| 4 Dec 2024 |
 Rick (Mindavi) | Likely you need to patch that, but maybe read the hydra tests to see how they handle this? | 16:20:01 |
 ma27 | fwiw I wrote a patch for this a while ago: https://github.com/NixOS/hydra/pull/1360 | 16:24:15 |
|  @belak:matrix.org left the room. | 17:39:24 |
 rhelmot | that looks like it's just replacing the local store with the remote store, which is not an option that I'm using right now - should I be? | 20:03:42 |
| rhelmot | I've just been setting store = daemon?real=/nix/hydra-store | 20:04:05 |
| rhelmot | or local?real=/nix/hydra-store in the daemon process | 20:04:20 |
| rhelmot | unless store and store_uri are synonyms...? | 20:04:41 |
| 5 Dec 2024 |
 hexa | does hydra gc build logs ever? 😄 | 11:11:20 |
| hexa | because h.n.o has like 420 GB worth of build logs and that feels a wee bit heavy | 11:11:58 |
| Rick (Mindavi) | I don't think it does. | 13:11:33 |
 das_j | In reply to @hexa:lossy.network
does hydra gc build logs ever? 😄 no | 13:31:34 |
| das_j | but you should just be able to compress them | 13:31:42 |
| das_j | depending on your version even with zstd | 13:31:57 |
| hexa | they are bz2 compressed | 13:32:55 |
| hexa | * they are bz2 compressed | 13:33:01 |
| hexa | and then zfs comes and compresses that again to 1.08x | 13:33:18 |
| hexa | so I think the best outcome would be to prune those older than say 3 years? | 13:33:42 |
 toonn | With fs compression enabled isn't the file-level compression counterproductive? | 14:22:47 |
| hexa | The zfs compression is "on" | 14:44:28 |
| hexa | That should be dirt cheap | 14:44:36 |
| hexa | But yeah, ideally don't compress and compress the fs with zstd-19 | 14:45:36 |
| Rick (Mindavi) | In reply to @hexa:lossy.network
so I think the best outcome would be to prune those older than say 3 years? I wouldn't mind that, only recent logs are typically accessed | 15:03:34 |
| Rick (Mindavi) | And people may always rebuild themselves to generate the logs | 15:03:49 |
 vcunat | Aren't they in S3 as well? | 16:36:30 |
| 6 Dec 2024 |
| hexa | systemd.services.hydra-prune-build-logs = {
description = "Clean up old build logs";
startAt = "weekly";
serviceConfig = {
User = "hydra-queue-runner";
Group = "hydra";
ExecStart = lib.concatStringsSep " " [
(lib.getExe pkgs.findutils)
"/var/lib/hydra/build-logs/"
"-type" "f"
"-mtime" "+${toString (3 * 365)}"
"-delete"
];
};
};
| 03:14:04 |
| hexa | In reply to @vcunat:matrix.org
Aren't they in S3 as well?
upload_logs_to_binary_cache = true
| 03:14:31 |
| hexa | only slightly wasteful | 03:14:43 |
| hexa | * systemd.services.hydra-prune-build-logs = {
description = "Clean up old build logs";
startAt = "weekly";
serviceConfig = {
User = "hydra-queue-runner";
Group = "hydra";
ExecStart = lib.concatStringsSep " " [
(lib.getExe pkgs.findutils)
"/var/lib/hydra/build-logs/"
"-type" "f"
"-mtime" "+${toString (3 * 365)}"
"-delete"
];
};
};
deployed to hydra.nixos.org, will run the first time on monday
| 03:15:28 |
| vcunat | It's pretty small compared to the (compressed) NARs, at a glance. But we could have some GC policy there as well. | 06:25:12 |
| das_j | That works but Hydra doesn't fetch them from there. nix logs should do through | 08:47:07 |
