feat: 添加结构化日志系统，更新LLM配置与全部文档
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:
+138
-11
@@ -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")
|
||||
@@ -0,0 +1,167 @@
|
||||
"""集中日志模块。
|
||||
|
||||
提供：
|
||||
- 结构化 JSON 日志（每行一条记录）
|
||||
- 请求级 trace_id（通过 contextvars 自动传播）
|
||||
- 独立的 LLM 调用日志文件
|
||||
- 日志轮转（按大小 10MB，保留 5 个备份）
|
||||
|
||||
用法:
|
||||
from backend.logger import get_logger, set_trace_id
|
||||
|
||||
# 业务日志
|
||||
log = get_logger("agent")
|
||||
log.info("节点开始执行", extra={"node": "classify_intent", "session_id": "xxx"})
|
||||
|
||||
# LLM 日志
|
||||
llm_log = get_logger("llm")
|
||||
llm_log.info("LLM 请求", extra={"prompt": "...", "model": "gpt-4o"})
|
||||
"""
|
||||
|
||||
import json
|
||||
import logging
|
||||
import os
|
||||
import sys
|
||||
import uuid
|
||||
from contextvars import ContextVar
|
||||
from datetime import datetime, timezone, timedelta
|
||||
from logging.handlers import RotatingFileHandler
|
||||
from pathlib import Path
|
||||
from typing import Optional
|
||||
|
||||
LOG_DIR = Path(os.getenv("LOG_DIR", "./logs"))
|
||||
LOG_LEVEL = os.getenv("LOG_LEVEL", "DEBUG")
|
||||
LLM_LOG_FILE = "llm.log"
|
||||
APP_LOG_FILE = "app.log"
|
||||
|
||||
CHINA_TZ = timezone(timedelta(hours=8))
|
||||
|
||||
_trace_id_var: ContextVar[str] = ContextVar("trace_id", default="")
|
||||
|
||||
|
||||
def generate_trace_id() -> str:
|
||||
return uuid.uuid4().hex[:16]
|
||||
|
||||
|
||||
def get_trace_id() -> str:
|
||||
tid = _trace_id_var.get()
|
||||
if not tid:
|
||||
tid = generate_trace_id()
|
||||
_trace_id_var.set(tid)
|
||||
return tid
|
||||
|
||||
|
||||
def set_trace_id(trace_id: str):
|
||||
_trace_id_var.set(trace_id)
|
||||
|
||||
|
||||
class JsonFormatter(logging.Formatter):
|
||||
"""将日志记录格式化为单行 JSON，便于后续分析。
|
||||
|
||||
LogRecord 标准属性的键（不放入 extra）。
|
||||
通过 logging.Logger.debug(msg, extra={...}) 传入的键会自动设为
|
||||
LogRecord 属性，由本格式化器收集到 extra 字段中。
|
||||
"""
|
||||
|
||||
_STANDARD_ATTRS: set[str] = frozenset({
|
||||
"args", "asctime", "created", "exc_info", "exc_text", "filename",
|
||||
"funcName", "levelname", "levelno", "lineno", "module", "msecs",
|
||||
"message", "msg", "name", "pathname", "process", "processName",
|
||||
"relativeCreated", "stack_info", "thread", "threadName",
|
||||
"extra_fields", "taskName",
|
||||
})
|
||||
|
||||
def _collect_extra(self, record: logging.LogRecord) -> dict:
|
||||
"""从 LogRecord 上收集非标准属性 → 合并为 extra dict。"""
|
||||
extra = dict(getattr(record, "extra_fields", {}))
|
||||
for key, val in record.__dict__.items():
|
||||
if key not in self._STANDARD_ATTRS and not key.startswith("_"):
|
||||
extra[key] = val
|
||||
return extra
|
||||
|
||||
def format(self, record: logging.LogRecord) -> str:
|
||||
log_entry = {
|
||||
"timestamp": datetime.now(CHINA_TZ).isoformat(),
|
||||
"level": record.levelname,
|
||||
"logger": record.name,
|
||||
"trace_id": get_trace_id(),
|
||||
"message": record.getMessage(),
|
||||
"module": record.module,
|
||||
"function": record.funcName,
|
||||
"line": record.lineno,
|
||||
}
|
||||
|
||||
extra = self._collect_extra(record)
|
||||
if extra:
|
||||
log_entry["extra"] = extra
|
||||
|
||||
if record.exc_info and record.exc_info[0]:
|
||||
import traceback
|
||||
log_entry["exception"] = traceback.format_exception(
|
||||
record.exc_info[0], record.exc_info[1], record.exc_info[2]
|
||||
)
|
||||
|
||||
return json.dumps(log_entry, ensure_ascii=False)
|
||||
|
||||
|
||||
def _create_handler(filename: str, level: int) -> RotatingFileHandler:
|
||||
handler = RotatingFileHandler(
|
||||
filename=str(LOG_DIR / filename),
|
||||
maxBytes=10 * 1024 * 1024,
|
||||
backupCount=5,
|
||||
encoding="utf-8",
|
||||
)
|
||||
handler.setLevel(level)
|
||||
handler.setFormatter(JsonFormatter())
|
||||
return handler
|
||||
|
||||
|
||||
def _get_level() -> int:
|
||||
return getattr(logging, LOG_LEVEL.upper(), logging.DEBUG)
