fabriziosalmi/proxmox-lxc-autoscale v1.2.0 on GitHub

What's Changed

CPU core pinning for Intel hybrid CPUs (#7) — Pin containers to P-cores (performance) or E-cores (efficiency) on Intel Alder Lake / Raptor Lake and newer. New cpu_pinning tier config option accepts p-cores, e-cores, all, or explicit ranges like 0-11. Core topology is auto-detected from the kernel at startup.
VitePress documentation site — Full documentation rewrite with 15 structured pages, local search, sidebar navigation, and edit-on-GitHub links. Live at fabriziosalmi.github.io/proxmox-lxc-autoscale.

Fix CPU measurement always showing ~100% (#6, #23, #25) — Replaced the broken /proc/stat method with host-side cgroup accounting (cgroup v2/v1). The old method used pct exec to read /proc/stat inside containers, which without LXCFS showed host CPU (not container CPU), and the overhead itself saturated CPU on low-core hosts. The new method reads directly from the kernel with zero overhead.
Fix tier lookup never finding configured tiers (#25) — resource_manager.py used a nonexistent config key instead of LXC_TIER_ASSOCIATIONS.
Fix misleading scaling log messages — Now distinguishes "at max_cores" from "no host cores available".
Fix missing scale_in() function — Was called but never defined (horizontal scaling was broken).
Fix get_container_config() crash — Name-shadowing bug caused UnboundLocalError.
Remove dead duplicate code between lxc_utils.py and resource_manager.py.

LXCFS is no longer required — Cgroup CPU measurement works without LXCFS.
Much faster data collection — No pct exec calls for CPU measurement; just host-side file reads.
Smarter caching — Cgroup paths and previous CPU readings cached across poll cycles.
Cleaner logs — Reduced noisy INFO messages to DEBUG; compact single-line container status.

CI now passes for the first time in months — Added requirements.txt, dropped EOL Python 3.8, added Python 3.12.
VitePress docs deploy — Automated build and deploy to GitHub Pages on every push to docs/.

Full Changelog: v1.1.0...v1.2.0