fix(worker): 注册 ai_avatar_render Celery 任务到 Worker，修复 AI数字人渲染卡 0% (#1846)

Co-authored-by: xiaoxia <dev@xiaoxiajianji.com>
Co-committed-by: xiaoxia <dev@xiaoxiajianji.com>

.gitea/workflows/debug-web-v2.yml

@@ -0,0 +1,60 @@
+name: "Debug: Web container v2 (mount conflict)"
+on:
+  push:
+    branches: [debug/web-crash-v2]
+  workflow_dispatch:
+jobs:
+  web-diag:
+    runs-on: runtime-builder
+    timeout-minutes: 10
+    steps:
+      - name: Setup SSH and diagnose
+        shell: bash
+        env:
+          STAGING_SSH_KEY: ${{ secrets.PREVIEW_SSH_KEY }}
+        run: |
+          set -x
+          which ssh || (apt-get update -qq && apt-get install -y -qq openssh-client)
+          mkdir -p ~/.ssh && chmod 700 ~/.ssh
+          printf "%s" "$STAGING_SSH_KEY" > ~/.ssh/id_rsa
+          chmod 600 ~/.ssh/id_rsa
+          H=47.98.113.167; P=22222
+          ssh-keyscan -p $P -H $H >> ~/.ssh/known_hosts 2>/dev/null
+          ssh -p $P -i ~/.ssh/id_rsa -o StrictHostKeyChecking=no root@$H 'bash -s' <<'REMOTE'
+          set -x
+          echo "=== Current staging containers ==="
+          docker ps -a --filter name=xiaoxia-*-staging --format "table {{.Names}}\t{{.Status}}\t{{.Image}}"
+          echo ""
+          echo "=== Web container logs (current/current-rolledback) ==="
+          docker logs xiaoxia-web-staging 2>&1 | tail -40
+          echo ""
+          echo "=== Web inspect: env & mounts ==="
+          docker inspect xiaoxia-web-staging --format 'Entrypoint: {{.Config.Entrypoint}} Cmd: {{.Config.Cmd}}'
+          docker inspect xiaoxia-web-staging --format '{{range .Config.Env}}{{.}}{{"\n"}}{{end}}' | grep -E "APP_ENV|VERSION"
+          echo "Mounts:"
+          docker inspect xiaoxia-web-staging --format '{{range .Mounts}}{{.Type}} {{.Source}} -> {{.Destination}} (rw={{.RW}}){{"\n"}}{{end}}'
+          echo ""
+          echo "=== Reproduce: rm on read-only bind mount ==="
+          docker run --rm --name nginx-ro-test \
+            -v /var/lib/xiaoxia-saas-staging/nginx-staging.conf:/etc/nginx/conf.d/default.conf:ro \
+            git.xiaoxiajianji.com/xiaoxia/xiaoxia-saas/xiaoxia-saas-web:387514c \
+            sh -c '
+              set -x
+              echo "Before:"
+              ls -la /etc/nginx/conf.d/
+              echo "Try rm (as entrypoint does):"
+              rm -f /etc/nginx/conf.d/default.conf
+              echo "rm exitcode=$?"
+              echo "After rm:"
+              ls -la /etc/nginx/conf.d/
+              echo "Test ln:"
+              ln -s /etc/nginx/nginx-staging.conf /etc/nginx/conf.d/default.conf
+              echo "ln exitcode=$?"
+              ls -la /etc/nginx/conf.d/
+              echo "nginx -t:"
+              nginx -t 2>&1
+            ' 2>&1
+          echo ""
+          echo "=== Also test with NEW fixed image (9c0d4b1 if present) ==="
+          docker images | grep xiaoxia-saas-web | head -5
+REMOTE