|
||||
|
||||
|
||||
def get_logger(name: str) -> logging.Logger:
|
||||
"""获取指定名称的 logger，自动配置了 JSON 格式化 + 文件轮转。
|
||||
|
||||
name="llm" → 输出到 logs/llm.log（仅 LLM 调用相关）
|
||||
其他 name → 输出到 logs/app.log
|
||||
"""
|
||||
logger = logging.getLogger(f"jrxml.{name}")
|
||||
|
||||
if logger.handlers:
|
||||
return logger
|
||||
|
||||
LOG_DIR.mkdir(parents=True, exist_ok=True)
|
||||
|
||||
level = _get_level()
|
||||
logger.setLevel(level)
|
||||
logger.propagate = False
|
||||
|
||||
if name == "llm":
|
||||
logger.addHandler(_create_handler(LLM_LOG_FILE, level))
|
||||
else:
|
||||
logger.addHandler(_create_handler(APP_LOG_FILE, level))
|
||||
|
||||
return logger
|
||||
|
||||
|
||||
class _ExtraAdapter(logging.LoggerAdapter):
|
||||
"""支持通过 adapter.extra 合并 extra 字段的适配器。"""
|
||||
|
||||
def process(self, msg, kwargs):
|
||||
extra = kwargs.pop("extra", {})
|
||||
merged = {**self.extra, **extra} if self.extra or extra else None
|
||||
if merged:
|
||||
kwargs["extra"] = {"extra_fields": merged}
|
||||
return msg, kwargs
|
||||
|
||||
|
||||
def get_trace_logger(name: str) -> _ExtraAdapter:
|
||||
"""返回一个自动附带 trace_id 的 logger 适配器。
|
||||
|
||||
用法:
|
||||
log = get_trace_logger("agent")
|
||||
log.info("节点完成", extra={"node": "generate"})
|
||||
"""
|
||||
logger = get_logger(name)
|
||||
return _ExtraAdapter(logger, {"trace_id": get_trace_id()})
|
||||
@@ -12,8 +12,12 @@ from typing import Optional
|
||||
|
||||
from dotenv import load_dotenv
|
||||
|
||||
from backend.logger import get_logger
|
||||
|
||||
load_dotenv()
|
||||
|
||||
_session_log = get_logger("session")
|
||||
|
||||
SESSIONS_DIR = Path(os.getenv("SESSIONS_DIR", "./sessions"))
|
||||
|
||||
|
||||
@@ -43,6 +47,7 @@ def create_session(name: str = "", agent_state: Optional[dict] = None) -> dict:
|
||||
}
|
||||
with open(_session_path(sid), "w", encoding="utf-8") as f:
|
||||
json.dump(data, f, ensure_ascii=False, indent=2)
|
||||
_session_log.info("创建会话", extra={"session_id": sid, "session_name": data["session_name"]})
|
||||
return data
|
||||
|
||||
|
||||
@@ -104,6 +109,7 @@ def delete_session(session_id: str) -> bool:
|
||||
fp = _session_path(session_id)
|
||||
if fp.exists():
|
||||
fp.unlink()
|
||||
_session_log.info("删除会话", extra={"session_id": session_id})
|
||||
return True
|
||||
return False
|
||||
|
||||
|
||||
+21
-6
@@ -5,22 +5,37 @@ import os
|
||||
import httpx
|
||||
from dotenv import load_dotenv
|
||||
|
||||
from backend.logger import get_logger
|
||||
|
||||
load_dotenv()
|
||||
|
||||
_val_log = get_logger("validation")
|
||||
|
||||
VALIDATION_URL = os.getenv("VALIDATION_SERVICE_URL", "http://localhost:8001/validate")
|
||||
|
||||
|
||||
def validate_jrxml(jrxml_text: str) -> dict:
|
||||
"""将 JRXML 发送到验证服务并返回 {valid: bool, error: str}。"""
|
||||
jrxml_length = len(jrxml_text)
|
||||
try:
|
||||
with httpx.Client(timeout=30.0) as client:
|
||||
resp = client.post(VALIDATION_URL, json={"jrxml": jrxml_text})
|
||||
resp.raise_for_status()
|
||||
return resp.json()
|
||||
result = resp.json()
|
||||
_val_log.info(
|
||||
"验证完成",
|
||||
extra={
|
||||
"valid": result.get("valid"),
|
||||
"error": result.get("error", ""),
|
||||
"jrxml_length": jrxml_length,
|
||||
},
|
||||
)
|
||||
return result
|
||||
except httpx.ConnectError:
|
||||
return {
|
||||
"valid": False,
|
||||
"error": f"无法连接到验证服务 ({VALIDATION_URL})。是否正在运行？",
|
||||
}
|
||||
error_msg = f"无法连接到验证服务 ({VALIDATION_URL})。是否正在运行？"
|
||||
_val_log.error("验证服务连接失败", extra={"error": error_msg, "url": VALIDATION_URL})
|
||||
return {"valid": False, "error": error_msg}
|
||||
except Exception as e:
|
||||
return {"valid": False, "error": f"验证请求失败: {str(e)}"}
|
||||
error_msg = f"验证请求失败: {str(e)}"
|
||||
_val_log.error("验证请求异常", extra={"error": str(e), "url": VALIDATION_URL})
|
||||
return {"valid": False, "error": error_msg}
|
||||
|
||||
Reference in New Issue
Block a user