| 3 Jun 2021 |
 ajs124 | maybe they're just a big fan of statsd | 12:57:28 |
 das_j | Yeah, investing in a perl codebase is unarguably an idea that results in mixed feelings | 13:12:31 |
 @grahamc:nixos.org | for sure | 13:26:23 |
| @grahamc:nixos.org | it certainly isn't my idea of career development lol | 13:26:43 |
 Sandro | do you want to maintain it or use something already there? | 14:12:35 |
 cransom | imo, i try not to stack bandaids. if you have to go through code change anyway and the end result is prometheus, going through an intermediate statsd step is cumbersome. regardless of if it's perl or not. | 15:50:37 |
| @grahamc:nixos.org | I agree cransom | 17:29:48 |
| @grahamc:nixos.org | I don't suppose you would be interested in making a Prometheus version of that Plack middleware? | 17:30:12 |
| @grahamc:nixos.org | it isn't actually so complicated to write, but I'd want to write tests for it too, and I think it'd take more time than I can really commit :( | 17:30:31 |
| 4 Jun 2021 |
 andi- | Do the hydra tests have any concept of test fixtures? Just looking at writing a test around a jobset and so far the tests look like they depend on each other (being executed in order). Usually I'd just setup a jobset and discard of it after my test again. | 14:28:49 |
| andi- | cole-h: ^ you wrote most of the code there :D | 14:29:02 |
| @grahamc:nixos.org | every .t file runs its own postgres database | 14:31:39 |
| @grahamc:nixos.org | * every .t file runs its own postgres server | 14:31:47 |
| @grahamc:nixos.org | and hydra-init runs then you have to create everything from scratch from there | 14:32:00 |
| andi- | Ah nice, I'll just add another file then. | 14:34:08 |
| 6 Jun 2021 |
|  Valodim joined the room. | 11:01:00 |
| Valodim | https://hydra.nixos.org/search?query=aaaa it seems that just about any search term that isn't very short will run into query timeouts. also, "aaaa" finds results with four as anywhere in the Job name. is the hydra search doing an unoptimized fuzzy search? | 11:13:30 |
| das_j | In reply to @Valodim:stratum0.org
https://hydra.nixos.org/search?query=aaaa it seems that just about any search term that isn't very short will run into query timeouts. also, "aaaa" finds results with four as anywhere in the Job name. is the hydra search doing an unoptimized fuzzy search? https://github.com/NixOS/hydra/blob/master/src/lib/Hydra/Controller/Root.pm#L431 | 11:19:11 |
| Valodim | ah, there's actually "aaaa" in those results elsewhere. so it's really just an ILIKE "%query%" that consistently times out | 11:24:38 |
| @grahamc:nixos.org | in this PR I included some potential improvements to the search behavior: https://github.com/NixOS/hydra/pull/853 | 14:45:47 |
| Valodim | queries that aren't left-anchored don't profit from btree indexes. are you currently using a trigram gin, or any other sort of specialized index for this operation? | 17:35:49 |
| Valodim | with a bit of luck, your woes might be gone with just CREATE INDEX expIndexBuildOutputsPathGin ON BuildOutputs USING GIN (path gin_trgm_ops); | 17:38:44 |
| Valodim | alternatively, maybe just don't allow fully fuzzy LIKEs but use a more specialized matching? I imagine searches by path will all look more or less alike, i.e. for a path /nix/store/hash-name-version the user will likely search either by that full string, or hash-name-version, or hash, or name-version | 17:50:55 |
| Valodim | I have next to no perl experience so probably won't have the time to encode that idea into a PR :\ | 17:51:42 |
| Valodim | Has anyone looked more closely into optimizing those queries yet? If not, I could probably do some experiments and see what works. | 17:56:05 |
| 7 Jun 2021 |
|  David Arnold joined the room. | 03:57:47 |
|  manveru joined the room. | 08:09:39 |
| @grahamc:nixos.org | https://github.com/SysPete/Catalyst-Plugin-PrometheusTiny/pull/2 | 16:37:13 |
 Mic92 | grahamc (he/him): what is your personal roadmap for hydra? | 16:58:17 |
| Mic92 | We don't have enough roadmaps in the nix community. | 16:58:35 |
