feat: 性能插桩、缓存清除端点、LLM top_p 与 JSON 强制输出
- chat_routes.py: 新增 8 阶段性能计时(stages_ms)+ SSE 输出 - sync_routes.py: 新增 POST /cache/clear 缓存清除端点 - engine.py: LLM 调用增加 top_p 参数 - intent_analyzer.py: response_format 加配置开关(INTENT_RESPONSE_FORMAT)
This commit is contained in:
@@ -315,3 +315,48 @@ def stop_sync_monitor() -> Tuple[Any, int]:
|
||||
message="操作失败",
|
||||
http_status=500
|
||||
)
|
||||
|
||||
|
||||
# ==================== 缓存管理 API ====================
|
||||
|
||||
@sync_bp.route('/cache/clear', methods=['POST'])
|
||||
@require_gateway_auth
|
||||
def clear_all_caches() -> Tuple[Any, int]:
|
||||
"""
|
||||
清除所有缓存层(Query / Embedding / Rerank / Semantic / Intent 精确缓存)
|
||||
|
||||
用于测试和调试,不影响向量库数据。
|
||||
|
||||
Returns:
|
||||
{"success": true, "message": "缓存已清除", "cleared": [...]}
|
||||
"""
|
||||
cleared = []
|
||||
|
||||
# 1. 精确匹配缓存(Query / Embedding / Rerank)
|
||||
try:
|
||||
from core.cache import get_cache_manager
|
||||
cache = get_cache_manager()
|
||||
if cache:
|
||||
cache.clear_all()
|
||||
cleared.append("query_cache")
|
||||
cleared.append("embedding_cache")
|
||||
cleared.append("rerank_cache")
|
||||
except Exception as e:
|
||||
logger.warning(f"清除精确缓存失败: {e}")
|
||||
|
||||
# 2. 语义缓存(FAISS)
|
||||
try:
|
||||
from core.semantic_cache import get_semantic_cache
|
||||
sc = get_semantic_cache()
|
||||
if sc:
|
||||
sc.clear()
|
||||
cleared.append("semantic_cache")
|
||||
except Exception as e:
|
||||
logger.warning(f"清除语义缓存失败: {e}")
|
||||
|
||||
logger.info(f"缓存清除完成: {cleared}")
|
||||
return jsonify({
|
||||
"success": True,
|
||||
"message": f"已清除 {len(cleared)} 层缓存",
|
||||
"cleared": cleared
|
||||
})
|
||||
|
||||
Reference in New Issue
Block a user