Skip to content

ROLE_LAB.md

Latest commit

 

History

History
51 lines (37 loc) · 1.43 KB

ROLE_LAB.md

File metadata and controls

51 lines (37 loc) · 1.43 KB
Raw

VPS Role Lab - train baseline packs without VMs

Collect metric spectrograms for common VPS roles on your builder PC, then train profile packs. No KVM/libvirt required.

Roles

Role Shape
ssh Quiet idle + rare bursts
nginx HTTP-like CPU/net (local http.server; nginx configtest if installed)
python Sustained mid CPU + memory waves
docker Bursty load + process churn
wireguard Mostly idle + rare UDP-like net bursts
panel 3x-ui style: light HTTP + periodic admin bursts

Collect

python -m sysspectogram role-lab list
python -m sysspectogram role-lab run --role nginx --out artifacts/rolelab/nginx \
  --duration 600 --synthetic-only

Listeners bind 127.0.0.1 only. WAN is not opened.

Interrupted collect is OK: role-lab train recovers a short anomaly clip if needed.

Train on PC

pip install -e '.[ml]'
python -m sysspectogram role-lab train \
  --role-dir artifacts/rolelab/nginx \
  --out artifacts/profiles/nginx/host \
  --pack dist/profile-nginx-v1.tar.gz \
  --epochs 12

Outputs:

  • artifacts/profiles/<role>/host/
  • dist/profile-<role>-v1.tar.gz + .sha256

Install on VPS with profiles install (see PROFILES.md).

Design notes

  • Anomaly clip (at least 75s) is collected after the normal window.
  • Quality filters are off for role-lab so synthetic runs still train.
  • Fine-tune on the real VPS before production trust.