mineru_vllm_daemon.sh 12 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390
  1. #!/bin/bash
  2. # filepath: /home/ubuntu/zhch/MinerU/zhch/mineru_vllm_daemon.sh
  3. # MinerU vLLM 服务守护进程脚本
  4. LOGDIR="/home/ubuntu/zhch/logs"
  5. mkdir -p $LOGDIR
  6. PIDFILE="$LOGDIR/mineru_vllm.pid"
  7. LOGFILE="$LOGDIR/mineru_vllm.log"
  8. # 配置参数
  9. CONDA_ENV="mineru2"
  10. PORT="8121"
  11. HOST="0.0.0.0"
  12. MODEL_PATH="/home/ubuntu/models/modelscope_cache/models/OpenDataLab/MinerU2___5-2509-1___2B"
  13. MODEL_NAME="MinerU2.5"
  14. # GPU 配置
  15. GPU_MEMORY_UTILIZATION="0.3"
  16. CUDA_VISIBLE_DEVICES="4"
  17. MAX_MODEL_LEN="16384"
  18. MAX_NUM_BATCHED_TOKENS="8192"
  19. MAX_NUM_SEQS="8"
  20. # MinerU 配置
  21. export MINERU_TOOLS_CONFIG_JSON="/home/ubuntu/zhch/MinerU/mineru.json"
  22. export MODELSCOPE_CACHE="/home/ubuntu/models/modelscope_cache"
  23. export USE_MODELSCOPE_HUB=1
  24. # export CUDA_VISIBLE_DEVICES="$CUDA_VISIBLE_DEVICES"
  25. # export NLTK_DATA="/home/ubuntu/nltk_data"
  26. # export HF_HOME="/home/ubuntu/models/hf_home"
  27. # export HF_ENDPOINT="https://hf-mirror.com"
  28. # export TORCH_HOME="/home/ubuntu/models/torch/"
  29. # 正确初始化和激活conda环境
  30. if [ -f "/home/ubuntu/anaconda3/etc/profile.d/conda.sh" ]; then
  31. source /home/ubuntu/anaconda3/etc/profile.d/conda.sh
  32. conda activate $CONDA_ENV
  33. elif [ -f "/opt/conda/etc/profile.d/conda.sh" ]; then
  34. source /opt/conda/etc/profile.d/conda.sh
  35. conda activate $CONDA_ENV
  36. else
  37. # 方法2:直接使用conda可执行文件路径
  38. echo "Warning: Using direct conda path activation"
  39. export PATH="/home/ubuntu/anaconda3/envs/$CONDA_ENV/bin:$PATH"
  40. fi
  41. # 设置环境变量
  42. # export LD_LIBRARY_PATH="/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH"
  43. # export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
  44. start() {
  45. if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
  46. echo "MinerU vLLM is already running"
  47. return 1
  48. fi
  49. echo "Starting MinerU vLLM daemon..."
  50. echo "Host: $HOST, Port: $PORT"
  51. echo "Model path: $MODEL_PATH"
  52. echo "GPU memory utilization: $GPU_MEMORY_UTILIZATION"
  53. echo "CUDA devices: $CUDA_VISIBLE_DEVICES"
  54. # 检查模型文件是否存在
  55. if [ ! -d "$MODEL_PATH" ]; then
  56. echo "❌ Model path not found: $MODEL_PATH"
  57. echo "Please download the model first:"
  58. echo "python -m mineru.cli.models_download"
  59. return 1
  60. fi
  61. # 检查conda环境
  62. if ! command -v python >/dev/null 2>&1; then
  63. echo "❌ Python not found. Check conda environment activation."
  64. return 1
  65. fi
  66. # 检查mineru-vllm-server命令
  67. if ! command -v mineru-vllm-server >/dev/null 2>&1; then
  68. echo "❌ mineru-vllm-server not found. Check installation and environment."
