dots.ocr/zhch/vllm_daemon.sh

#!/bin/bash
# filepath: /home/ubuntu/zhch/dots.ocr/zhch/vllm_daemon.sh

# DotsOCR vLLM 服务守护进程脚本

LOGDIR="/home/ubuntu/zhch/logs"
mkdir -p $LOGDIR
PIDFILE="$LOGDIR/vllm.pid"
LOGFILE="$LOGDIR/vllm.log"

# 配置参数
CONDA_ENV="dots.ocr"
PORT="8101"
HOST="0.0.0.0"
HF_MODEL_PATH="/home/ubuntu/zhch/dots.ocr/weights/DotsOCR"
MODEL_NAME="DotsOCR"

# GPU 配置
GPU_MEMORY_UTILIZATION="0.70"
DATA_PARALLEL_SIZE="1"  # 从3改为1，避免内存不足
MAX_MODEL_LEN="32768"
MAX_NUM_BATCHED_TOKENS="32768"
MAX_NUM_SEQS="16"

# 正确初始化和激活conda环境
if [ -f "/home/ubuntu/anaconda3/etc/profile.d/conda.sh" ]; then
    source /home/ubuntu/anaconda3/etc/profile.d/conda.sh
    conda activate $CONDA_ENV
elif [ -f "/opt/conda/etc/profile.d/conda.sh" ]; then
    source /opt/conda/etc/profile.d/conda.sh
    conda activate $CONDA_ENV
else
    # 方法2：直接使用conda可执行文件路径
    echo "Warning: Using direct conda path activation"
    export PATH="/home/ubuntu/anaconda3/envs/$CONDA_ENV/bin:$PATH"
fi

# 设置环境变量
export PYTHONPATH=$(dirname "$HF_MODEL_PATH"):$PYTHONPATH
export LD_LIBRARY_PATH="/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH"
export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True

# 注册 DotsOCR 模型到 vLLM
register_model() {
    echo "🔧 注册 DotsOCR 模型到 vLLM..."
    vllm_path=$(which vllm)
    if [ -z "$vllm_path" ]; then
        echo "❌ vLLM 未找到，请检查安装和环境激活"
        return 1
    fi
    
    if ! grep -q "from DotsOCR import modeling_dots_ocr_vllm" "$vllm_path"; then
        sed -i '/^from vllm\.entrypoints\.cli\.main import main$/a\
from DotsOCR import modeling_dots_ocr_vllm' "$vllm_path"
        echo "✅ DotsOCR 模型已注册到 vLLM"
    else
        echo "✅ DotsOCR 模型已经注册过了"
    fi
}

start() {
    if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "vLLM DotsOCR is already running"
        return 1
    fi
    
    echo "Starting vLLM DotsOCR daemon..."
    echo "Host: $HOST, Port: $PORT"
    echo "Model path: $HF_MODEL_PATH"
    echo "GPU memory utilization: $GPU_MEMORY_UTILIZATION"
    echo "Data parallel size: $DATA_PARALLEL_SIZE"
    
    # 检查模型文件是否存在
    if [ ! -d "$HF_MODEL_PATH" ]; then
        echo "❌ Model path not found: $HF_MODEL_PATH"
        return 1
    fi
    
    # 检查conda环境
    if ! command -v python >/dev/null 2>&1; then
        echo "❌ Python not found. Check conda environment activation."
        return 1
    fi
    
    # 检查vllm命令
    if ! command -v vllm >/dev/null 2>&1; then
        echo "❌ vLLM not found. Check installation and environment."
        return 1
    fi
    
    echo "🔧 Using Python: $(which python)"
    echo "🔧 Using vLLM: $(which vllm)"
    
    # 注册模型
    register_model
    if [ $? -ne 0 ]; then
        echo "❌ Model registration failed"
        return 1
    fi
    
    # 显示GPU状态
    echo "📊 GPU 状态检查:"
    if command -v nvidia-smi >/dev/null 2>&1; then
        nvidia-smi --query-gpu=index,name,memory.used,memory.total --format=csv,noheader,nounits | \
        awk -F',' '{printf "  GPU %s: %s - 内存: %sMB/%sMB\n", $1, $2, $3, $4}'
    else
        echo "⚠️  nvidia-smi not available"
    fi
    
