feat(observability): Phase2.3 媒体/LLM 可观测面板(后端归因聚合+前端页) 并修复 token_usage_logs trace_id 写入(chat_service/im_llm) [v2026.07.19.5]
CI / gate (push) Has been cancelled
CI / gate (push) Has been cancelled
This commit is contained in:
QihuAI Design committed
1 parent
491f31cead
commit
2969544d54
9 files changed
+3357
-3079
No files matched your search
+234
-232
@@ -1,232 +1,234 @@
|
||||
"""
|
||||
IM Bot 真实 LLM 调用。
|
||||
|
||||
原实现位于 channels.py（通用频道管理模块），该模块已移除。
|
||||
为保持 IM Bot（微信机器人等）正常生成回复，将 _call_real_llm 迁移至此独立模块。
|
||||
|
||||
行为（对齐 AI 聊天）：
|
||||
- Issue 7：图像/视频/语音意图优先走平台媒体模型（与 AI 聊天同一套 Auto 路由），
|
||||
不再对"制作图片"这类请求只回文本。
|
||||
- Issue 8：默认 Auto 模型；每轮先自动发现本账号已安装且平台已部署的技能，
|
||||
多个适用时由模型自行判断选 1 个最优，通过 function calling 执行。
|
||||
- 2026-07-15：修复 trust_env=False + proxy 处理，避免容器坏代理导致调用失败；
|
||||
媒体生成前先经 _optimize_media_prompt 精炼提示词，提升 AGNES 生成质量。
|
||||
"""
|
||||
import asyncio
|
||||
import json
|
||||
import logging
|
||||
import re as _re
|
||||
from datetime import datetime, timezone, timedelta
|
||||
from typing import Optional
|
||||
from urllib.parse import urlparse as _urlparse
|
||||
|
||||
import httpx
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from ..capability_registry import resolve_auto_model
|
||||
from ..models import AIModel, ModelProvider, TokenUsageLog, gen_uuid
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
# 香港出网代理(实测对 AGNES 延迟更低，2026-07-15 用户实测)，与 chat.py 同值。
|
||||
_HK_PROXY = "http://qyyfw:XHR521yhf@103.242.14.208:11495"
|
||||
|
||||
|
||||
def _proxy_for_url(url: str) -> str | None:
|
||||
"""按目标域名选择是否走代理（与 chat.py 同逻辑）。
|
||||
|
||||
2026-07-15 调整：AGNES 改走香港代理(实测延迟更低)，不再直连。
|
||||
国内域名(尊重 NO_PROXY)直连，更稳定。
|
||||
"""
|
||||
import os
|
||||
host = (_urlparse(url).hostname or "").lower()
|
||||
if "agnes-ai.com" in host or "agnes-ai.space" in host:
|
||||
return _HK_PROXY
|
||||
if "qihuyun.com" in host:
|
||||
return None
|
||||
_no_proxy = [h.strip().lower() for h in os.environ.get("NO_PROXY", "").split(",") if h.strip()]
|
||||
for _np in _no_proxy:
|
||||
if host == _np or host.endswith("." + _np) or _np in host:
|
||||
return None
|
||||
return os.environ.get("HTTPS_PROXY") or os.environ.get("HTTP_PROXY") or "http://172.18.0.1:7890"
|
||||
|
||||
|
||||
def _run_async(coro, timeout=None):
|
||||
"""在同步上下文中运行协程，兼容「无事件循环」与「已有事件循环」两种场景。
|
||||
|
||||
IM 个人微信在独立线程中调用（无事件循环）→ asyncio.run；
|
||||
IM 多平台 Webhook 可能在 FastAPI 协程内调用（已有事件循环）→ 丢到线程池执行。
|
||||
|
||||
timeout: 可选，秒数；超时抛出 asyncio.TimeoutError（由调用方捕获）。
|
||||
"""
|
||||
try:
|
||||
loop = asyncio.get_running_loop()
|
||||
except RuntimeError:
|
||||
loop = None
|
||||
|
||||
if timeout is not None:
|
||||
async def _timed():
|
||||
return await asyncio.wait_for(coro, timeout=timeout)
|
||||
target = _timed()
|
||||
else:
|
||||
target = coro
|
