panda/agent_jrxml
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: 添加结构化日志系统，更新LLM配置与全部文档

Browse Source 
新增:
- backend/logger.py — 集中日志模块 (JSON格式 + trace_id + 独立llm.log)
- @log_node / @_log_route 装饰器覆盖17个节点和8个路由

改进:
- backend/llm.py — _LLMLoggingWrapper 自动记录LLM输入输出
- backend/llm.py — API Key优先读ANTHROPIC_API_KEY，模型名改为MiniMax-M2.7
- backend/llm.py — get_llm() 新增caller参数标识调用来源
- backend/validation.py — 新增验证结果/连接失败日志
- backend/session.py — 新增会话创建/删除日志
- app.py — 新增用户交互日志 (输入/执行/异常/会话操作)
- app.py — 提前导入torchvision抑制transformers懒加载报错
- .env.example — 新增LOG_DIR/LOG_LEVEL/ANTHROPIC_API_KEY等配置项
- .gitignore — 新增logs/和db/忽略规则

文档:
- ROADMAP.md — 新增阶段四: 可观测性
- README.md — 补充日志架构/LLM配置/项目结构
- CLAUDE.md — 同步最新配置/日志/MAX_RETRY(3)
- CODE_GUIDE.md — 新增第15章日志系统，更新架构图/LLM/配置
This commit is contained in:
panda
2026-05-19 23:40:01 +08:00
parent 6467fd4ae5
commit 067880bf2e
13 changed files with 753 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files
backend/llm.py
+138 -11
View File
@@ -1,12 +1,17 @@
"""大语言模型工厂：支持 OpenAI 兼容的云端 API、Anthropic 兼容 API 和本地 Ollama。"""
import os
import time
from typing import Any
from dotenv import load_dotenv
from backend.logger import get_logger
load_dotenv()
_llm_log = get_logger("llm")
class _BaseLLM:
"""LLM 统一接口基类 — 所有后端都提供 invoke() 和 stream()。"""
@@ -18,7 +23,124 @@ class _BaseLLM:
raise NotImplementedError
def get_llm():
class _LLMLoggingWrapper(_BaseLLM):
"""包装任何 LLM 后端，自动记录输入/输出到 llm.log。"""
def __init__(self, inner: _BaseLLM, model: str, backend: str, caller: str = ""):
self._inner = inner
self._model = model
self._backend = backend
self._caller = caller
def invoke(self, prompt: str) -> Any:
t0 = time.time()
prompt_len = len(prompt)
prompt_preview = prompt[:500]
_llm_log.debug(
"LLM invoke 请求",
extra={
"direction": "request",
"model": self._model,
"backend": self._backend,
"caller": self._caller,
"prompt_length": prompt_len,
"prompt_preview": prompt_preview,
"prompt": prompt[:10000],
},
)
try:
result = self._inner.invoke(prompt)
elapsed = round((time.time() - t0) * 1000)
content = getattr(result, "content", str(result))
resp_len = len(content)
resp_preview = content[:500]
_llm_log.info(
"LLM invoke 完成",
extra={
"direction": "response",
"model": self._model,
"backend": self._backend,
"caller": self._caller,
"duration_ms": elapsed,
"response_length": resp_len,
"response_preview": resp_preview,
"response": content[:10000],
},
)
return result
except Exception as e:
elapsed = round((time.time() - t0) * 1000)
_llm_log.error(
"LLM invoke 异常",
extra={
"direction": "error",
"model": self._model,
"backend": self._backend,
"caller": self._caller,
"duration_ms": elapsed,
"error": str(e),
"prompt": prompt[:10000],
},
)
raise
def stream(self, prompt: str):
t0 = time.time()
prompt_len = len(prompt)
prompt_preview = prompt[:500]
_llm_log.debug(
"LLM stream 请求",
extra={
"direction": "request",
"model": self._model,
"backend": self._backend,
"caller": self._caller,
"prompt_length": prompt_len,
"prompt_preview": prompt_preview,
"prompt": prompt[:10000],
},
)
full = []
try:
for chunk in self._inner.stream(prompt):
full.append(chunk)
yield chunk
elapsed = round((time.time() - t0) * 1000)
resp_text = "".join(full)
resp_len = len(resp_text)
resp_preview = resp_text[:500]
_llm_log.info(
"LLM stream 完成",
extra={
"direction": "response",
"model": self._model,
"backend": self._backend,
"caller": self._caller,
"duration_ms": elapsed,
"response_length": resp_len,
"response_preview": resp_preview,
"response": resp_text[:10000],
},
)
except Exception as e:
elapsed = round((time.time() - t0) * 1000)
_llm_log.error(
"LLM stream 异常",
extra={
"direction": "error",
"model": self._model,
"backend": self._backend,
"caller": self._caller,
"duration_ms": elapsed,
"error": str(e),
"prompt": prompt[:10000],
},
)
raise
def _build_raw_llm(caller: str = "") -> tuple[_BaseLLM, str, str]:
"""构造原始 LLM 实例，返回 (实例, model名, backend名)。"""
backend = os.getenv("LLM_BACKEND", "cloud")
if backend == "local":
from langchain_ollama import ChatOllama
@@ -34,20 +156,18 @@ def get_llm():
for chunk in raw.stream(prompt):
yield chunk.content
return OllamaWrapper()
return OllamaWrapper(), model, f"local/{model}"
provider = os.getenv("LLM_PROVIDER", "openai")
if provider == "anthropic":
from anthropic import Anthropic
api_key = os.getenv("OPENAI_API_KEY", "")
base_url = os.getenv("OPENAI_BASE_URL", "https://api.minimaxi.com/anthropic")
model = os.getenv("LLM_MODEL", "minimax-2.7")
api_key = os.getenv("ANTHROPIC_API_KEY") or os.getenv("OPENAI_API_KEY", "")
base_url = os.getenv("ANTHROPIC_BASE_URL") or os.getenv("OPENAI_BASE_URL", "https://api.minimaxi.com/anthropic")
model = os.getenv("LLM_MODEL", "MiniMax-M2.7")
temperature = 0.1
max_tokens = 4096
os.environ["NO_PROXY"] = "*"
client = Anthropic(api_key=api_key, base_url=base_url, timeout=120)
class MiniMaxLLM(_BaseLLM):
@@ -80,12 +200,13 @@ def get_llm():
)
return resp.input_tokens
return MiniMaxLLM()
return MiniMaxLLM(), model, f"cloud/anthropic/{model}"
else:
from langchain_openai import ChatOpenAI
model = os.getenv("LLM_MODEL", "gpt-4o")
raw = ChatOpenAI(
model=os.getenv("LLM_MODEL", "gpt-4o"),
model=model,
api_key=os.getenv("OPENAI_API_KEY"),
base_url=os.getenv("OPENAI_BASE_URL", "https://api.openai.com/v1"),
temperature=0.1,
@@ -99,8 +220,14 @@ def get_llm():
for chunk in raw.stream(prompt):
yield chunk.content
return OpenAIWrapper()
return OpenAIWrapper(), model, f"cloud/openai/{model}"
def get_llm(caller: str = "") -> _BaseLLM:
"""返回带日志的 LLM 实例。caller 用于标识调用来源（如 generate、classify_intent）。"""
inner, model, backend = _build_raw_llm(caller)
return _LLMLoggingWrapper(inner, model=model, backend=backend, caller=caller)
def get_llm_for_correction():
return get_llm()
return get_llm(caller="correction")