apps/api/app/services/lipsync_service.py

@@ -211,12 +211,18 @@ class LipsyncService:
                         normalize_emotion(emotion),
                     )
                 )
-            except Exception:
-                logger.warning(
-                    "Celery 任务提交失败，TTS 任务已创建但未触发执行: %s",
+            except Exception as exc:
+                # 投递失败时立即把 job 标成 failed 并写入 error_message，
+                # 前端轮询时能直接看到失败原因，不会无限卡在 tts_processing。
+                logger.exception(
+                    "Celery 任务提交失败，TTS 任务已创建但未触发执行: job_id=%s err=%s",
                     job_id,
-                    exc_info=True,
+                    exc,
                 )
+                job.status = "failed"
+                job.error_message = f"Celery 任务投递失败: {exc}"
+                job.error_code = "AsyncDispatchFailed"
+                job.updated_at = datetime.now(timezone.utc)
         else:
             # 2b. 直接音频模式：同步签名并提交 MediaKit
             video_url = self._sign_media_url(video_url)

apps/api/app/tasks/lipsync_tts.py

@@ -76,11 +76,21 @@ def tts_synthesize_and_submit(
     from app.services.mediakit_client import MediaKitError, get_mediakit_client
     from sqlalchemy.orm import Session as DBSession
 
-    from packages.adapters.sqlalchemy_impl.database import SessionLocal
     from packages.adapters.sqlalchemy_impl.models import LipsyncJobModel
     from packages.application.cosyvoice_service import CosyVoiceError, CosyVoiceService
     from packages.shared.url_security import safe_download_bytes
 
+    # SessionLocal 获取：
+    # - API 容器：app.db.SessionLocal（环境变量完整，导入即建引擎）
+    # - Worker 容器：worker_app.db.SessionLocal（Worker 自己的 settings 初始化引擎）
+    # API 侧没有 worker_app 模块 → ImportError 直接回退；
+    # Worker 侧 app.db 会因缺少 API 专有环境变量抛 pydantic ValidationError，
+    # 此时也要回退到 worker_app.db。
+    try:
+        from worker_app.db import SessionLocal  # type: ignore
+    except Exception:  # noqa: BLE001
+        from app.db import SessionLocal  # type: ignore
+
     db: DBSession = SessionLocal()
     try:
         job = (

apps/worker/worker_app/celery_app.py

@@ -34,6 +34,8 @@ celery_app.conf.imports = (
     "worker_app.tasks.tts_synthesis",
     "worker_app.tasks.batch_download",
     "worker_app.tasks.duplication_check",
+    # #1798 AI 数字人渲染：必须在 Worker 实例上注册同名任务，否则消息无人消费（渲染卡 0%）
+    "worker_app.tasks.ai_avatar_render",
     "worker_app.tasks._startup",
     "apps.worker.video_processing.dedup",
     "worker_app.tasks.cleanup",

apps/worker/worker_app/tasks/ai_avatar_render.py

@@ -0,0 +1,48 @@
+"""AI 数字人渲染任务 — Worker 侧 Celery 任务注册.
+
+#1798 渲染进度卡在 0% 的根因：渲染任务定义在 API 侧（`app.tasks.ai_avatar_render`），
+装饰在 API 自己的 Celery 实例（`xiaoxia-saas-api`）上；而 Worker 用的是
+`worker_app.celery_app` 实例，`conf.imports` 从未导入该任务，Worker 的任务
+注册表里没有 `ai_avatar_render.execute`，消息被路由到默认 `celery` 队列后
+无人消费，任务永远停在 0%。
+
+修复：在 Worker 侧用 `worker_app.celery_app` 注册同名任务，直接调用与 API
+服务一致的 `AiAvatarRenderService.execute_render` 核心管线（业务逻辑在
+`apps.api.app.services`，worker 镜像已复制 `apps/api/app`）。任务名保持
+`ai_avatar_render.execute`，与 API 生产端 `.delay()` 的消息路由一致；未在
+task_routes 显式配置，走默认 `celery` 队列，由 transcode worker 消费。
+"""
+
+from __future__ import annotations
+
+import logging
+
+from worker_app.celery_app import celery_app
+from worker_app.db import SessionLocal
+
+logger = logging.getLogger(__name__)
+
+
+@celery_app.task(bind=True, name="ai_avatar_render.execute", max_retries=2)
+def execute_ai_avatar_render(self, job_id: str) -> dict:
+    """执行 AI 数字人渲染管线（Worker 侧入口）.
+
+    进度由 service 直接写入 DB（AiAvatarRenderJob.progress:
+    0→5→20→40→80→90→95→100），API 通过轮询 progress 字段展示。
+    """
+    logger.info("开始执行渲染任务: %s", job_id)
+    self.update_state(state="PROCESSING", meta={"progress": 0, "job_id": job_id})
+
+    session = SessionLocal()
+    try:
+        from app.services.ai_avatar_render_service import AiAvatarRenderService
+
+        service = AiAvatarRenderService(session)
+        service.execute_render(job_id)
+        return {"status": "completed", "job_id": job_id}
+    except Exception as exc:
+        logger.exception("渲染任务执行异常 [%s]: %s", job_id, exc)
+        self.update_state(state="FAILED", meta={"progress": 0, "error": str(exc)})
+        raise
+    finally:
+        session.close()

