Dashboard Overview

Purpose

Real-time monitoring dashboard for the OVH5 dev server (15.204.90.153) and its 5 LXC containers. Provides at-a-glance visibility into container health, Claude Code usage, Docker status, and system resources.

URL: https://admin.ipnoelp.io Deployed: 2026-02-28

Architecture

Two systemd-driven collector scripts gather metrics and write atomic JSON files. A zero-dependency Node.js server reads the JSON and serves a single-page dashboard through nginx.

[systemd timer: 30s] → collect-dashboard.sh → /tmp/dashboard-status.json
[systemd timer: 5m]  → collect-dashboard-slow.sh → /tmp/dashboard-status-slow.json
                                                          ↓
[browser] ← [Cloudflare Access] ← [nginx :443] ← [Node.js :3580] ← reads JSON

Design Principles

• Collectors run as root — needed for lxc-attach and cgroup reads
• Node.js runs unprivileged (user ubuntu) — reads pre-collected JSON only
• Zero npm dependencies — uses only built-in Node.js modules
• Atomic writes — collectors write to .tmp then mv to prevent partial reads
• Cloudflare Access for authentication (same policy as term.ipnoelp.com)

What Gets Monitored

Host Level (every 30 seconds)

• CPU load (1/5/15 min averages) and core count
• Memory used vs available
• Disk usage for / and /data partitions
• Service status: nginx, claude-net-tunnel, fail2ban
• Fail2ban banned IP count and detailed list

Per Container (every 30 seconds)

• State (running/stopped), IP address
• CPU load, memory (from host cgroup — accurate for containers)
• Network TX/RX bytes, uptime
• Claude Code process detection (running/PID)
• tmux session status
• Docker containers (names, images, status)
• Last 10 SSH sessions with IP, date, and duration

Per Container — Slow (every 5 minutes)

• Disk usage (rootfs size via du -sb)
• Claude Code version
• Token usage from JSONL session files (today’s total output tokens)
• Snapshot age and status

Security Model