  69. return 1
  70. fi
  71. echo "🔧 Using Python: $(which python)"
  72. echo "🔧 Using mineru-vllm-server: $(which mineru-vllm-server)"
  73. # 显示GPU状态
  74. echo "📊 GPU 状态检查:"
  75. if command -v nvidia-smi >/dev/null 2>&1; then
  76. nvidia-smi --query-gpu=index,name,memory.used,memory.total --format=csv,noheader,nounits | \
  77. awk -F',' '{printf " GPU %s: %s - 内存: %sMB/%sMB\n", $1, $2, $3, $4}'
  78. else
  79. echo "⚠️ nvidia-smi not available"
  80. fi
  81. # 启动MinerU vLLM服务
  82. nohup $CUDA_VISIBLE_DEVICES; mineru-vllm-server \
  83. --host $HOST \
  84. --port $PORT \
  85. --gpu-memory-utilization $GPU_MEMORY_UTILIZATION \
  86. > $LOGFILE 2>&1 &
  87. echo $! > $PIDFILE
  88. echo "✅ MinerU vLLM started with PID: $(cat $PIDFILE)"
  89. echo "📋 Log file: $LOGFILE"
  90. echo "🌐 Service URL: http://$HOST:$PORT"
  91. echo "📖 API Documentation: http://localhost:$PORT/docs"
  92. echo ""
  93. echo "Waiting for service to start..."
  94. sleep 5
  95. status
  96. }
  97. stop() {
  98. if [ ! -f $PIDFILE ]; then
  99. echo "MinerU vLLM is not running"
  100. return 1
  101. fi
  102. PID=$(cat $PIDFILE)
  103. echo "Stopping MinerU vLLM (PID: $PID)..."
  104. # 优雅停止
  105. kill $PID
  106. # 等待进程结束
  107. for i in {1..10}; do
  108. if ! kill -0 $PID 2>/dev/null; then
  109. break
  110. fi
  111. echo "Waiting for process to stop... ($i/10)"
  112. sleep 1
  113. done
  114. # 如果进程仍在运行,强制结束
  115. if kill -0 $PID 2>/dev/null; then
  116. echo "Force killing process..."
  117. kill -9 $PID
  118. fi
  119. rm -f $PIDFILE
  120. echo "✅ MinerU vLLM stopped"
  121. }
  122. status() {
  123. if [ -f $PIDFILE ] && kill -0 $(cat $PIDFILE) 2>/dev/null; then
  124. PID=$(cat $PIDFILE)
  125. echo "✅ MinerU vLLM is running (PID: $PID)"
  126. echo "🌐 Service URL: http://$HOST:$PORT"
  127. echo "📋 Log file: $LOGFILE"
  128. # 检查端口是否被监听
  129. if command -v ss >/dev/null 2>&1; then
  130. if ss -tuln | grep -q ":$PORT "; then
  131. echo "🔗 Port $PORT is being listened"
  132. else
  133. echo "⚠️ Port $PORT is not being listened (service may be starting up)"
  134. fi
  135. elif command -v netstat >/dev/null 2>&1; then
  136. if netstat -tuln | grep -q ":$PORT "; then
  137. echo "🔗 Port $PORT is being listened"
  138. else
  139. echo "⚠️ Port $PORT is not being listened (service may be starting up)"
  140. fi
  141. fi
  142. # 检查API响应
  143. if command -v curl >/dev/null 2>&1; then
  144. if curl -s --connect-timeout 2 http://127.0.0.1:$PORT/v1/models > /dev/null 2>&1; then
  145. echo "🎯 API 响应正常"
  146. else
  147. echo "⚠️ API 无响应 (service may be starting up)"
  148. fi
  149. fi
  150. # 显示GPU使用情况
  151. if command -v nvidia-smi >/dev/null 2>&1; then
  152. echo "📊 GPU 使用情况:"
  153. nvidia-smi --query-gpu=index,utilization.gpu,utilization.memory,memory.used,memory.total --format=csv,noheader,nounits | \
  154. 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}'
  155. fi
  156. # 显示最新日志
  157. if [ -f $LOGFILE ]; then
  158. echo "📄 Latest logs (last 3 lines):"
  159. tail -3 $LOGFILE | sed 's/^/ /'
  160. fi
  161. else
  162. echo "❌ MinerU vLLM is not running"
  163. if [ -f $PIDFILE ]; then
  164. echo "Removing stale PID file..."
