| 30 Jun 2026 |
 Mic92 | you would drop .zstd anyway from the url | 08:35:48 |
 flokli | And store more data, sure. | 08:35:55 |
| flokli | I would probably do it for newly-uploaded Nars only. | 08:36:14 |
| Mic92 | Surely | 08:36:34 |
 Arian | With my approach we don't need if-non-match all all though.
Yes we might reupload a NAR but the content-length header now lives in the object instead of the narinfo so doesn't matter | 08:38:56 |
| Mic92 | In that case we might as well drop FileSize and FileHash | 08:40:34 |
| Arian | Yes. We should. That was my whole thesis :D | 08:40:58 |
| Arian | Move them to the object. They dont belong in the narinfo | 08:41:18 |
 Vladimír Čunát | At this point I do think that it is better or maybe even solved. We certainly have built some previously risky stuff like graphical ISOs, and I haven't noticed anything looking very much like a GC issue. | 16:34:53 |
 hexa | nixos:release-26.05:nixos.tests.boot.uefiNetboot.x86_64-linux 333646028 -test-run-boot-netboot-uefi.drv Building 1m 38s
| 21:21:13 |
| hexa | weird .drv name starting with a dash | 21:21:19 |
| hexa | we now accumulate a few builds that are clearly finished but still allocate a build slot | 21:21:47 |
| hexa | most in the 10-11h range with the queue-runner being up for 20h | 21:22:11 |
| hexa | like 10-15 jobs | 21:22:20 |
| hexa | and no matching processes on the build host any more | 21:22:51 |
| hexa | Redacted or Malformed Event | 21:23:18 |
| hexa | restarted the queue-builder on goofy-hopcroft and elated-minsk | 21:23:31 |
| hexa | Redacted or Malformed Event | 21:23:31 |
| hexa | and rebooted a few macs, because that was easier to remember how to kickstart the … what was the unit name again? | 21:23:47 |
|  Epic Eric joined the room. | 23:52:48 |
|  Sapii/Saperson joined the room. | 23:53:16 |
| 1 Jul 2026 |
|  @rooster:darkstar.network left the room. | 01:42:29 |
| Mic92 | hexa (signing key rotation when): Looks like build slots are still handled on the queue runner. So if the response from builder to runner has any errors, it might went stale without any reconcilation. | 07:54:18 |
| Mic92 | hexa (signing key rotation when): remove_job for a fire-and-forget rpc call: https://github.com/NixOS/hydra/commit/457b8f861a4c90068bfb961f6ecbc6bba1669368 now it should be retried on error | 08:22:23 |
| Mic92 | * hexa (signing key rotation when): remove_job was a fire-and-forget rpc call: https://github.com/NixOS/hydra/commit/457b8f861a4c90068bfb961f6ecbc6bba1669368 now it should be retried on error | 08:22:32 |
| Mic92 | Historically that part was using ffi code, which is why there was a spawn task | 08:23:05 |
| Mic92 | build_step_update follows the same pattern but I assume we don't care as much if we loose one step update and gain more performance from not stalling on this | 08:30:04 |
| Mic92 | * build_step_update follows the same pattern but I assume we don't care as much if we loose one step update in the frontend and gain more performance from not stalling on this | 08:30:20 |
|  hsjobeki joined the room. | 11:01:09 |
| hsjobeki | Hey guys 👋 | 11:01:31 |