infra/docker/nginx-entrypoint.sh

@@ -1,18 +1,49 @@
 #!/bin/sh
 # Select nginx config based on APP_ENV (staging/production).
-# Both configs are baked into the image at well-known paths.
-# nginx reads config only at startup, so symlink before exec.
+#
+# 两种运行模式：
+# 1. CI/CD 部署（staging/production）：部署脚本通过 `-v 宿主机文件:/etc/nginx/conf.d/default.conf:ro`
+#    把宿主机生成的带 resolver/docker upstream 的配置 bind mount 进来，entrypoint 不应改动。
+#    bind mount 的文件是 readonly 的，rm 会报 EBUSY ("Resource busy")，直接 exec nginx 即可。
+# 2. 本地 docker-compose / 直接 `docker run`（无外部挂载）：镜像烤入了 nginx-staging.conf 与
+#    nginx-production.conf 到 /etc/nginx/，entrypoint 根据 APP_ENV 把 default.conf 换成正确的 symlink。
+#
+# 策略：
+# - 如果 /etc/nginx/conf.d/default.conf 已经是指向目标 conf 的 symlink，什么都不做；
+# - 否则尝试 rm -f 再 ln -s；rm 失败说明是外部 bind mount（已有正确配置），不阻塞启动；
+# - 兜底：只要 conf.d 目录里有 .conf 文件（含 bind mount 来的），就直接启动 nginx。
 set -e
 
 NGINX_CONF_DIR="/etc/nginx/conf.d"
+TARGET_CONF=""
 
 case "${APP_ENV:-production}" in
   staging)
-    ln -sf /etc/nginx/nginx-staging.conf "$NGINX_CONF_DIR/default.conf"
+    TARGET_CONF="/etc/nginx/nginx-staging.conf"
     ;;
   *)
-    ln -sf /etc/nginx/nginx-production.conf "$NGINX_CONF_DIR/default.conf"
+    TARGET_CONF="/etc/nginx/nginx-production.conf"
     ;;
 esac
 
+DEFAULT_CONF="$NGINX_CONF_DIR/default.conf"
+
+# 1. 已经是正确的 symlink：直接启动
+if [ -L "$DEFAULT_CONF" ] && [ "$(readlink "$DEFAULT_CONF" 2>/dev/null)" = "$TARGET_CONF" ]; then
+  exec nginx -g "daemon off;"
+fi
+
+# 2. 尝试替换为目标 symlink（无 bind mount 的场景）
+#    若 rm 失败（bind mount readonly，EBUSY/EPERM），则认为外部已注入配置，不阻塞。
+rm -f "$DEFAULT_CONF" 2>/dev/null || true
+if [ -f "$TARGET_CONF" ] && [ ! -e "$DEFAULT_CONF" ]; then
+  ln -s "$TARGET_CONF" "$DEFAULT_CONF" 2>/dev/null || true
+fi
+
+# 3. 兜底：至少要有一个 .conf 文件，否则 nginx 起不来
+if ! ls "$NGINX_CONF_DIR"/*.conf >/dev/null 2>&1; then
+  echo "ERROR: no nginx config found in $NGINX_CONF_DIR (tried $TARGET_CONF and external bind mount)" >&2
+  exit 1
+fi
+
 exec nginx -g "daemon off;"

packages/__init__.py

@@ -0,0 +1 @@
+# xiaoxia-saas shared packages namespace

packages/adapters/__init__.py

@@ -0,0 +1 @@
+# adapter implementations namespace

tests/unit/test_lipsync_speed_optimization.py

@@ -143,8 +143,8 @@ class TestCreateJobAsyncTTS:
         assert args[2] == "v-1"  # voice_id
         assert args[3] == "测试文本"  # script_text
 