  165. rm -f $PIDFILE
  166. fi
  167. fi
  168. }
  169. logs() {
  170. if [ -f $LOGFILE ]; then
  171. echo "📄 MinerU vLLM logs:"
  172. echo "=================="
  173. tail -f $LOGFILE
  174. else
  175. echo "❌ Log file not found: $LOGFILE"
  176. fi
  177. }
  178. config() {
  179. echo "📋 Current configuration:"
  180. echo " Conda Environment: $CONDA_ENV"
  181. echo " Host: $HOST"
  182. echo " Port: $PORT"
  183. echo " Model Path: $MODEL_PATH"
  184. echo " Model Name: $MODEL_NAME"
  185. echo " GPU Memory Utilization: $GPU_MEMORY_UTILIZATION"
  186. echo " CUDA Visible Devices: $CUDA_VISIBLE_DEVICES"
  187. echo " Max Model Length: $MAX_MODEL_LEN"
  188. echo " Max Num Seqs: $MAX_NUM_SEQS"
  189. echo " PID File: $PIDFILE"
  190. echo " Log File: $LOGFILE"
  191. echo ""
  192. echo " MinerU Config: $MINERU_TOOLS_CONFIG_JSON"
  193. echo " ModelScope Cache: $MODELSCOPE_CACHE"
  194. if [ -d "$MODEL_PATH" ]; then
  195. echo "✅ Model path exists"
  196. echo " Model files:"
  197. ls -la "$MODEL_PATH" | head -10 | sed 's/^/ /'
  198. if [ $(ls -1 "$MODEL_PATH" | wc -l) -gt 10 ]; then
  199. echo " ... and more files"
  200. fi
  201. else
  202. echo "❌ Model path not found"
  203. fi
  204. # 检查MinerU配置文件
  205. if [ -f "$MINERU_TOOLS_CONFIG_JSON" ]; then
  206. echo "✅ MinerU config file exists"
  207. else
  208. echo "❌ MinerU config file not found: $MINERU_TOOLS_CONFIG_JSON"
  209. fi
  210. # 显示环境信息
  211. echo ""
  212. echo "🔧 Environment:"
  213. echo " Python: $(which python 2>/dev/null || echo 'Not found')"
  214. echo " mineru-vllm-server: $(which mineru-vllm-server 2>/dev/null || echo 'Not found')"
  215. echo " Conda: $(which conda 2>/dev/null || echo 'Not found')"
  216. echo " CUDA: $(which nvcc 2>/dev/null || echo 'Not found')"
  217. # 显示GPU信息
  218. if command -v nvidia-smi >/dev/null 2>&1; then
  219. echo ""
  220. echo "🔥 GPU Information:"
  221. nvidia-smi --query-gpu=index,name,driver_version,memory.total --format=csv,noheader,nounits | \
  222. awk -F',' '{printf " GPU %s: %s (Driver: %s, Memory: %sMB)\n", $1, $2, $3, $4}'
  223. fi
  224. }
  225. test_api() {
  226. echo "🧪 Testing MinerU vLLM API..."
  227. if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
  228. echo "❌ MinerU vLLM service is not running"
  229. return 1
  230. fi
  231. if ! command -v curl >/dev/null 2>&1; then
  232. echo "❌ curl command not found"
  233. return 1
  234. fi
  235. echo "📡 Testing /v1/models endpoint..."
  236. response=$(curl -s --connect-timeout 5 http://127.0.0.1:$PORT/v1/models)
  237. if [ $? -eq 0 ]; then
  238. echo "✅ Models endpoint accessible"
  239. echo "$response" | python -m json.tool 2>/dev/null || echo "$response"
  240. else
  241. echo "❌ Models endpoint not accessible"
  242. fi
  243. echo ""
  244. echo "📡 Testing health endpoint..."
