From ccc5f799e1c8fa4a2f0b2e7dd2436f44c7c84ceb Mon Sep 17 00:00:00 2001
From: deploy <deploy@qihuyun.com>
Date: Wed, 15 Jul 2026 19:58:06 +0800
Subject: [PATCH] =?UTF-8?q?fix(media):=20AGNES=20=E5=9B=BE=E5=83=8F/?=
 =?UTF-8?q?=E8=A7=86=E9=A2=91=E7=94=9F=E6=88=90=E5=8A=A0=E9=9F=A7=E6=80=A7?=
 =?UTF-8?q?=E5=B1=82=E2=80=94=E2=80=94=E6=8C=87=E6=95=B0=E9=80=80=E9=81=BF?=
 =?UTF-8?q?=E9=87=8D=E8=AF=95+=E5=8F=8B=E5=A5=BD=E9=94=99=E8=AF=AF?=
 =?UTF-8?q?=E7=BF=BB=E8=AF=91+Bot=E8=A7=A3=E6=9E=90=E5=8A=A0=E5=9B=BA?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

根因(实测复现): AGNES 外部服务不稳定, 非平台代码问题:
- 视频创建常返 500 publish_video_queue_failed / context deadline exceeded
- 连接偶发重置 Server disconnected
- 视频硬限频 每分钟 1 次 (429 rate limit)

改动:
1. chat.py 新增 _agnes_post_json(): 对 5xx/连接中断做 3 次指数退避重试(2/4s);
   4xx(参数错误)不重试直接返回
2. 新增 _friendly_media_err(): 把 AGNES 原始 HTTP/JSON 错误
   转译为对用户友好的中文(限频/连接失败/队列繁忙), 不再透传原始报错
3. _generate_media 图像/视频分支改用上述重试+友好错误
4. im_llm.py Bot 媒体结果解析加固: strip 后判 <img>/<video>,
   成功返回干净链接; 失败透传友好中文, 绝不把原始提示词/标签甩给用户

验证(容器内实测):
- WEB_IMG OK / WEB_VID OK / BOT_IMG OK (均返回真实 png/mp4)
- 429/断开/队列三类错误均正确转译为友好中文
---
 backend/python/app/api/chat.py   | 146 +++++++++++++++++++++----------
 backend/python/app/api/im_llm.py |  14 ++-
 2 files changed, 109 insertions(+), 51 deletions(-)

diff --git a/backend/python/app/api/chat.py b/backend/python/app/api/chat.py
index 98b78b5..1df62c3 100644
--- a/backend/python/app/api/chat.py
+++ b/backend/python/app/api/chat.py
@@ -239,6 +239,49 @@ async def _optimize_media_prompt(raw_prompt: str, media_type: str, db=None) -> s
     return stripped
 