    # 启动vLLM服务
    nohup vllm serve $HF_MODEL_PATH \
        --host $HOST \
        --port $PORT \
        --gpu-memory-utilization $GPU_MEMORY_UTILIZATION \
        --max-log-len 1000 \
        --trust-remote-code \
        --max-model-len $MAX_MODEL_LEN \
        --max-num-batched-tokens $MAX_NUM_BATCHED_TOKENS \
        --uvicorn-log-level info \
        --limit-mm-per-prompt '{"image": 1}' \
        --chat-template-content-format string \
        --data-parallel-size $DATA_PARALLEL_SIZE \
        --max-num-seqs $MAX_NUM_SEQS \
        --enable-prefix-caching \
        --served-model-name $MODEL_NAME \
        > $LOGFILE 2>&1 &
    
    echo $! > $PIDFILE
    echo "✅ vLLM DotsOCR started with PID: $(cat $PIDFILE)"
    echo "📋 Log file: $LOGFILE"
    echo "🌐 Service URL: http://$HOST:$PORT"
    echo "📖 API Documentation: http://localhost:$PORT/docs"
}

stop() {
    if [ ! -f $PIDFILE ]; then
        echo "vLLM DotsOCR is not running"
        return 1
    fi
    
    PID=$(cat $PIDFILE)
    echo "Stopping vLLM DotsOCR (PID: $PID)..."
    
    # 优雅停止
    kill $PID
    
    # 等待进程结束
    for i in {1..10}; do
        if ! kill -0 $PID 2>/dev/null; then
            break
        fi
        echo "Waiting for process to stop... ($i/10)"
        sleep 1
    done
    
    # 如果进程仍在运行，强制结束
    if kill -0 $PID 2>/dev/null; then
        echo "Force killing process..."
        kill -9 $PID
    fi
    
    rm -f $PIDFILE
    echo "✅ vLLM DotsOCR stopped"
}

status() {
    if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
        PID=$(cat $PIDFILE)
        echo "✅ vLLM DotsOCR is running (PID: $PID)"
        echo "🌐 Service URL: http://$HOST:$PORT"
        echo "📋 Log file: $LOGFILE"
        
        # 检查端口是否被监听
        if command -v ss >/dev/null 2>&1; then
            if ss -tuln | grep -q ":$PORT "; then
                echo "🔗 Port $PORT is being listened"
            else
                echo "⚠️  Port $PORT is not being listened (service may be starting up)"
            fi
        elif command -v netstat >/dev/null 2>&1; then
            if netstat -tuln | grep -q ":$PORT "; then
                echo "🔗 Port $PORT is being listened"
            else
                echo "⚠️  Port $PORT is not being listened (service may be starting up)"
            fi
        fi
        
        # 检查API响应
        if command -v curl >/dev/null 2>&1; then
            if curl -s --connect-timeout 2 http://127.0.0.1:$PORT/v1/models > /dev/null 2>&1; then
                echo "🎯 API 响应正常"
            else
                echo "⚠️  API 无响应 (service may be starting up)"
            fi
        fi
        
        # 显示GPU使用情况
        if command -v nvidia-smi >/dev/null 2>&1; then
            echo "📊 GPU 使用情况:"
            nvidia-smi --query-gpu=index,utilization.gpu,utilization.memory,memory.used,memory.total --format=csv,noheader,nounits | \
            awk -F',' '{printf "  GPU %s: GPU利用率 %s%%, 内存利用率 %s%%, 显存 %sMB/%sMB\n", $1, $2, $3, $4, $5}'
        fi
        
