app.py

from flask import Flask, request, jsonify
from sentence_transformers import SentenceTransformer, util
import logging
import sys
import signal

# 初始化Flask应用
app = Flask(__name__)

# 配置日志，日志级别设为INFO
logging.basicConfig(level=logging.INFO)
app.logger = logging.getLogger("CodeSearchAPI")

# 预定义代码片段
CODE_SNIPPETS = [
    "def sort_list(x): return sorted(x)",
    """def count_above_threshold(elements, threshold=0):
    return sum(1 for e in elements if e > threshold)""",
    """def find_min_max(elements):
    return min(elements), max(elements)"""
]

# 全局服务状态
service_ready = False

# 优雅关闭处理
def handle_shutdown(signum, frame):
    app.logger.info("收到终止信号，开始关闭...")
    sys.exit(0)

signal.signal(signal.SIGTERM, handle_shutdown)
signal.signal(signal.SIGINT, handle_shutdown)

# 初始化模型和编码
try:
    # Hugging Face Spaces专用缓存路径
    model = SentenceTransformer(
        "flax-sentence-embeddings/st-codesearch-distilroberta-base",
        cache_folder="/model-cache"
    )
    
    # 预计算编码（强制使用CPU）
    code_emb = model.encode(CODE_SNIPPETS, 
                            convert_to_tensor=True,
                            device="cpu")
    
    service_ready = True
    app.logger.info("服务初始化完成")
except Exception as e:
    app.logger.error(f"初始化失败: {str(e)}")
    raise

# Hugging Face健康检查端点
@app.route('/')
def hf_health_check():
    """必须响应根路径的健康检查"""
    if service_ready:
        return jsonify({"status": "ready"}), 200
    else:
        return jsonify({"status": "initializing"}), 503

# 支持GET和POST请求的搜索API端点
@app.route('/search', methods=['GET', 'POST'])
def handle_search():
    if not service_ready:
        return jsonify({"error": "服务正在初始化"}), 503

    try:
        # 区分GET和POST请求，GET从URL参数中获取query，POST从JSON体中获取
        if request.method == 'GET':
            query = request.args.get('query', '').strip()
        else:
            data = request.get_json() or {}
            query = data.get('query', '').strip()

        if not query:
            app.logger.info("收到空的查询请求")
            return jsonify({"error": "查询不能为空"}), 400
        
        # 记录接收到的查询
        app.logger.info("收到查询请求: %s", query)

        # 对查询进行编码，并搜索最匹配的代码片段
        query_emb = model.encode(query, 
                                 convert_to_tensor=True,
                                 device="cpu")
        hits = util.semantic_search(query_emb, code_emb, top_k=1)[0]
        best = hits[0]

        result = {
            "code": CODE_SNIPPETS[best['corpus_id']],
            "score": round(float(best['score']), 4)
        }
        # 记录返回结果
        app.logger.info("返回结果: %s", result)

        return jsonify(result)
        
    except Exception as e:
        app.logger.error("请求处理失败: %s", str(e))
        return jsonify({"error": "服务器内部错误"}), 500

if __name__ == "__main__":
    # Hugging Face Spaces会通过gunicorn启动，此处仅为本地测试保留
    app.run(host='0.0.0.0', port=7860)