Merge pull request 'fix: OSS sign_url 添加 slash_safe=True 防止签名URL 403' (#2244) from fix/oss-sign-url-slash-safe into develop

sign_url() 默认将路径中的 / 编码为 %2F，导致 OSS 服务端签名校验
不匹配（SignatureDoesNotMatch 403），所有预签名 URL 无法访问。

影响范围：Ditto 口型视频 output_video_url、uploads 目录所有签名 URL。
修复：传递 slash_safe=True 保持路径中的 / 不被编码。

apps/api/app/services/lipsync_service.py

@@ -845,6 +845,30 @@ class LipsyncService:
         if job.status in (STATUS_COMPLETED, "failed"):
             return job
 
+        # Ditto 异步路径：mediakit_task_id 以 "ditto:" 开头，由 Celery 任务异步更新
+        # 不做 MediaKit 轮询，只检查是否卡住太久（>10 分钟）则标失败
+        if job.mediakit_task_id and job.mediakit_task_id.startswith("ditto:"):
+            if job.status in ("processing", "submitted"):
+                _now = datetime.now(UTC)
+                _upd = job.updated_at
+                if _upd is not None and _upd.tzinfo is None:
+                    _upd = _upd.replace(tzinfo=UTC)
+                stale_minutes = 10
+                if _upd and (_now - _upd).total_seconds() > stale_minutes * 60:
+                    logger.warning(
+                        "Ditto 异步任务超时（>%d 分钟），标记失败: job_id=%s",
+                        stale_minutes,
+                        job_id,
+                    )
+                    job.status = "failed"
+                    job.error_message = f"Ditto 处理超时（>{stale_minutes} 分钟）"
+                    job.error_code = "DittoTimeout"
+                    job.completed_at = _now
+                    job.updated_at = _now
+                    self.db.commit()
+                    self._refund_lip_sync(job)
+            return job
+
         # GPU 异步路径：mediakit_task_id 以 "gpu:" 开头，由 Celery 任务异步更新
         # 不做 MediaKit 轮询，只检查是否卡住太久（>30 分钟）则标失败
         if job.mediakit_task_id and job.mediakit_task_id.startswith("gpu:"):

apps/api/app/tasks/lipsync_ditto.py

@@ -246,8 +246,8 @@ def lipsync_ditto_process_async(self, job_id: str, user_id: str) -> None:
             # video_url 不传则用默认模板
         )
 
-        # Ditto 返回的 MP4 自带音频，直接标记完成
-        job.output_video_url = result.video_url
+        # Ditto 返回的 MP4 自带音频，签名 OSS URL（7天有效）后标记完成
+        job.output_video_url = _sign_media_url(result.video_url)
         # 探测时长（用于计费）
         duration = _probe_video_duration(result.video_bytes)
         if duration <= 0:

deploy/configs/.env.staging

@@ -311,3 +311,10 @@ GPU_ENCODE_CRF=23
 GPU_ENCODE_FALLBACK_CPU=true
 GPU_ENCODE_MEZZANINE_TRANSPORT=oss
 GPU_ENCODE_OSS_TMP_PREFIX=tmp/gpu-mezzanine/
+
+# ==================== Ditto 蚂蚁数字人口型 ====================
+# 注意：这些值必须写死在模板里（不是 CI Secret），否则每次 CI 重新渲染 .env 都会被丢弃，
+# 导致 staging 发版后 Ditto 口型服务静默降级到 GPU/MediaKit（P0 防复发）。
+USE_DITTO_LIPSYNC=true
+DITTO_API_BASE_URL=http://100.76.80.23:8000
+DITTO_DEFAULT_VIDEO_URL=https://xiaoxia-autocut.oss-cn-hangzhou.aliyuncs.com/uploads/default_avatar.mp4

packages/application/ditto_service.py

@@ -129,7 +129,9 @@ class DittoClient:
         for attempt in range(self.max_retries + 1):
             try:
                 start = time.monotonic()
-                with httpx.Client(timeout=self.timeout, follow_redirects=True) as client:
+                # 精细化超时：connect=10s（网络不通快速失败），read=120s（最长音频~45s按RTF=2.8推算）
+                _timeout = httpx.Timeout(connect=10.0, read=self.timeout, write=30.0, pool=10.0)
+                with httpx.Client(timeout=_timeout, follow_redirects=True) as client:
                     resp = client.post(url, json=payload)
                 elapsed = time.monotonic() - start
 