-    def test_tts_mode_celery_dispatch_failure_still_creates_job(self):
-        """Celery dispatch 失败时，job 记录已创建，状态保持 tts_processing."""
+    def test_tts_mode_celery_dispatch_failure_marks_job_failed(self):
+        """Celery dispatch 失败时，job 标为 failed 并写入 error_message，前端轮询能直接看到错误."""
         svc, client, cosy = _make_service_with_mocks()
 
         with patch("app.services.lipsync_service.tts_synthesize_and_submit") as mock_task:
@@ -157,9 +157,11 @@ class TestCreateJobAsyncTTS:
                 script_text="测试文本",
             )
 
-        # job 已创建
+        # job 已创建且状态标为 failed
         assert job is not None
-        assert job.status == "tts_processing"
+        assert job.status == "failed"
+        assert "Celery 任务投递失败" in job.error_message
+        assert job.error_code == "AsyncDispatchFailed"
         # MediaKit 未被调用
         client.submit_lipsync.assert_not_called()
 

tests/unit/test_lipsync_tts.py

@@ -73,12 +73,16 @@ def _apply_all_patches(
     lipsync_tts.run() 在函数体内部懒 import 多个模块，通过 sys.modules 注入
     伪造包路径避免真实导入；对存在的模块用 patch() 替换返回值/side_effect。
     """
-    fake_db_mod = ModuleType("packages.adapters.sqlalchemy_impl.database")
+    # SessionLocal 通过懒探测获取（Worker 用 worker_app.db，API 用 app.db），
+    # 测试环境里两个模块都能被真实导入，必须同时 mock 保证用的是 fake session。
+    fake_app_db = ModuleType("app.db")
+    fake_worker_db = ModuleType("worker_app.db")
     session, factory = _build_session(job)
-    fake_db_mod.SessionLocal = factory
+    fake_app_db.SessionLocal = factory
+    fake_worker_db.SessionLocal = factory
 
     patches = [
-        patch.dict(sys.modules, {"packages.adapters.sqlalchemy_impl.database": fake_db_mod}),
+        patch.dict(sys.modules, {"app.db": fake_app_db, "worker_app.db": fake_worker_db}),
         patch(
             "app.tasks.lipsync_tts._sign_media_url",
             side_effect=lambda url: url + "?signed" if url else url,
@@ -289,9 +293,11 @@ class TestTtsSynthesizeAndSubmit:
         from app.tasks.lipsync_tts import tts_synthesize_and_submit
 
         job = _make_fake_job()
-        fake_db_mod = ModuleType("packages.adapters.sqlalchemy_impl.database")
+        fake_app_db = ModuleType("app.db")
+        fake_worker_db = ModuleType("worker_app.db")
         session, factory = _build_session(job)
-        fake_db_mod.SessionLocal = factory
+        fake_app_db.SessionLocal = factory
+        fake_worker_db.SessionLocal = factory
 
         # CosyVoiceService 在 __init__ 抛 RuntimeError（非 CosyVoiceError/ValueError）
         fake_cosy_mod = ModuleType("packages.application.cosyvoice_service")
@@ -309,7 +315,8 @@ class TestTtsSynthesizeAndSubmit:
         with patch.dict(
             sys.modules,
             {
-                "packages.adapters.sqlalchemy_impl.database": fake_db_mod,
+                "app.db": fake_app_db,
+                "worker_app.db": fake_worker_db,
                 "packages.application.cosyvoice_service": fake_cosy_mod,
             },
         ):