!RROtHmAaQIkiJzJZZE:nixos.org

NixOS Infrastructure

388 Members
Next Infra call: 2024-07-11, 18:00 CEST (UTC+2) | Infra operational issues backlog: https://github.com/orgs/NixOS/projects/52 | See #infra-alerts:nixos.org for real time alerts from Prometheus.120 Servers

Load older messages


SenderMessageTime
24 Mar 2025
@emilazy:matrix.orgemilyso current growth rate is ~6.56 TiB/month, so cache growth size would be 1.15–1.3× current?14:20:34
@emilazy:matrix.orgemilyhas there been any work or even proposals for things to reduce cache size growth other than "GC everything" that would have anything like a 15–30% impact/14:24:43
@emilazy:matrix.orgemily* has there been any work or even proposals for things to reduce cache size growth other than "GC everything" that would have anything like a 15–30% impact?14:24:46
@hexa:lossy.networkhexaDownload BucketSizeBytes-2023_12_26_00_00_00-2025_03_23_00_00_00-UTC.csv14:28:22
@hexa:lossy.networkhexadaily bucket sizes for the last 36 months14:28:31
@hexa:lossy.networkhexa* daily bucket sizes for the last 15 months14:29:00
@hexa:lossy.networkhexa * that is a manual thing iirc 14:29:29
@vcunat:matrix.orgVladimír ČunátI've certainly heard proposals like GC everything that's never "been in a channel".14:29:58
@emilazy:matrix.orgemilyright, that's what i meant re: GC14:34:40
@elvishjerricco:matrix.orgElvishJerriccothis seems to indicate the cache grows by like 300G a day, doesn't it?14:43:05
@hexa:lossy.networkhexahttps://docs.google.com/spreadsheets/d/1KA2tbPsi1kQuwlHJC89iWUQBXPv6p1N3n0t73bbK7n4/edit?usp=sharing14:44:00
@emilazy:matrix.orgemilyabout in line with the 6.56 TiB/month figure14:45:16
@emilazy:matrix.orgemilyso seems accurate14:45:25
@elvishjerricco:matrix.orgElvishJerricco hexa (signing key rotation when): i think your columns are backward; you have the IA labeled as standard and vice versa 14:45:30
@hexa:lossy.networkhexalol 🙂 14:45:38
@hexa:lossy.networkhexafeel free to edit14:45:41
@hexa:lossy.networkhexaat least the trendline for daily added goes down 😄 14:50:37
@emilazy:matrix.orgemilynow add the first day with an eval that adds ~500 GiB of NVIDIA :P14:51:56
@hexa:lossy.networkhexaI'm still waiting for the rocm kernel build that exceeds 24 hours14:53:42
@hexa:lossy.networkhexabut apparently https://github.com/NixOS/nixpkgs/pull/367695 has not landed yet14:54:24
@hexa:lossy.networkhexa* but apparently https://github.com/NixOS/nixpkgs/pull/367695 has not landed so far14:54:33
@vcunat:matrix.orgVladimír Čunát That would be ~1 TB / month just from staging* rebuilds for this. 14:59:17
@emilazy:matrix.orgemilyright. I commented as much on the RFC. (from the RFC diff after the conversation earlier it seems like your earlier comment was misinterpreted as "the current normal cache growth rate is 2 TB/month compared to this being 700 GB one time"?)15:00:13
@emilazy:matrix.orgemily(rather than "this would add 2 TB/month rather than 700 GB one time, because of rebuilds")15:00:31
@emilazy:matrix.orgemilyat least, the RFC claims that the current growth rate is 2 TB/month so we shouldn't worry about a one-time increase, but neither of those seems accurate15:00:51
@elvishjerricco:matrix.orgElvishJerriccoSo how much of a problem is this? Can we absorb a 15-30% hit to cache size growth? Or does that make the RFC a non-starter from an infra perspective?15:08:38
@emilazy:matrix.orgemilythe infra team isn't the one paying the bills15:09:03
@toonn:matrix.orgtoonn Wasn't the sponsorship extension based on a commitment to reduce the size of the cache? 15:14:08
@emilazy:matrix.orgemilythat's my understanding, which I also mentioned in https://github.com/NixOS/rfcs/pull/185#issuecomment-274833920015:14:33
@emilazy:matrix.orgemilyI just mean that's more Foundation than it is infra team, AFAIU15:15:05

Show newer messages


Back to Room ListRoom Version: 6