| 23 Jan 2025 |
|  Kamilla 'ova joined the room. | 18:58:46 |
| 31 Jan 2025 |
|  Jonas Chevalier changed their display name from Jonas Chevalier to Jonas Chevalier (FOSDEM). | 19:12:38 |
|  SomeoneSerge (back on matrix) changed their display name from SomeoneSerge to SomeoneSerge (Bruxelles). | 19:34:21 |
| 2 Feb 2025 |
|  pbsds changed their display name from pbsds to pbsds (FOSDEM). | 16:04:30 |
| 3 Feb 2025 |
| Jonas Chevalier changed their display name from Jonas Chevalier (FOSDEM) to Jonas Chevalier. | 08:24:12 |
| SomeoneSerge (back on matrix) changed their display name from SomeoneSerge (Bruxelles) to SomeoneSerge (Gand St. Pieters). | 13:39:58 |
| pbsds changed their display name from pbsds (FOSDEM) to pbsds. | 16:25:08 |
| 6 Feb 2025 |
| SomeoneSerge (back on matrix) changed their display name from SomeoneSerge (Gand St. Pieters) to SomeoneSerge (UTC+U[-12,12]). | 17:49:23 |
| 13 Feb 2025 |
|  connor (he/him) (UTC-7) changed their display name from connor (he/him) (UTC-7) to connor (he/him) (UTC-8). | 06:59:20 |
| 15 Feb 2025 |
|  BenjB83 joined the room. | 10:20:29 |
| BenjB83 changed their display name from Benjamín Buske to BenjB83. | 10:43:03 |
| 17 Feb 2025 |
|  p14 joined the room. | 21:49:42 |
| 4 Mar 2025 |
|  fpletz joined the room. | 07:35:17 |
| fpletz | Are there still interested people in this room in getting something done about our cache issues, in particular to get the GC/Glacierization going? I would like to revive the meetings to document the status quo and draft a plan to move forward with the efforts. | 08:29:59 |
 flokli | Yes, I'm still interested. The problem what got me to move on was a bit of a lack of being able to make decisions, having predictability on timelines and / or commitments to a certain solution. | 11:41:52 |
| flokli | I personally also still think we should dedicate some time to properly build tooling that allows us to determine closures and how they are connected to channel bumps, so we can play with various parameters and simulate the expected outcome. Moving to different buckets per channel, another proposal I've seen is only gonna increase the amount of complexity to reason about the system as a whole, while potentially adding even more duplicates. | 11:45:59 |
 nh2 | I'm also still interested in setting up a Hetzner Ceph cluster where everything fits. | 12:28:43 |
| fpletz | If there are any decisions to be made I will try to fast track them in the SC. That's my job after all. :) But I also want to help in a technical capability. | 13:38:37 |
| flokli | We might not need an awfully large cluster to begin with. nixos.tvix.store still uses less than 1% of its disks 😄 | 14:17:07 |
 tomberek | @flokli:matrix.org: I've still not seen a comparison. Is there an estimate of the dedup and compression ratio? Compared to just NAR compression we do now? | 14:23:42 |
| flokli | How good you can dedup largely depends on your dataset, and the size of it. We did some test drives on some channel bumps, it's somewhere in discourse, need to dig up the link. | 14:27:34 |
| flokli | Someone who's more into statistics and visualization should also probably take a look at the chunking parameters and play with them a bit, the current chunking params are just some gut feeling numbers. | 14:28:33 |
| flokli | It should probably be not too much work to expose the list of store paths present, and the uncompressed nar size (uncompressed) of all Pathinfos aggregated. If you compare that with the disk usage you get the current dedup ratio. | 14:29:55 |
| flokli | It would just need to be wired up into the metrics. | 14:30:26 |
| tomberek | Okay. I still think it is reasonable to incur the cost of copying nearly everything to Glacier. Then do a GC. (Or GC first if we have high confidence.) It's something that helps and something we'd likely want regardless of any other approaches we take. | 14:35:42 |
| flokli | We would need to first find out *what* to copy to glacier. We don't have the tooling for that | 14:38:17 |
| flokli | And if we accidentally glacier something that's being requested, have an idea about the costs for this. | 14:38:38 |
| flokli | And I'd honestly rather copy the to-be-glaciered data to a self-hosted object storage and serve from there, rather than only having it somewhere where you have to pay premium to AWS to get it out. | 14:39:40 |
| flokli | But well, I can share my opinions here all day long, ultimately we need to agree on a path forward and realistic timelines attached to it, and then do this plan / empower people to do it. | 14:40:49 |
|  lassulus changed their profile picture. | 17:49:06 |
