vps-guardian（vps 守护者）

VPS Guardian — Self-Healing Server Agent

What It Does

VPS Guardian doesn't just alert you when something's wrong — it fixes it. An autonomous agent that monitors your VPS and takes action: killing runaway processes, freeing disk space, restarting dead services, and hardening security.

Philosophy: Alerts are for humans. Guardians act.

Quick Start

Install

# Download
sudo curl -o /usr/local/bin/vps-guardian https://raw.githubusercontent.com/vps-guardian/guardian/main/src/guardian.py
sudo chmod +x /usr/local/bin/vps-guardian

# Configure (optional — defaults are sensible)
sudo cp /usr/local/bin/vps-guardian.conf.example /etc/vps-guardian.conf

Run

# One-shot scan + remediate
vps-guardian --scan

# Daemon mode (continuous monitoring)
vps-guardian --daemon

# Dry run (report only, no action)
vps-guardian --dry-run

Schedule with Cron

# Every 5 minutes
*/5 * * * * /usr/local/bin/vps-guardian --scan >> /var/log/vps-guardian.log 2>&1

What It Fixes

1. Runaway Processes

• Detects processes consuming >90% CPU for 5+ minutes
• Identifies zombie processes (state Z)
• Finds OOM candidates (highest RSS on low memory)
• Action: SIGTERM, then SIGKILL after 10s if unresponsive
• Logs PID, command, CPU%, memory%, and duration before killing

2. Disk Space Recovery

• Warns at 80% usage, acts at 90%
• Rotates logs older than 7 days (gzip)
• Clears /tmp files older than 3 days
• Removes orphaned apt caches
• Truncates journal logs >500MB
• Action: Sequential cleanup until below threshold
• Logs bytes freed and what was cleaned

3. Dead Service Restart

• Monitors critical systemd services
• Detects failed/inactive states
• Attempts systemctl restart up to 3 times
• Escalates to alert after 3 failures
• Action: Auto-restart with exponential backoff (5s, 30s, 120s)

4. Security Hardening

• Detects unexpected listening ports (diffs against baseline)
• Identifies brute-force SSH attempts (fail2ban integration)
• Flags world-writable files in /etc, /root
• Checks for suspicious processes (crypto miners, reverse shells)
• Action: Blocks IPs via iptables, chmods files, kills suspicious processes
• Never auto-blocks — always requires approval (prevents lockout)

5. Memory Recovery

• Detects memory pressure (available <10%)
• Identifies top memory consumers
• Clears page cache, dentries, inodes
• Action: Cache drop first, process kill as last resort

Configuration

# /etc/vps-guardian.conf

[thresholds]
cpu_kill_percent = 90
cpu_kill_duration = 300
disk_warn_percent = 80
disk_act_percent = 90
memory_act_percent = 90
zombie_kill = true

[services]
# Services to auto-restart when down
critical = ssh, cron, systemd-resolved, rsyslog
# Services to monitor but NOT auto-restart
watched = nginx, docker, fail2ban

[security]
block_bruteforce = true
max_ssh_failures = 10
scan_suspicious_procs = true
auto_block = false    # Always requires approval

[cleanup]
rotate_logs_days = 7
tmp_max_age_days = 3
journal_max_size_mb = 500
apt_cache_clean = true

[alerts]
telegram_bot_token =
telegram_chat_id =
webhook_url =

Exit Codes

Logging

All actions are logged with before/after state:

[2026-04-17T06:30:00Z] RUNAWAY pid=4523 cmd="node server.js" cpu=94.2% mem=1.2G duration=420s → SIGTERM → exited
[2026-04-17T06:30:05Z] DISK / 92% → rotated /var/log/syslog.1 (freed 340MB) → 87%
[2026-04-17T06:30:10Z] SERVICE ssh: inactive → restarted → active
[2026-04-17T06:30:15Z] SECURITY 5 SSH failures from 45.33.xx.xx → blocked via iptables

Safety Mechanisms

• Protected processes: Never kills pid 1, kernel threads, or self
• Protected services: Never restarts systemd itself
• Rate limiting: Max 5 process kills per scan cycle
• Dry run mode: --dry-run shows what would happen without acting
• Whitelist: Exclude processes/services from remediation
• Approval gate: Security actions always require confirmation

Requirements

• Linux (systemd-based)
• Python 3.8+
• Root or sudo access
• No external dependencies (stdlib only)

Differences from Monitoring Tools

Support

• Issues: https://github.com/vps-guardian/guardian/issues
• Docs: https://vps-guardian.io/docs