 
+def _friendly_media_err(kind: str, raw: str) -> str:
+    """把 AGNES 原始错误转译为对用户友好的中文。
+
+    AGNES 常见错误：429 限频(每分钟1次视频) / Server disconnected(连接重置) /
+    publish_video_queue_failed + context deadline exceeded(队列超时)。这些都不是
+    平台代码问题，统一给出可操作的提示，不把原始 HTTP/JSON 甩给用户。
+    """
+    r = (raw or "").lower()
+    if "429" in raw or "rate limit" in r:
+        return (f"⚠️ {kind}生成请求过于频繁（AGNES 限制每分钟 1 次），"
+                f"请稍等 1 分钟后再试。")
+    if "server disconnected" in r or "连接" in raw or "reset" in r:
+        return f"⚠️ {kind}生成服务暂时连接失败，请稍后重试。"
+    if "publish_video" in r or "queue" in r or "deadline" in r or "context" in r:
+        return f"⚠️ {kind}生成队列繁忙（AGNES 处理超时），请稍后重试。"
+    return f"⚠️ {kind}生成失败（请稍后重试）：{(raw or '')[:120]}"
+
+
+async def _agnes_post_json(api_url: str, headers: dict, payload: dict, timeout) -> tuple:
+    """直连 AGNES 发 POST，对 5xx / 连接中断做指数退避重试。
+
+    AGNES 视频/图像队列极不稳定：经常返回 500 publish_video_queue_failed /
+    context deadline exceeded，或连接被重置(Server disconnected)。这类错误是 AGNES
+    侧瞬时抖动，重试通常立刻成功。4xx（参数错误）不重试，直接返回。
+    返回 (resp, err)：成功(resp.status_code<500) 时 err=None；耗尽重试时 resp=None。
+    """
+    last_err = None
+    for attempt in range(3):
+        try:
+            async with httpx.AsyncClient(
+                timeout=timeout, proxy=_proxy_for_url(api_url), trust_env=False
+            ) as client:
+                resp = await client.post(api_url, headers=headers, json=payload)
+            if resp.status_code < 500:
+                return resp, None
+            last_err = f"HTTP {resp.status_code}: {resp.text[:200]}"
+        except Exception as e:
+            last_err = str(e)[:200]
+        if attempt < 2:
+            await asyncio.sleep(2 * (attempt + 1))
+    return None, last_err
+
+
 async def _generate_media(db, ai_model, prompt: str, user_id: str = None) -> str:
     """图像/视频生成模型统一入口：按 capabilities 调用 AGNES 等媒体生成 API。
     返回可直接在 AI 聊天中渲染的富文本(含 <img>/<video> 标签)。"""
@@ -265,25 +308,27 @@ async def _generate_media(db, ai_model, prompt: str, user_id: str = None) -> str
             "prompt": optimized_prompt,
             "size": "1024x768",
         }
+        resp, err = await _agnes_post_json(api_url, headers, payload, timeout)
+        if resp is None:
+            # 3 次重试后仍失败 → AGNES 侧持续抖动，提示稍后重试而非甩原始错误
+            return _friendly_media_err("图像", err)
+        if resp.status_code != 200:
+            return _friendly_media_err("图像", f"HTTP {resp.status_code}: {resp.text[:200]}")
         try:
-            async with httpx.AsyncClient(timeout=timeout, proxy=_proxy_for_url(api_url), trust_env=False) as client:
-                resp = await client.post(api_url, headers=headers, json=payload)
-            if resp.status_code != 200:
-                return f"⚠️ 图像生成失败 (HTTP {resp.status_code}): {resp.text[:200]}"
             data = resp.json()
-            items = data.get("data") or []
-            if not items:
-                return f"⚠️ 图像生成失败：响应无图片数据 {str(data)[:200]}"
-            item = items[0]
-            url = item.get("url") or item.get("b64_json")
-            if not url:
-                return f"⚠️ 图像生成失败：未返回图片地址 {str(data)[:200]}"
-            if not url.startswith("http"):
-                url = f"data:image/png;base64,{url}"
-            return (f'<img class="md-img" src="{url}" alt="生成图片"/>\n\n'
-                    f'**提示词：** {_html_escape(prompt)}')
-        except Exception as e:
-            return f"⚠️ 图像生成异常：{str(e)[:200]}"
+        except Exception:
+            return f"⚠️ 图像生成失败：响应解析错误 {resp.text[:200]}"
+        items = data.get("data") or []
+        if not items:
+            return f"⚠️ 图像生成失败：响应无图片数据 {str(data)[:200]}"
+        item = items[0]
+        url = item.get("url") or item.get("b64_json")
+        if not url:
+            return f"⚠️ 图像生成失败：未返回图片地址 {str(data)[:200]}"
+        if not url.startswith("http"):
+            url = f"data:image/png;base64,{url}"
+        return (f'<img class="md-img" src="{url}" alt="生成图片"/>\n\n'
+                f'**提示词：** {_html_escape(prompt)}')
 