||||
|
||||
if loop is not None and loop.is_running():
|
||||
import concurrent.futures
|
||||
with concurrent.futures.ThreadPoolExecutor(1) as ex:
|
||||
return ex.submit(lambda: asyncio.run(target)).result(timeout=(timeout + 10) if timeout else None)
|
||||
return asyncio.run(target)
|
||||
|
||||
|
||||
def _log_media_usage(db: Session, user_id: str, ai_model, media_type: str, prompt: str):
|
||||
"""IM Bot 媒体生成记账：把图片/视频消耗写入 TokenUsageLog，供驾驶舱统计。
|
||||
|
||||
此前 IM Bot 路径完全不写 TokenUsageLog，导致微信生成的图片/视频在驾驶舱里“查无此项”。
|
||||
媒体 API 不返回 token 用量，这里用与 chat.py _generate_media_task 一致的估算。
|
||||
"""
|
||||
if not user_id or not ai_model:
|
||||
return
|
||||
try:
|
||||
base = 1200 if media_type == "video" else 400
|
||||
est = base + (len(prompt or "") // 4)
|
||||
db.add(TokenUsageLog(
|
||||
id=gen_uuid(),
|
||||
user_id=user_id,
|
||||
model_id=ai_model.id,
|
||||
model_name=ai_model.name,
|
||||
prompt_tokens=est,
|
||||
completion_tokens=0,
|
||||
total_tokens=est,
|
||||
created_at=datetime.now(timezone(timedelta(hours=8))).replace(tzinfo=None),
|
||||
))
|
||||
db.commit()
|
||||
except Exception as e:
|
||||
try:
|
||||
db.rollback()
|
||||
except Exception:
|
||||
pass
|
||||
logger.warning(f"[im_llm] 媒体 Token 记账失败(忽略): {e}")
|
||||
|
||||
|
||||
def _call_real_llm(db: Session, user=None, prompt: str = "", platform: str = "im", user_id: str = None) -> "str | dict":
|
||||
"""调用七虎AI 生成回复，供 IM Bot 复用。支持媒体意图路由与已安装技能自动发现。"""
|
||||
try:
|
||||
from app.api.chat import (
|
||||
_generate_media,
|
||||
detect_media_intent,
|
||||
_discover_installed_skills,
|
||||
_build_system_prompt,
|
||||
_execute_skill_tool,
|
||||
)
|
||||
|
||||
# ===== 1) 媒体意图路由（Issue 7）=====
|
||||
intent = detect_media_intent(prompt)
|
||||
if intent:
|
||||
ai_model = resolve_auto_model(intent, db)
|
||||
if ai_model:
|
||||
try:
|
||||
result = _run_async(_generate_media(db, ai_model, prompt, user_id), timeout=300)
|
||||
except Exception:
|
||||
# 任何未预期异常(如 AGNES 连接重置)统一兜底为友好提示，
|
||||
# 绝不直接把原始异常(如 "Server disconnected")甩给用户
|
||||
result = "⚠️ 媒体生成暂时失败（连接或服务波动），请稍后重试。"
|
||||
if result:
|
||||
# 媒体生成成功：返回结构化对象，交由 im_bot_user 在微信内联发送图片/视频
|
||||
s = result.strip()
|
||||
if s.startswith("<img") or s.startswith("<video"):
|
||||
m = _re.search(r'src="([^"]+)"', s)
|
||||
url = m.group(1) if m else ""
|
||||
media_type = "image" if s.startswith("<img") else "video"
|
||||
if not url:
|
||||
return "（生成完成，但未能提取媒体地址）"
|
||||
# IM Bot 路径此前完全不记账 → 补写媒体消耗，驾驶舱才看得到图片/视频数据
|
||||
_log_media_usage(db, user_id, ai_model, media_type, prompt)
|
||||
return {"__media__": True, "media_type": media_type, "url": url}
|
||||
# 媒体生成失败：_generate_media 已返回友好中文（如限频/队列繁忙），
|
||||
# 直接透传，绝不把原始提示词/标签/JSON 透传给用户
|
||||
if s.startswith("⚠️") or s.startswith("（媒体"):
|
||||
return s
|
||||
