| 27 Jun 2026 |
 emily | guess that's not really a problem if big-parallel is sufficiently rare | 14:03:47 |
 hexa (signing key rotation when) | and if they are small they'll pass through much faster | 14:03:50 |
| hexa (signing key rotation when) | so eventually they get clogged with big-parallel builds | 14:04:08 |
| emily | more thinking about ones that will take roughly the same time but hog the slots | 14:04:13 |
| hexa (signing key rotation when) | ❯ rg big-parallel -l | wc -l
60
| 14:04:23 |
| emily | right | 14:04:30 |
| emily | ideally you wouldn't need to statically assign this stuff at all but … well, preaching to the choir | 14:04:47 |
| hexa (signing key rotation when) | https://prometheus.nixos.org/query?g0.expr=hydraqueuerunner_machine_type_runnable&g0.show_tree=0&g0.tab=table&g0.range_input=1h&g0.res_type=auto&g0.res_density=medium&g0.display_mode=lines&g0.show_exemplars=0 | 14:06:28 |
| hexa (signing key rotation when) | so I think we just don't have the required-system-features as a dimension any more | 14:06:48 |
| emily | the failure mode I can see is if e.g. we need LLVMs and stuff for staging/-next but big-parallel builders get busy building a billion backlog things for unstable/releases that don't really benefit from it much instead | 14:07:13 |
| emily | but maybe that's fine since staging/-next getting delayed by other branches is normal, just thinking it could make the initial bottleneck worse but I guess that's what manual management is for | 14:07:53 |
| hexa (signing key rotation when) | we can always bump to front | 14:10:02 |
| hexa (signing key rotation when) | and I say we do | 14:10:06 |
 vcunat | If the overall speed gets 2x higher or more, slight occasional delays won't be such a big deal 😉 | 14:10:33 |
| emily | have we had big-parallel builder just sitting idle most of the time until now? | 14:11:05 |
| hexa (signing key rotation when) | they remain as an architectural problem | 14:11:05 |
| hexa (signing key rotation when) | no, mixed slots | 14:11:11 |
| hexa (signing key rotation when) | 4 bp, 6 normal | 14:11:14 |
| emily | or has the whole system been clogged up enough that they always have a Firefox or something to do? | 14:11:19 |
| emily | ah, ok | 14:11:25 |
| hexa (signing key rotation when) | for that we'd need to run multiple queue-builders now | 14:11:27 |
| hexa (signing key rotation when) | but the module does not support that yet | 14:11:38 |
| emily | so they'd operate below capacity because big-parallel slots are frequently unused, but everything operated below capacity so it was less visible? | 14:11:46 |
| hexa (signing key rotation when) | or the queue-builder would need support for better slot definitions | 14:11:47 |
| hexa (signing key rotation when) | yea | 14:11:54 |
| hexa (signing key rotation when) | Redacted or Malformed Event | 14:11:55 |
| emily | going to supported but not required makes sense to me then. I wonder if it makes sense to slightly decrease the per-job core count at the same time in exchange for the extra jobs you get. (sorry, this is just curiosity on my part, not trying to meddle with the decisions :) ) | 14:14:35 |
 Myria | Uhhh, it seems the problems I had with stdenv on the gcc-15.3.0 changes is also affecting staging-next. gfortran is failing build on aarch64-darwin (https://hydra.nixos.org/build/333247000) but locally it builds for me whe I ignore the cache | 14:16:46 |
| Myria | Actually, may just be a gc error | 14:17:43 |
| Myria | But this likely causes a large amount of failures | 14:18:00 |
