PaddleX/zhch/paddle_vllm_daemon.sh

#!/bin/bash
# filepath: /home/ubuntu/zhch/PaddleX/zhch/paddle_vllm_daemon.sh

# PaddleOCR-VL vLLM 服务守护进程脚本

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

# 配置参数
CONDA_ENV="paddle"  # 根据你的环境调整
PORT="8110"
HOST="0.0.0.0"
MODEL_NAME="PaddleOCR-VL-0.9B"
BACKEND="vllm"

# GPU 配置
GPU_MEMORY_UTILIZATION="0.3"
CUDA_VISIBLE_DEVICES="3"  # 使用3号显卡
MAX_MODEL_LEN="16384"
MAX_NUM_BATCHED_TOKENS="8192"
MAX_NUM_SEQS="8"

# PaddleX 环境变量
export PADDLE_PDX_MODEL_SOURCE="bos"
export PYTHONWARNINGS="ignore::UserWarning"

# 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 $CONDA_ENV
else
    echo "Warning: Using direct conda path activation"
    export PATH="/home/ubuntu/anaconda3/envs/$CONDA_ENV/bin:$PATH"
fi

start() {
    if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "PaddleOCR-VL vLLM is already running"
        return 1
    fi
    
    echo "Starting PaddleOCR-VL vLLM daemon..."
    echo "Host: $HOST, Port: $PORT"
    echo "Model: $MODEL_NAME, Backend: $BACKEND"
    echo "GPU memory utilization: $GPU_MEMORY_UTILIZATION"
    echo "CUDA devices: $CUDA_VISIBLE_DEVICES"
    
    # 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paddlex_genai_server命令
    if ! command -v paddlex_genai_server >/dev/null 2>&1; then
        echo "❌ paddlex_genai_server not found. Please install vllm-server plugin:"
        echo "   paddlex --install genai-vllm-server"
        return 1
    fi
    
    echo "🔧 Using Python: $(which python)"
    echo "🔧 Using paddlex_genai_server: $(which paddlex_genai_server)"
    
    # 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 | \
        grep "^$CUDA_VISIBLE_DEVICES," | \
        awk -F',' '{printf "  GPU %s: %s - 内存: %sMB/%sMB\n", $1, $2, $3, $4}'
    else
        echo "⚠️  nvidia-smi not available"
    fi
    
    # 启动PaddleOCR-VL vLLM服务
    CUDA_VISIBLE_DEVICES=$CUDA_VISIBLE_DEVICES nohup paddlex_genai_server \
        --model_name $MODEL_NAME \
        --backend $BACKEND \
        --host $HOST \
        --port $PORT \
        --backend_config <(cat <<EOF
gpu-memory-utilization: $GPU_MEMORY_UTILIZATION
EOF
) > $LOGFILE 2>&1 &
    
    echo $! > $PIDFILE
    echo "✅ PaddleOCR-VL vLLM started with PID: $(cat $PIDFILE)"
    echo "📋 Log file: $LOGFILE"
    echo "🌐 Service URL: http://$HOST:$PORT"
    echo "📖 API Documentation: http://localhost:$PORT/docs"
    echo ""
    echo "Waiting for service to start..."
    sleep 5
    status
}

stop() {
    if [ ! -f $PIDFILE ]; then
        echo "PaddleOCR-VL vLLM is not running"
        return 1
    fi
    
    PID=$(cat $PIDFILE)
    echo "Stopping PaddleOCR-VL vLLM (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 "✅ PaddleOCR-VL vLLM stopped"
}

status() {
    if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
        PID=$(cat $PIDFILE)
        echo "✅ PaddleOCR-VL vLLM 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 | \
            grep "^$CUDA_VISIBLE_DEVICES," | \
            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 "❌ PaddleOCR-VL vLLM is not running"
        if [ -f $PIDFILE ]; then
            echo "Removing stale PID file..."
            rm -f $PIDFILE
        fi
    fi
}

logs() {
    if [ -f $LOGFILE ]; then
        echo "📄 PaddleOCR-VL vLLM 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 Name: $MODEL_NAME"
    echo "  Backend: $BACKEND"
    echo "  GPU Memory Utilization: $GPU_MEMORY_UTILIZATION"
    echo "  CUDA Visible Devices: $CUDA_VISIBLE_DEVICES"
    echo "  Max Model Length: $MAX_MODEL_LEN"
    echo "  Max Num Seqs: $MAX_NUM_SEQS"
    echo "  PID File: $PIDFILE"
    echo "  Log File: $LOGFILE"
    echo ""
    echo "  Model Source: ${PADDLE_PDX_MODEL_SOURCE:-default}"
    