return "（媒体生成未能返回有效内容，请重试）"
|
||||
|
||||
# ===== 2) 文本回复 + 技能自动发现（Issue 8）=====
|
||||
ai_model = resolve_auto_model("text", db)
|
||||
if not ai_model:
|
||||
return "（未配置默认文本模型，无法生成回复）"
|
||||
provider = db.query(ModelProvider).filter(ModelProvider.id == ai_model.provider_id).first()
|
||||
api_key = ai_model.api_key or (provider.api_key if provider else None)
|
||||
api_url = ai_model.endpoint
|
||||
if not api_url and provider and provider.api_base_url:
|
||||
api_url = f"{provider.api_base_url.rstrip('/')}/v1/chat/completions"
|
||||
if not (api_key and api_url):
|
||||
return "（模型未配置 API Key 或地址，无法生成回复）"
|
||||
|
||||
# 自动发现本账号已安装且平台已部署的技能
|
||||
skills = []
|
||||
if user_id:
|
||||
try:
|
||||
skills = _discover_installed_skills(db, user_id)[:15]
|
||||
except Exception:
|
||||
skills = []
|
||||
|
||||
system_prompt, function_definitions = _build_system_prompt(
|
||||
skills=skills,
|
||||
agents=[],
|
||||
kb_context="",
|
||||
user_content=prompt,
|
||||
mcp=[],
|
||||
auto_discovered=bool(skills),
|
||||
)
|
||||
messages = [
|
||||
{"role": "system", "content": system_prompt},
|
||||
{"role": "user", "content": prompt},
|
||||
]
|
||||
|
||||
max_rounds = 3
|
||||
ai_content = ""
|
||||
for _ in range(max_rounds):
|
||||
request_body = {"model": ai_model.name, "messages": messages, "max_tokens": 2048}
|
||||
if function_definitions and skills:
|
||||
request_body["tools"] = [
|
||||
{"type": "function", "function": fd} for fd in function_definitions
|
||||
]
|
||||
request_body["tool_choice"] = "auto"
|
||||
with httpx.Client(
|
||||
timeout=ai_model.timeout or 120,
|
||||
proxy=_proxy_for_url(api_url),
|
||||
trust_env=False,
|
||||
) as client:
|
||||
resp = client.post(
|
||||
api_url,
|
||||
headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
|
||||
json=request_body,
|
||||
)
|
||||
if resp.status_code != 200:
|
||||
return f"（AI 模型调用失败 HTTP {resp.status_code}）"
|
||||
data = resp.json()
|
||||
message = data.get("choices", [{}])[0].get("message", {})
|
||||
tool_calls = message.get("tool_calls", [])
|
||||
if tool_calls and skills:
|
||||
messages.append({
|
||||
"role": "assistant",
|
||||
"content": message.get("content") or "",
|
||||
"tool_calls": tool_calls,
|
||||
})
|
||||
for tc in tool_calls:
|
||||
func_name = tc.get("function", {}).get("name", "")
|
||||
try:
|
||||
func_args = json.loads(tc.get("function", {}).get("arguments", "{}"))
|
||||
except Exception:
|
||||
func_args = {"input": tc.get("function", {}).get("arguments", "")}
|
||||
tool_result = _execute_skill_tool(db, func_name, func_args, skills, prompt)
|
||||
messages.append({
|
||||
"role": "tool",
|
||||
"tool_call_id": tc.get("id", "call_1"),
|
||||
"content": tool_result,
|
||||
})
|
||||
continue
|
||||
ai_content = message.get("content", "")
|