     # ============ 视频生成 (异步任务 + 轮询) ============
     if "video" in caps:
@@ -303,38 +348,45 @@ async def _generate_media(db, ai_model, prompt: str, user_id: str = None) -> str
             "num_frames": 121,
             "frame_rate": 24,
         }
+        resp, err = await _agnes_post_json(api_url, headers, payload, timeout)
+        if resp is None:
+            return _friendly_media_err("视频", err)
+        if resp.status_code != 200:
+            return _friendly_media_err("视频", f"HTTP {resp.status_code}: {resp.text[:200]}")
         try:
-            async with httpx.AsyncClient(timeout=timeout, proxy=_proxy_for_url(api_url), trust_env=False) as client:
-                resp = await client.post(api_url, headers=headers, json=payload)
-                if resp.status_code != 200:
-                    return f"⚠️ 视频生成失败 (HTTP {resp.status_code}): {resp.text[:200]}"
-                data = resp.json()
-                video_id = data.get("video_id") or data.get("task_id") or data.get("id")
-                if not video_id:
-                    return f"⚠️ 视频生成失败：未返回任务ID {str(data)[:200]}"
-                # 轮询结果(最多约 5 分钟)。AGNES 视频状态端点为 GET {base}/v1/videos/{video_id}
-                # 该端点顶层返回 status(queued/in_progress/completed/failed)，成片地址在 metadata.url
-                result_url = None
-                err_msg = None
-                for _ in range(60):
-                    await asyncio.sleep(5)
+            data = resp.json()
+        except Exception:
+            return f"⚠️ 视频生成失败：响应解析错误 {resp.text[:200]}"
+        video_id = data.get("video_id") or data.get("task_id") or data.get("id")
+        if not video_id:
+            return f"⚠️ 视频生成失败：未返回任务ID {str(data)[:200]}"
+        # 轮询结果(最多约 5 分钟)。AGNES 视频状态端点为 GET {base}/v1/videos/{video_id}
+        # 该端点顶层返回 status(queued/in_progress/completed/failed)，成片地址在 metadata.url
+        result_url = None
+        err_msg = None
+        for _ in range(60):
+            await asyncio.sleep(5)
+            try:
+                async with httpx.AsyncClient(
+                    timeout=timeout, proxy=_proxy_for_url(f"{base}/v1/videos/{video_id}"), trust_env=False
+                ) as client:
                     r2 = await client.get(f"{base}/v1/videos/{video_id}", headers=headers)
-                    if r2.status_code != 200:
-                        continue
-                    rd = r2.json()
-                    st = rd.get("status")
-                    if st == "completed":
-                        result_url = (rd.get("metadata") or {}).get("url") or rd.get("url")
-                        break
-                    if st == "failed":
-                        err_msg = rd.get("error") or (rd.get("metadata") or {}).get("error")
-                        break
-                if result_url:
-                    return (f'<video class="md-video" controls src="{result_url}"></video>\n\n'
-                            f'**提示词：** {_html_escape(prompt)}')
-                return f"⚠️ 视频生成失败：{err_msg or '生成超时，请稍后在对话中重试'}"
-        except Exception as e:
-            return f"⚠️ 视频生成异常：{str(e)[:200]}"
+            except Exception:
+                continue
+            if r2.status_code != 200:
+                continue
+            rd = r2.json()
+            st = rd.get("status")
+            if st == "completed":
+                result_url = (rd.get("metadata") or {}).get("url") or rd.get("url")
+                break
+            if st == "failed":
+                err_msg = rd.get("error") or (rd.get("metadata") or {}).get("error")
+                break
+        if result_url:
+            return (f'<video class="md-video" controls src="{result_url}"></video>\n\n'
+                    f'**提示词：** {_html_escape(prompt)}')
+        return f"⚠️ 视频生成失败：{err_msg or '生成超时，请稍后在对话中重试'}"
 
     return "⚠️ 未知的媒体模型类型。"
 
diff --git a/backend/python/app/api/im_llm.py b/backend/python/app/api/im_llm.py
index 8e222ff..a80a95c 100644
--- a/backend/python/app/api/im_llm.py
+++ b/backend/python/app/api/im_llm.py
@@ -78,11 +78,17 @@ def _call_real_llm(db: Session, user=None, prompt: str = "", platform: str = "im
                     result = f"（媒体生成失败：{e}）"
                 if result:
                     # IM 不渲染 HTML，把 <img>/<video> 还原为纯文本链接返回
-                    if result.startswith("<img") or result.startswith("<video"):
-                        m = _re.search(r'src="([^"]+)"', result)
+                    s = result.strip()
+                    if s.startswith("<img") or s.startswith("<video"):
+                        m = _re.search(r'src="([^"]+)"', s)
                         url = m.group(1) if m else ""
-                        label = "图片" if result.startswith("<img") else "视频"
-                        return f"已为你生成{label}：{url}" if url else result
+                        label = "图片" if s.startswith("<img") else "视频"
+                        return f"已为你生成{label}：{url}" if url else "（生成完成，但未能提取媒体地址）"
+                    # 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）"
                     return result
 
         # ===== 2) 文本回复 + 技能自动发现（Issue 8）=====