sample telemetry

bees × prom

A Grafana import and a Prometheus textfile exporter. Screenshots use generic bucket names and live sizes from 2026-08-15. No LAN IPs, no home paths.

What you are looking at

Live du -xhd1 + btrfs fi df from a two-NVMe RAID0 pool on 2026-08-15. bees was running. Apparent tree 3.2T vs btrfs Data used 2.08 TiB (bees + zstd). Hash occupancy omitted (beesstats unreadable after reboot). Grafana/Prometheus themselves were down — panels rebuilt from those numbers, not a logged-in admin UI.

Sample Grafana-style overview: bees up, 2.08 TiB of 2.22 TiB Data used
docs/assets/grafana-overview.png — health + capacity
Sample bucket sizes: models 1.6T, pinokio 436G, opt 407G, uv-cache 226G, llms 202G; apparent 3.2T vs 2.08 TiB on disk
docs/assets/grafana-buckets.png — generic pool buckets
fast-models / bees dedupe sample · 2026-08-15 · last 24h
bees
UP
Data used
2.08 TiB
Data allocated
2.22 TiB
devices (raw)
3.68 TiB
models
1.6T
pinokio
436G
opt
407G
uv-cache
226G
llms
202G
_dl
111G
github
67G
work
55G

Import

  1. Cron deploy/observability/ai_data_stats_exporter.sh every 15 minutes on the NAS host.
  2. node_exporter --collector.textfile.directory on that output dir.
  3. Prometheus scrapes the node_exporter — see prometheus-scrape.example.yml.
  4. Grafana → Import → grafana-bees-dashboard.json (uid fast-models-bees-dedupe).

Do not treat content-hash “reclaimable GB” as bees savings. That series is optional and only appears if you pass AUDIT_JSON.

Full notes: docs/observability/README.md