||||
break
|
||||
return ai_content or "（AI 回复为空）"
|
||||
except Exception as e:
|
||||
return f"（AI 回复生成失败：{e}）"
|
||||
"""
|
||||
IM Bot 真实 LLM 调用。
|
||||
|
||||
原实现位于 channels.py（通用频道管理模块），该模块已移除。
|
||||
为保持 IM Bot（微信机器人等）正常生成回复，将 _call_real_llm 迁移至此独立模块。
|
||||
|
||||
行为（对齐 AI 聊天）：
|
||||
- Issue 7：图像/视频/语音意图优先走平台媒体模型（与 AI 聊天同一套 Auto 路由），
|
||||
不再对"制作图片"这类请求只回文本。
|
||||
- Issue 8：默认 Auto 模型；每轮先自动发现本账号已安装且平台已部署的技能，
|
||||
多个适用时由模型自行判断选 1 个最优，通过 function calling 执行。
|
||||
- 2026-07-15：修复 trust_env=False + proxy 处理，避免容器坏代理导致调用失败；
|
||||
媒体生成前先经 _optimize_media_prompt 精炼提示词，提升 AGNES 生成质量。
|
||||
"""
|
||||
import asyncio
|
||||
import json
|
||||
import logging
|
||||
import re as _re
|
||||
from datetime import datetime, timezone, timedelta
|
||||
from typing import Optional
|
||||
from urllib.parse import urlparse as _urlparse
|
||||
|
||||
import httpx
|
||||
from sqlalchemy.orm import Session
|
||||
|
||||
from ..capability_registry import resolve_auto_model
|
||||
from ..models import AIModel, ModelProvider, TokenUsageLog, gen_uuid
|
||||
from app.core.observability import get_trace_id
|
||||
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
# 香港出网代理(实测对 AGNES 延迟更低，2026-07-15 用户实测)，与 chat.py 同值。
|
||||
_HK_PROXY = "http://qyyfw:XHR521yhf@103.242.14.208:11495"
|
||||
|
||||
|
||||
def _proxy_for_url(url: str) -> str | None:
|
||||
"""按目标域名选择是否走代理（与 chat.py 同逻辑）。
|
||||
|
||||
2026-07-15 调整：AGNES 改走香港代理(实测延迟更低)，不再直连。
|
||||
国内域名(尊重 NO_PROXY)直连，更稳定。
|
||||
"""
|
||||
import os
|
||||
host = (_urlparse(url).hostname or "").lower()
|
||||
if "agnes-ai.com" in host or "agnes-ai.space" in host:
|
||||
return _HK_PROXY
|
||||
if "qihuyun.com" in host:
|
||||
return None
|
||||
_no_proxy = [h.strip().lower() for h in os.environ.get("NO_PROXY", "").split(",") if h.strip()]
|
||||
for _np in _no_proxy:
|
||||
if host == _np or host.endswith("." + _np) or _np in host:
|
||||
return None
|
||||
return os.environ.get("HTTPS_PROXY") or os.environ.get("HTTP_PROXY") or "http://172.18.0.1:7890"
|
||||
|
||||
|
||||
def _run_async(coro, timeout=None):
|
||||
"""在同步上下文中运行协程，兼容「无事件循环」与「已有事件循环」两种场景。
|
||||
|
||||
IM 个人微信在独立线程中调用（无事件循环）→ asyncio.run；
|
||||
IM 多平台 Webhook 可能在 FastAPI 协程内调用（已有事件循环）→ 丢到线程池执行。
|
||||
|
||||
timeout: 可选，秒数；超时抛出 asyncio.TimeoutError（由调用方捕获）。
|
||||
"""
|
||||
try:
|
||||
loop = asyncio.get_running_loop()
|
||||
except RuntimeError:
|
||||
loop = None
|
||||
|
||||
if timeout is not None:
|
||||
async def _timed():
|
||||
return await asyncio.wait_for(coro, timeout=timeout)
|
||||
target = _timed()
|
||||
else:
|
||||
target = coro
|
||||
|
||||
if loop is not None and loop.is_running():
|
||||
import concurrent.futures
|
||||
with concurrent.futures.ThreadPoolExecutor(1) as ex:
|
||||
