| 22 May 2024 |
 cole-h | In reply to @qyliss:fairydust.space
A while a go there was a discussion in this room about raising the timeout on Darwin. On staging PRs it never gets past LLVM, which is part of stdenv, so it's essentially useless — but also, that means that every staging PR is going to take at least an hour of OfBorg time, so I and others suspect that raising the timeout to allow it to finish might actually free up capacity, because it will be able to share LLVM builds. Right, I remember now... the only problem I have with that is the Darwin queue frequently grows to the hundreds, and I'm afraid that if given e.g. 2 hours (or more) that means the build queue will become even more unmanageable... I dunno, I suppose I'm willing to give it a shot, just let me what you think we should start with (2? 3? 6 hours before timeout?) | 22:45:13 |
| 23 May 2024 |
 Alyssa Ross | shouldn't need to be more than 3 | 06:42:05 |
| Alyssa Ross | not sure if 2 would be enough | 06:42:08 |
 hexa | are the builders sharing a cache of any kind? | 13:01:10 |
| hexa | or would every individual builder have to build their own llvm | 13:01:23 |
 K900 | Even if they share a cache they don't share locks | 13:01:43 |
| K900 | So they'll attempt to race | 13:01:47 |
| hexa | yes, but you can't have it all 🙂 | 13:04:12 |
| cole-h | ofborg has a queue, so if one builder picks up the only LLVM build, nobody else can pick it up | 14:38:40 |
| cole-h | In reply to @qyliss:fairydust.space
shouldn't need to be more than 3 I'll give it a shot later today and we'll see how it goes. | 14:39:29 |
 7c6f434c | It's not an LLVM build, it's an LLVM-dependent build though | 14:40:49 |
| cole-h | Gotcha, I misunderstood the concern | 14:41:15 |
| hexa | and caching? | 18:10:57 |
| hexa | if one builder has the result, can others reuse it? 😛 | 18:11:03 |
| cole-h | No, no caching between machines at this moment | 18:11:34 |
| Alyssa Ross | ah, I assumed there would be | 18:12:48 |
| Alyssa Ross | that makes me less confident it would save time | 18:12:57 |
| hexa | wondering, does haumea run on darwin? | 18:13:49 |
| hexa | that would allow sharing directly from the /nix/store | 18:14:05 |
| cole-h | No clue what haumea is, so can't comment | 18:14:38 |
| hexa | harmonia | 18:14:48 |
| hexa | sorry 😄 | 18:14:48 |
| cole-h | (Other than it was at one point? the name of the hydra db I think) | 18:14:56 |
| hexa | https://github.com/nix-community/harmonia | 18:14:59 |
| hexa | yeah, my head wandered a bit | 18:15:10 |
| cole-h | Does harmonia support GCing? The core machine doesn't have much free disk space to store artifacts, and I imagine running on every PR will fill up what little it has very quick... lol | 18:18:58 |
| cole-h | (Not much is ~500GiB, fwiw, so not "too little" either) | 18:19:26 |
| hexa | I thin not, in that I had to fiddle with the ttl in nix | 18:25:25 |
| hexa | because nix is not kind about failed substitution | 18:25:34 |
| hexa | narinfo-cache-positive-ttl = 3600;
| 18:26:09 |