  245. health_response=$(curl -s --connect-timeout 5 http://127.0.0.1:$PORT/health)
  246. if [ $? -eq 0 ]; then
  247. echo "✅ Health endpoint accessible"
  248. echo "$health_response"
  249. else
  250. echo "❌ Health endpoint not accessible"
  251. fi
  252. }
  253. test_client() {
  254. echo "🧪 Testing MinerU client with vLLM server..."
  255. if [ ! -f $PIDFILE ] || ! kill -0 $(cat $PIDFILE) 2>/dev/null; then
  256. echo "❌ MinerU vLLM service is not running. Start it first with: $0 start"
  257. return 1
  258. fi
  259. # 测试用例文件路径(需要根据实际情况调整)
  260. TEST_IMAGE="/home/ubuntu/zhch/data/test/sample.png"
  261. TEST_OUTPUT="/tmp/mineru_vllm_test_output"
  262. if [ ! -f "$TEST_IMAGE" ]; then
  263. echo "⚠️ Test image not found: $TEST_IMAGE"
  264. echo "Please provide a test image or update the TEST_IMAGE path in the script"
  265. return 1
  266. fi
  267. echo "📄 Testing with image: $TEST_IMAGE"
  268. echo "📁 Output directory: $TEST_OUTPUT"
  269. # 使用HTTP客户端连接到vLLM服务器
  270. python -m mineru.cli.client \
  271. -p "$TEST_IMAGE" \
  272. -o "$TEST_OUTPUT" \
  273. --backend vlm-http-client \
  274. --server-url "http://127.0.0.1:$PORT"
  275. if [ $? -eq 0 ]; then
  276. echo "✅ Client test completed successfully"
  277. echo "📁 Check output in: $TEST_OUTPUT"
  278. else
  279. echo "❌ Client test failed"
  280. fi
  281. }
  282. # 显示使用帮助
  283. usage() {
  284. echo "MinerU vLLM Service Daemon"
  285. echo "=========================="
  286. echo "Usage: $0 {start|stop|restart|status|logs|config|test|test-client}"
  287. echo ""
  288. echo "Commands:"
  289. echo " start - Start the MinerU vLLM service"
  290. echo " stop - Stop the MinerU vLLM service"
  291. echo " restart - Restart the MinerU vLLM service"
  292. echo " status - Show service status and resource usage"
  293. echo " logs - Show service logs (follow mode)"
  294. echo " config - Show current configuration"
  295. echo " test - Test API endpoints"
  296. echo " test-client - Test MinerU client with vLLM server"
  297. echo ""
  298. echo "Configuration (edit script to modify):"
  299. echo " Host: $HOST"
  300. echo " Port: $PORT"
  301. echo " Model: $MODEL_PATH"
  302. echo " GPU Memory: $GPU_MEMORY_UTILIZATION"
  303. echo " CUDA Devices: $CUDA_VISIBLE_DEVICES"
  304. echo ""
  305. echo "Examples:"
  306. echo " ./mineru_vllm_daemon.sh start"
  307. echo " ./mineru_vllm_daemon.sh status"
  308. echo " ./mineru_vllm_daemon.sh logs"
  309. echo " ./mineru_vllm_daemon.sh test"
  310. echo " ./mineru_vllm_daemon.sh test-client"
  311. }
  312. case "$1" in
  313. start)
  314. start
  315. ;;
  316. stop)
  317. stop
  318. ;;
  319. restart)
  320. stop
  321. sleep 3
  322. start
  323. ;;
  324. status)
  325. status
  326. ;;
  327. logs)
  328. logs
  329. ;;
  330. config)
  331. config
  332. ;;
  333. test)
  334. test_api
  335. ;;
  336. test-client)
  337. test_client
  338. ;;
  339. *)
  340. usage
  341. exit 1
  342. ;;
  343. esac