| 14 Feb 2022 |
 das_j | * Also if the code looks like I really wanted to learn some go reflection - that's what the project was initially for | 12:55:09 |
 hexa | this looks much more complete from what I remember when I touched the upstream prometheus instance a few days ago 🙂 | 12:59:32 |
| das_j | Not by a lot actually. There's some stuff like nrUnsupportedSomething but the upstream exporter is mostly complete | 13:00:27 |
| das_j | Maybe you want --web.disable-exporter-metrics? :D | 13:00:49 |
|  Tseb joined the room. | 13:34:33 |
 @grahamc:nixos.org | nice | 14:15:18 |
| @grahamc:nixos.org | I'm hoping we could add a proper exporter directly to the evaluator and queue runner some day, but this looks great | 14:15:37 |
| das_j | yeah I was hoping the evaluator somehow dumps its state into the db as well but I couldn't find anything | 14:16:19 |
| @grahamc:nixos.org | one tricky bit of that is of course the process breakdown of hydra-evaulator calling hydra-evaluate-jobset calling hydra-eval-jobs, and potentially each of these could feasibly contain their own exporter, which would also be weird | 14:27:52 |
| das_j | What about this: The main daemon opens a fd for its subprocesses (FD 5 or something), and the subprocesses write their metrics there, maybe as kv pairs? Would be shitty but better than nothing | 14:29:04 |