        # 显示最新日志
        if [ -f $LOGFILE ]; then
            echo "📄 Latest logs (last 3 lines):"
            tail -3 $LOGFILE | sed 's/^/  /'
        fi
    else
        echo "❌ vLLM DotsOCR is not running"
        if [ -f $PIDFILE ]; then
            echo "Removing stale PID file..."
            rm -f $PIDFILE
        fi
    fi
}

logs() {
    if [ -f $LOGFILE ]; then
        echo "📄 vLLM DotsOCR logs:"
        echo "=================="
        tail -f $LOGFILE
    else
        echo "❌ Log file not found: $LOGFILE"
    fi
}

config() {
    echo "📋 Current configuration:"
    echo "  Conda Environment: $CONDA_ENV"
    echo "  Host: $HOST"
    echo "  Port: $PORT"
    echo "  Model Path: $HF_MODEL_PATH"
    echo "  Model Name: $MODEL_NAME"
    echo "  GPU Memory Utilization: $GPU_MEMORY_UTILIZATION"
    echo "  Data Parallel Size: $DATA_PARALLEL_SIZE"
    echo "  Max Model Length: $MAX_MODEL_LEN"
    echo "  Max Num Seqs: $MAX_NUM_SEQS"
    echo "  PID File: $PIDFILE"
    echo "  Log File: $LOGFILE"
    
    if [ -d "$HF_MODEL_PATH" ]; then
        echo "✅ Model path exists"
        echo "  Model files:"
        ls -la "$HF_MODEL_PATH" | head -10 | sed 's/^/    /'
        if [ $(ls -1 "$HF_MODEL_PATH" | wc -l) -gt 10 ]; then
            echo "    ... and more files"
        fi
    else
        echo "❌ Model path not found"
    fi
    
    # 显示环境信息
    echo ""
    echo "🔧 Environment:"
    echo "  Python: $(which python 2>/dev/null || echo 'Not found')"
    echo "  vLLM: $(which vllm 2>/dev/null || echo 'Not found')"
    echo "  Conda: $(which conda 2>/dev/null || echo 'Not found')"
    echo "  CUDA: $(which nvcc 2>/dev/null || echo 'Not found')"
    
    # 显示GPU信息
    if command -v nvidia-smi >/dev/null 2>&1; then
        echo ""
        echo "🔥 GPU Information:"
        nvidia-smi --query-gpu=index,name,driver_version,memory.total --format=csv,noheader,nounits | \
        awk -F',' '{printf "  GPU %s: %s (Driver: %s, Memory: %sMB)\n", $1, $2, $3, $4}'
    fi
}

test_api() {
    echo "🧪 Testing vLLM DotsOCR API..."
    
    if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "❌ vLLM service is not running"
        return 1
    fi
    
    if ! command -v curl >/dev/null 2>&1; then
        echo "❌ curl command not found"
        return 1
    fi
    
    echo "📡 Testing /v1/models endpoint..."
    response=$(curl -s --connect-timeout 5 http://127.0.0.1:$PORT/v1/models)
    if [ $? -eq 0 ]; then
        echo "✅ Models endpoint accessible"
        echo "$response" | python -m json.tool 2>/dev/null || echo "$response"
    else
        echo "❌ Models endpoint not accessible"
    fi
}

# 显示使用帮助
usage() {
    echo "vLLM DotsOCR Service Daemon"
    echo "============================"
    echo "Usage: $0 {start|stop|restart|status|logs|config|test}"
    echo ""
    echo "Commands:"
    echo "  start   - Start the vLLM DotsOCR service"
    echo "  stop    - Stop the vLLM DotsOCR service"
    echo "  restart - Restart the vLLM DotsOCR service"
    echo "  status  - Show service status and resource usage"
    echo "  logs    - Show service logs (follow mode)"
    echo "  config  - Show current configuration"
    echo "  test    - Test API endpoints"
    echo ""
    echo "Configuration (edit script to modify):"
    echo "  Host: $HOST"
    echo "  Port: $PORT"
    echo "  Model: $HF_MODEL_PATH"
    echo "  GPU Memory: $GPU_MEMORY_UTILIZATION"
    echo "  Parallel Size: $DATA_PARALLEL_SIZE"
    echo ""
    echo "Examples:"
    echo "  ./vllm_daemon.sh start"
    echo "  ./vllm_daemon.sh status"
    echo "  ./vllm_daemon.sh logs"
    echo "  ./vllm_daemon.sh test"
}

case "$1" in
    start)
        start
        ;;
    stop)
        stop
        ;;
    restart)
        stop
        sleep 3
        start
        ;;
    status)
        status
        ;;
    logs)
        logs
        ;;
    config)
        config
        ;;
    test)
        test_api
        ;;
    *)
        usage
        exit 1
        ;;
esac