return ex.submit(lambda: asyncio.run(target)).result(timeout=(timeout + 10) if timeout else None)
|
||||
return asyncio.run(target)
|
||||
|
||||
|
||||
def _log_media_usage(db: Session, user_id: str, ai_model, media_type: str, prompt: str):
|
||||
"""IM Bot 媒体生成记账：把图片/视频消耗写入 TokenUsageLog，供驾驶舱统计。
|
||||
|
||||
此前 IM Bot 路径完全不写 TokenUsageLog，导致微信生成的图片/视频在驾驶舱里“查无此项”。
|
||||
媒体 API 不返回 token 用量，这里用与 chat.py _generate_media_task 一致的估算。
|
||||
"""
|
||||
if not user_id or not ai_model:
|
||||
return
|
||||
try:
|
||||
base = 1200 if media_type == "video" else 400
|
||||
est = base + (len(prompt or "") // 4)
|
||||
db.add(TokenUsageLog(
|
||||
id=gen_uuid(),
|
||||
user_id=user_id,
|
||||
model_id=ai_model.id,
|
||||
model_name=ai_model.name,
|
||||
trace_id=get_trace_id(),
|
||||
prompt_tokens=est,
|
||||
completion_tokens=0,
|
||||
total_tokens=est,
|
||||
created_at=datetime.now(timezone(timedelta(hours=8))).replace(tzinfo=None),
|
||||
))
|
||||
db.commit()
|
||||
except Exception as e:
|
||||
try:
|
||||
db.rollback()
|
||||
except Exception:
|
||||
pass
|
||||
logger.warning(f"[im_llm] 媒体 Token 记账失败(忽略): {e}")
|
||||
|
||||
|
||||
def _call_real_llm(db: Session, user=None, prompt: str = "", platform: str = "im", user_id: str = None) -> "str | dict":
|
||||
"""调用七虎AI 生成回复，供 IM Bot 复用。支持媒体意图路由与已安装技能自动发现。"""
|
||||
try:
|
||||
from app.api.chat import (
|
||||
_generate_media,
|
||||
detect_media_intent,
|
||||
_discover_installed_skills,
|
||||
_build_system_prompt,
|
||||
_execute_skill_tool,
|
||||
)
|
||||
|
||||
# ===== 1) 媒体意图路由（Issue 7）=====
|
||||
intent = detect_media_intent(prompt)
|
||||
if intent:
|
||||
ai_model = resolve_auto_model(intent, db)
|
||||
if ai_model:
|
||||
try:
|
||||
result = _run_async(_generate_media(db, ai_model, prompt, user_id), timeout=300)
|
||||
except Exception:
|
||||
# 任何未预期异常(如 AGNES 连接重置)统一兜底为友好提示，
|
||||
# 绝不直接把原始异常(如 "Server disconnected")甩给用户
|
||||
result = "⚠️ 媒体生成暂时失败（连接或服务波动），请稍后重试。"
|
||||
if result:
|
||||
# 媒体生成成功：返回结构化对象，交由 im_bot_user 在微信内联发送图片/视频
|
||||
s = result.strip()
|
||||
if s.startswith("<img") or s.startswith("<video"):
|
||||
m = _re.search(r'src="([^"]+)"', s)
|
||||
url = m.group(1) if m else ""
|
||||
media_type = "image" if s.startswith("<img") else "video"
|
||||
if not url:
|
||||
return "（生成完成，但未能提取媒体地址）"
|
||||
# IM Bot 路径此前完全不记账 → 补写媒体消耗，驾驶舱才看得到图片/视频数据
|
||||
_log_media_usage(db, user_id, ai_model, media_type, prompt)
|
||||
return {"__media__": True, "media_type": media_type, "url": url}
|
||||
# 媒体生成失败：_generate_media 已返回友好中文（如限频/队列繁忙），
|
||||
# 直接透传，绝不把原始提示词/标签/JSON 透传给用户
|
||||
if s.startswith("⚠️") or s.startswith("（媒体"):
|
||||
return s
|
||||
return "（媒体生成未能返回有效内容，请重试）"
|
||||
|
||||
# ===== 2) 文本回复 + 技能自动发现（Issue 8）=====
|
||||
ai_model = resolve_auto_model("text", db)
|
||||
if not ai_model:
|
||||
