#!/bin/sh # 检查环境变量 if [ -z "$HF_TOKEN" ] || [ -z "$DATASET_ID" ]; then echo "未检测到 HF_TOKEN 或 DATASET_ID,备份功能不可用" exit 1 fi # 激活虚拟环境 . ${APP_HOME}/venv/bin/activate # 生成同步脚本 cat > /tmp/hf_sync.py << 'EOL' # HuggingFace 同步脚本 from huggingface_hub import HfApi import sys import os import tarfile import tempfile def manage_backups(api, repo_id, prefix, max_files=50): files = api.list_repo_files(repo_id=repo_id, repo_type="dataset") backup_files = [f for f in files if f.startswith(prefix) and f.endswith('.tar.gz')] backup_files.sort() if len(backup_files) >= max_files: files_to_delete = backup_files[:(len(backup_files) - max_files + 1)] for file_to_delete in files_to_delete: try: api.delete_file(path_in_repo=file_to_delete, repo_id=repo_id, repo_type="dataset") print(f'已删除旧备份: {file_to_delete}') except Exception as e: print(f'删除 {file_to_delete} 时出错: {str(e)}') def upload_backup(file_path, file_name, token, repo_id, prefix): api = HfApi(token=token) try: api.upload_file( path_or_fileobj=file_path, path_in_repo=file_name, repo_id=repo_id, repo_type="dataset" ) print(f"成功上传 {file_name}") manage_backups(api, repo_id, prefix) except Exception as e: print(f"上传文件出错: {str(e)}") def download_latest_backup(token, repo_id, prefix, extract_path): try: api = HfApi(token=token) files = api.list_repo_files(repo_id=repo_id, repo_type="dataset") backup_files = [f for f in files if f.startswith(prefix) and f.endswith('.tar.gz')] if not backup_files: print(f"未找到任何备份文件(前缀: {prefix})") return latest_backup = sorted(backup_files)[-1] with tempfile.TemporaryDirectory() as temp_dir: filepath = api.hf_hub_download( repo_id=repo_id, filename=latest_backup, repo_type="dataset", local_dir=temp_dir ) if filepath and os.path.exists(filepath): try: with tarfile.open(filepath, 'r:gz') as tar: tar.extractall(extract_path) print(f"已成功恢复备份: {latest_backup} 到 {extract_path}") except Exception as e: print(f"解压备份时出错: {str(e)}") except Exception as e: print(f"下载备份出错: {str(e)}") def super_squash_history(token, repo_id): try: api = HfApi(token=token) api.super_squash_history(repo_id=repo_id, repo_type="dataset") print("历史合并完成。") except Exception as e: print(f"合并历史出错: {str(e)}") if __name__ == "__main__": action = sys.argv[1] token = sys.argv[2] repo_id = sys.argv[3] if action == "upload": file_path = sys.argv[4] file_name = sys.argv[5] prefix = sys.argv[6] upload_backup(file_path, file_name, token, repo_id, prefix) elif action == "download": prefix = sys.argv[4] extract_path = sys.argv[5] download_latest_backup(token, repo_id, prefix, extract_path) elif action == "super_squash": super_squash_history(token, repo_id) EOL # 首次启动时分别还原 config 和 tools echo "正在从 HuggingFace 下载 config 备份..." python /tmp/hf_sync.py download "${HF_TOKEN}" "${DATASET_ID}" "config_backup_" "/mcp-proxy-server/config" echo "正在从 HuggingFace 下载 tools 备份..." python /tmp/hf_sync.py download "${HF_TOKEN}" "${DATASET_ID}" "tools_backup_" "/tools" # 同步函数 sync_data() { while true; do echo "同步进程启动于 $(date)" STORAGE_PATH1="/mcp-proxy-server/config" STORAGE_PATH2="/tools" timestamp=$(date +%Y%m%d_%H%M%S) backup_file1="config_backup_${timestamp}.tar.gz" backup_file2="tools_backup_${timestamp}.tar.gz" if [ -d "${STORAGE_PATH1}" ]; then tar -czf "/tmp/${backup_file1}" -C "/mcp-proxy-server" "config" echo "正在上传 config 备份到 HuggingFace..." python /tmp/hf_sync.py upload "${HF_TOKEN}" "${DATASET_ID}" "/tmp/${backup_file1}" "${backup_file1}" "config_backup_" rm -f "/tmp/${backup_file1}" fi if [ -d "${STORAGE_PATH2}" ]; then tar -czf "/tmp/${backup_file2}" -C "/" "tools" echo "正在上传 tools 备份到 HuggingFace..." python /tmp/hf_sync.py upload "${HF_TOKEN}" "${DATASET_ID}" "/tmp/${backup_file2}" "${backup_file2}" "tools_backup_" rm -f "/tmp/${backup_file2}" fi # 合并历史提交 SQUASH_FLAG_FILE="/tmp/last_squash_time" NOW=$(date +%s) SEVEN_DAYS=$((7*24*60*60)) if [ ! -f "$SQUASH_FLAG_FILE" ]; then echo $NOW > "$SQUASH_FLAG_FILE" echo "首次合并历史提交..." python /tmp/hf_sync.py super_squash "${HF_TOKEN}" "${DATASET_ID}" else LAST=$(cat "$SQUASH_FLAG_FILE") DIFF=$((NOW - LAST)) if [ $DIFF -ge $SEVEN_DAYS ]; then echo $NOW > "$SQUASH_FLAG_FILE" echo "距离上次合并已超过7天,正在合并历史提交..." python /tmp/hf_sync.py super_squash "${HF_TOKEN}" "${DATASET_ID}" else echo "距离上次合并未满7天,本次跳过合并历史提交。" fi fi SYNC_INTERVAL=${SYNC_INTERVAL:-7200} echo "下次同步将在 ${SYNC_INTERVAL} 秒后进行..." sleep $SYNC_INTERVAL done } # 启动同步进程 sync_data & echo "同步进程已在后台启动 (PID: $!)" echo "现在启动 MCP Proxy 主程序..." # 启动原始的 MCP Proxy 程序 echo "准备启动 MCP Proxy,端口: ${PORT:-7860},主机: ${HOST:-0.0.0.0}" if [ -f build/sse.js ]; then echo "启动: node build/sse.js" exec node build/sse.js elif [ -f /app/build/sse.js ]; then echo "启动: node /app/build/sse.js" exec node /app/build/sse.js elif [ -f /mcp-proxy-server/build/sse.js ]; then echo "启动: node /mcp-proxy-server/build/sse.js" exec node /mcp-proxy-server/build/sse.js else echo "错误: 找不到 build/sse.js 文件" echo "当前目录: $(pwd)" echo "查找可能的 Node.js 文件:" find / -name "sse.js" -type f 2>/dev/null | head -5 # 如果找不到,保持容器运行以便调试 echo "保持容器运行以便调试..." tail -f /dev/null fi