| 16 Oct 2025 |
 raitobezarius | i was going to call this out | 11:49:31 |
| raitobezarius | i'm confused why #narinfo >>> #nars | 11:49:40 |
| raitobezarius | but perhaps, nar uploads are averted | 11:49:49 |
 BMG | Looks like about 500k or so more narinfos a month than nars | 11:49:52 |
| raitobezarius | and narinfos are rewritten no matter what | 11:49:53 |
| BMG | Now I know the upload process is "messy" | 11:50:01 |
| raitobezarius | :"D | 11:50:06 |
| BMG | Less messy, more very simple. Not atomic. You can have a nar without a narinfo or vice versa because they are uploaded independently | 11:50:48 |
| BMG | These numbers would seem to indicate something is not quite right with the upload process. | 11:51:09 |
| BMG | I'm spending my afternoons on this stuff at the min. I'm gonna grab some lunch and dig into this properly. Need to brush up on my clickhouse skills a bit. | 11:51:58 |
| BMG | Might be faster to just finish the postgres import process and poke around it there, cos I'll have parsed the narinfos and have the references etc and be able to relate the nar entries to the nar infos | 11:53:20 |
| BMG | 🤔 | 11:53:30 |
 Vladimír Čunát | narinfo is per derivation. nar is per derivation output. Am I wrong? | 12:01:57 |
| BMG | 🤔 | 12:02:09 |
| Vladimír Čunát | We do use multi-output derivations pretty commonly, at least in the more-core packages. | 12:02:26 |
| BMG | That would make sense | 12:02:44 |
| BMG | But woudl that mean more nars than narinfos? | 12:03:17 |
| BMG | * But wouldn't that mean more nars than narinfos? | 12:03:25 |
 hexa | yes | 12:03:31 |
| BMG | We're seeing the opposite | 12:03:45 |
| BMG | If my clickhouse skills are to be believed | 12:03:51 |
| raitobezarius | <confusion intensifies> | 12:04:02 |
| Vladimír Čunát | Ah, I misread that. OK. | 12:04:20 |
| Vladimír Čunát | It does happen commonly on Hydra that we build the same-hashed derivation twice in parallel. | 12:06:02 |
| Vladimír Čunát | * It does happen commonly on Hydra that we build the same-hashed derivation twice in parallel (and both succeed). | 12:06:09 |
| Vladimír Čunát | I'm not sure what exactly happens in that case. But even so. | 12:06:26 |
| BMG | which would account for same nar but multiple narinfos? | 12:06:29 |
| BMG | assuming the build output was bit-wise reproducible | 12:06:48 |
| Vladimír Čunát | They often are not, but I'm not sure that really affects the count of re-uploads. | 12:07:23 |
| Vladimír Čunát | It doesn't affect the (S3/URL) paths. | 12:07:39 |