return "（未配置默认文本模型，无法生成回复）"
|
||||
provider = db.query(ModelProvider).filter(ModelProvider.id == ai_model.provider_id).first()
|
||||
api_key = ai_model.api_key or (provider.api_key if provider else None)
|
||||
api_url = ai_model.endpoint
|
||||
if not api_url and provider and provider.api_base_url:
|
||||
api_url = f"{provider.api_base_url.rstrip('/')}/v1/chat/completions"
|
||||
if not (api_key and api_url):
|
||||
return "（模型未配置 API Key 或地址，无法生成回复）"
|
||||
|
||||
# 自动发现本账号已安装且平台已部署的技能
|
||||
skills = []
|
||||
if user_id:
|
||||
try:
|
||||
skills = _discover_installed_skills(db, user_id)[:15]
|
||||
except Exception:
|
||||
skills = []
|
||||
|
||||
system_prompt, function_definitions = _build_system_prompt(
|
||||
skills=skills,
|
||||
agents=[],
|
||||
kb_context="",
|
||||
user_content=prompt,
|
||||
mcp=[],
|
||||
auto_discovered=bool(skills),
|
||||
)
|
||||
messages = [
|
||||
{"role": "system", "content": system_prompt},
|
||||
{"role": "user", "content": prompt},
|
||||
]
|
||||
|
||||
max_rounds = 3
|
||||
ai_content = ""
|
||||
for _ in range(max_rounds):
|
||||
request_body = {"model": ai_model.name, "messages": messages, "max_tokens": 2048}
|
||||
if function_definitions and skills:
|
||||
request_body["tools"] = [
|
||||
{"type": "function", "function": fd} for fd in function_definitions
|
||||
]
|
||||
request_body["tool_choice"] = "auto"
|
||||
with httpx.Client(
|
||||
timeout=ai_model.timeout or 120,
|
||||
proxy=_proxy_for_url(api_url),
|
||||
trust_env=False,
|
||||
) as client:
|
||||
resp = client.post(
|
||||
api_url,
|
||||
headers={"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"},
|
||||
json=request_body,
|
||||
)
|
||||
if resp.status_code != 200:
|
||||
return f"（AI 模型调用失败 HTTP {resp.status_code}）"
|
||||
data = resp.json()
|
||||
message = data.get("choices", [{}])[0].get("message", {})
|
||||
tool_calls = message.get("tool_calls", [])
|
||||
if tool_calls and skills:
|
||||
messages.append({
|
||||
"role": "assistant",
|
||||
"content": message.get("content") or "",
|
||||
"tool_calls": tool_calls,
|
||||
})
|
||||
for tc in tool_calls:
|
||||
func_name = tc.get("function", {}).get("name", "")
|
||||
try:
|
||||
func_args = json.loads(tc.get("function", {}).get("arguments", "{}"))
|
||||
except Exception:
|
||||
func_args = {"input": tc.get("function", {}).get("arguments", "")}
|
||||
tool_result = _execute_skill_tool(db, func_name, func_args, skills, prompt)
|
||||
messages.append({
|
||||
"role": "tool",
|
||||
"tool_call_id": tc.get("id", "call_1"),
|
||||
"content": tool_result,
|
||||
})
|
||||
continue
|
||||
ai_content = message.get("content", "")
|
||||
break
|
||||
return ai_content or "（AI 回复为空）"
|
||||
except Exception as e:
|
||||
return f"（AI 回复生成失败：{e}）"
|
||||
Reference in new issue
Block a user