    # 显示环境信息
    echo ""
    echo "🔧 Environment:"
    echo "  Python: $(which python 2>/dev/null || echo 'Not found')"
    echo "  paddlex_genai_server: $(which paddlex_genai_server 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 | \
        grep "^$CUDA_VISIBLE_DEVICES," | \
        awk -F',' '{printf "  GPU %s: %s (Driver: %s, Memory: %sMB)\n", $1, $2, $3, $4}'
    fi
}

test_api() {
    echo "🧪 Testing PaddleOCR-VL vLLM API..."
    
    if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "❌ PaddleOCR-VL 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
    
    echo ""
    echo "📡 Testing health endpoint..."
    health_response=$(curl -s --connect-timeout 5 http://127.0.0.1:$PORT/health)
    if [ $? -eq 0 ]; then
        echo "✅ Health endpoint accessible"
        echo "$health_response"
    else
        echo "❌ Health endpoint not accessible"
    fi
}

test_client() {
    echo "🧪 Testing PaddleOCR-VL client with vLLM server..."
    
    if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
        echo "❌ PaddleOCR-VL vLLM service is not running. Start it first with: $0 start"
        return 1
    fi
    
    # 测试用例文件路径
    TEST_IMAGE="/home/ubuntu/zhch/data/至远彩色印刷工业有限公司/2023年度报告母公司.img/2023年度报告母公司_page_006.png"
    TEST_OUTPUT="/tmp/paddleocr_vl_vllm_test_output"
    PIPELINE_CONFIG="/home/ubuntu/zhch/PaddleX/zhch/my_config/PaddleOCR-VL-Client.yaml"
    
    if [ ! -f "$TEST_IMAGE" ]; then
        echo "⚠️  Test image not found: $TEST_IMAGE"
        echo "Please provide a test image or update the TEST_IMAGE path in the script"
        return 1
    fi
    
    if [ ! -f "$PIPELINE_CONFIG" ]; then
        echo "⚠️  Pipeline config not found: $PIPELINE_CONFIG"
        echo "Please update the PIPELINE_CONFIG path in the script"
        return 1
    fi
    
    echo "📄 Testing with image: $TEST_IMAGE"
    echo "⚙️  Using pipeline config: $PIPELINE_CONFIG"
    echo "📁 Output directory: $TEST_OUTPUT"
    echo ""
    
    # 方法1: 使用 paddlex 命令行 (推荐)
    echo "🔧 Using paddlex CLI..."
    mkdir -p "$TEST_OUTPUT"
    
    paddlex --pipeline "$PIPELINE_CONFIG" \
            --input "$TEST_IMAGE" \
            --save_path "$TEST_OUTPUT" \
            --use_doc_orientation_classify False \
            --use_doc_unwarping False
    
    if [ $? -eq 0 ]; then
        echo "✅ CLI test completed successfully"
        echo "📁 Results saved to: $TEST_OUTPUT"
        
        # 显示生成的文件
        if [ -d "$TEST_OUTPUT" ]; then
            echo ""
            echo "📂 Generated files:"
            ls -lh "$TEST_OUTPUT" | tail -n +2 | awk '{print "  " $9 " (" $5 ")"}'
        fi
    else
        echo "❌ CLI test failed"
        return 1
    fi
    
}

# 显示使用帮助
usage() {
    echo "PaddleOCR-VL vLLM Service Daemon"
    echo "================================="
    echo "Usage: $0 {start|stop|restart|status|logs|config|test|test-client}"
    echo ""
    echo "Commands:"
    echo "  start       - Start the PaddleOCR-VL vLLM service"
    echo "  stop        - Stop the PaddleOCR-VL vLLM service"
    echo "  restart     - Restart the PaddleOCR-VL vLLM 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 "  test-client - Test PaddleX client with vLLM server"
    echo ""
    echo "Configuration (edit script to modify):"
    echo "  Host: $HOST"
    echo "  Port: $PORT"
    echo "  Model: $MODEL_NAME"
    echo "  Backend: $BACKEND"
    echo "  GPU Memory: $GPU_MEMORY_UTILIZATION"
    echo "  CUDA Devices: $CUDA_VISIBLE_DEVICES"
    echo ""
    echo "Examples:"
    echo "  ./paddle_vllm_daemon.sh start"
    echo "  ./paddle_vllm_daemon.sh status"
    echo "  ./paddle_vllm_daemon.sh logs"
    echo "  ./paddle_vllm_daemon.sh test"
    echo "  ./paddle_vllm_daemon.sh test-client"
}

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