a13labs.infra/ansible/files/scripts/auto_suspend.sh

#!/bin/bash

IDLE_TIMEOUT_MINUTES=20

# Check 1: Any logged-in users?
if who | grep -qv "tty"; then
    echo "Active user sessions found. Skipping suspend."
    exit 0
fi

# Check 2: CPU idle percentage
CPU_IDLE=$(mpstat 1 1 | tail -n 1 | awk '/:/ {print $12}' | sed 's/,/./g')
CPU_THRESHOLD=90

if (( $(echo "$CPU_IDLE < $CPU_THRESHOLD" | bc -l) )); then
    echo "CPU too active. Skipping suspend."
    exit 0
fi

if ! command -v nvidia-smi &> /dev/null; then
  echo "nvidia-smi not found, skipping GPU check"
else
    # Check 3: Is GPU idle?
    GPU_UTIL=$(nvidia-smi --query-gpu=utilization.gpu --format=csv,noheader,nounits | head -n 1)
    GPU_THRESHOLD=0

    if (( GPU_UTIL > GPU_THRESHOLD )); then
        echo "GPU is busy (utilization=${GPU_UTIL}%). Skipping suspend."
        exit 0
    fi

    # Check 4: Is OLLAMA active?
    OLLAMA_ACTIVE=$(nvidia-smi --query-compute-apps=name,used_memory --format=csv,noheader,nounits | grep ollama | wc -l)
    if (( OLLAMA_ACTIVE > 0 )); then
        echo "OLLMAMA is active. Skipping suspend."
        exit 0
    fi
fi

if | command -v amd-smi &> /dev/null; then
    # Check 5: Is AMD GPU idle?
    AMD_GPU_UTIL=$(amd-smi monitor --csv | head -n 2 | tail -n 1 | cut -d ',' -f 8)
    AMD_GPU_THRESHOLD=0

    if (( AMD_GPU_UTIL > AMD_GPU_THRESHOLD )); then
        echo "AMD GPU is busy (utilization=${AMD_GPU_UTIL}%). Skipping suspend."
        exit 0
    fi

    if amd-smi process --csv | head -n 2 | tail -n 1 | grep -q "No running processes detected"; then
        echo "No processes running on AMD GPU."
    else
        echo "Processes are running on AMD GPU. Skipping suspend."
        exit 0
    fi
else
    echo "amd-smi not found, skipping AMD GPU check"
fi

echo "System idle. Suspending..."
shutdown now