root would really speed it up though | 00:26:36 |
| @elvishjerricco:matrix.org | colemickens: NFS just kinda sucks in general in my limited experience | 00:28:35 |
| @elvishjerricco:matrix.org | for one thing all writes are sync writes on the server for some reason | 00:29:18 |
| colemickens | that's why I'm hoping it will be enough to kill the rootfs if the rest is just the RO nix store | 00:29:46 |
| @elvishjerricco:matrix.org | there's probably better network file systems, though I guess it's rare for them to be suitable for a rootfs | 00:30:16 |
| colemickens | but also I don't really need nfs at all, so... now I'm dreaming up the recursive-nix initrd builder combiend with this dwarFS business.. | 00:30:18 |
| @elvishjerricco:matrix.org | why would that need recursive nix? | 00:30:37 |
| colemickens | it doesn't, I'm conflating things | 00:31:00 |
| colemickens | part of nfs = quick turnaround times since there's no images to build/push around | 00:31:09 |
| colemickens | the recursive nix initrd builder would presumably help with that a bit ( I think) | 00:31:36 |
| @elvishjerricco:matrix.org | "the recursive nix initrd builder" I do not know what this is | 00:31:53 |
| colemickens | https://github.com/grahamc/netboot.nix | 00:32:03 |
| colemickens | (sorry) | 00:32:06 |
 @grahamc:nixos.org | I don’t use that anymore preferring a faster option | 00:32:37 |
| colemickens | oh! and... oh? | 00:32:54 |
| @grahamc:nixos.org | https://github.com/DeterminateSystems/nix-netboot-serve | 00:33:04 |
| @grahamc:nixos.org | In the next few weeks I’ll be pushing up a disk image builder which takes 2-10s too. | 00:34:10 |
| colemickens | very excellent, thanks for the heads up on both | 00:34:58 |
| @grahamc:nixos.org | Yea | 00:35:03 |
| @grahamc:nixos.org | At this point I can make, upload, and boot an AMI in ~30s, making the most drastically slow step terminating the instance. | 00:35:38 |
| @elvishjerricco:matrix.org | I would love to know more about that, though tonight is a night for things-that-aren't-that :P | 00:36:32 |
| @grahamc:nixos.org | I’m going to bed in a few minutes, and it skips stage-1 anyway … :) | 00:37:00 |
| @elvishjerricco:matrix.org | Y'know, over time, I've stumbled over various little tidbits that make me think we're handling file systems in initrd slightly wrong. And in particular, I've been trying to figure out why I don't have After=local-fs-pre.target on initrd file systems since not having that can fuck with hibernation resumption, and I found this: https://github.com/systemd/systemd/blob/3f504b892b92f54087feeb3fb35e3938567d7fa0/src/core/mount.c#L482-L484 | 03:25:05 |
| @elvishjerricco:matrix.org | I think we're really meant to be using initrd-fs.target the way we've been using local-fs.target | 03:25:59 |