@@ -205,6 +207,13 @@ class DittoClient:
 
             except DittoError:
                 raise
+            except (httpx.ConnectError, httpx.NetworkError, ConnectionError, OSError) as exc:
+                # 网络不通/连接被拒（如 GPU 断网/Tailscale 掉线），不重试，直接快速回退
+                logger.warning("[ditto] 网络不可达 attempt=%d err=%s", attempt + 1, exc)
+                raise DittoError(
+                    f"Ditto 网络不可达: {exc}",
+                    code="NetworkUnreachable",
+                ) from exc
             except httpx.TimeoutException as exc:
                 last_exc = exc
                 logger.warning("[ditto] 请求超时 attempt=%d err=%s", attempt + 1, exc)
@@ -212,7 +221,7 @@ class DittoClient:
                     time.sleep(min(2**attempt, 15))
                     continue
                 raise DittoError(
-                    f"Ditto 请求超时（{self.timeout}s），重试耗尽",
+                    f"Ditto 请求超时（read={self.timeout}s），重试耗尽",
                     code="Timeout",
                 ) from exc
             except Exception as exc:

packages/config/base.py

@@ -196,9 +196,10 @@ class SharedSettings(BaseSettings):
         default=3,
         validation_alias=AliasChoices("DITTO_MAX_RETRIES", "ditto_max_retries"),
     )
-    # Ditto 单次请求超时（秒）：数字人半身视频推理通常 30-120s
+    # Ditto 单次请求 read 超时（秒）：数字人半身视频推理通常 30-120s（RTF≈2.8，40s音频约112s）
+    # connect 超时固定 10s（代码硬编码，网络不通快速失败）
     ditto_request_timeout: int = Field(
-        default=300,
+        default=120,
         validation_alias=AliasChoices("DITTO_REQUEST_TIMEOUT", "ditto_request_timeout"),
     )
 

packages/shared/storage.py

@@ -334,7 +334,7 @@ class SharedStorageService(StoragePort):
 
         storage_key = self.normalize_storage_key(storage_key_or_url)
         try:
-            signed = sign_bucket.sign_url("GET", storage_key, expires_seconds)
+            signed = sign_bucket.sign_url("GET", storage_key, expires_seconds, slash_safe=True)
             logger.info(
                 "signed URL generated for key=%s prefix=%s",
                 storage_key[:80],

tests/unit/test_ditto_service.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import time
 from unittest.mock import MagicMock, patch
 
 import httpx
@@ -195,3 +196,34 @@ def test_empty_script_replaced_with_space():
         c.generate(audio_url="http://x/a.mp3", script="")
         payload = client.post.call_args.kwargs["json"]
         assert payload["script"] == " "
+
+
+def test_generate_network_error_fails_fast(monkeypatch):
+    """网络不通（ConnectError）时不重试，直接快速抛 NetworkUnreachable，避免用户等5分钟"""
+    import httpx
+
+    from packages.application import ditto_service as ds_mod
+
+    calls = {"n": 0}
+
+    def _fake_post(self, url, json=None):
+        calls["n"] += 1
+        raise httpx.ConnectError("[Errno 113] No route to host")
+
+    monkeypatch.setattr(httpx.Client, "post", _fake_post)
+
+    client = _make_client(
+        base_url="http://100.76.80.23:8000",
+        default_video_url="http://oss/tpl.mp4",
+        max_retries=2,
+        timeout=120,
+    )
+
+    t0 = time.monotonic()
+    with pytest.raises(ds_mod.DittoError) as exc:
+        client.generate(audio_url="http://oss/a.wav", script="你好")
+    elapsed = time.monotonic() - t0
+
+    assert exc.value.code == "NetworkUnreachable"
+    assert calls["n"] == 1  # 不重试
+    assert elapsed < 5  # 快速失败<5秒