refactor: 改造为通用产线执行引擎模块

- 去掉独立 aiohttp 服务器，改为标准模块（load_pipeline_service） - 存储从文件系统改 MySQL（sqlor） - 新增 3 张数据表：pipeline_tasks/task_steps/artifacts - 多租户隔离（tenant_id） - 通用 DAG 调度引擎（读 pipeline_steps 表，不硬编码业务） - 可插拔步骤处理器（register_handler by step_type） - artifact 版本管理 + 级联重跑 - init/data.json 标准 appcodes 格式 - 完整 README 文档
2026-06-11 17:30:06 +08:00 · 2026-06-11 17:30:06 +08:00 · 2448ad45f7
commit 2448ad45f7
parent 5124b4a047
16 changed files with 1120 additions and 854 deletions
--- a/.gitignore
+++ b/.gitignore
@ -1,4 +1,25 @@
+# Python
 __pycache__/
-*.pyc
+*.py[cod]
 *.egg-info/
+dist/
 build/
+*.egg
+
+# Virtual environment
+py3/
+pkgs/
+
+# Runtime
+logs/
+*.pid
+
+# IDE
+.vscode/
+.idea/
+
+# OS
+.DS_Store
+
+# DDL output
+mysql.ddl.sql
--- a/README.md
+++ b/README.md
@ -1,53 +1,117 @@
-# Pipeline-Service — 产线任务引擎（早期版本）
+# pipeline-service 通用产线执行引擎

-KTV/MV 视频制作流水线的后端引擎，负责编排和执行多步骤产线任务。
+## 定位

-## 功能
+通用产线执行引擎模块。把 Hermes Agent 验证过的业务流程固化为可重复、可并发的产线业务环境。

- **任务存储**：JSON 文件存储产线任务状态
- **步骤编排**：支持串行/并行步骤执行
- **状态管理**：pending → running → completed/failed
- **版本追踪**：每个步骤的输入/输出可追溯
+## 核心价值
+
+- Hermes Agent 中用 cron/delegate/terminal 跑通的流程 → 固化为产线步骤定义 → pipeline-service 自动调度执行
+- 一次验证，无限次自动执行
+- 多租户并发：同一产线，不同租户同时使用，数据完全隔离

 ## 架构

-```python
-pipeline_service/
-├── engine.py        # 产线执行引擎（步骤调度）
-├── models.py        # 数据模型（Task, Step, Node）
-├── storage.py       # JSON 文件存储层
-└── __init__.py
+```
+宿主应用 (pipeline-app / sage / 其他)
+  │
+  ├── load_pipeline_service()  ← 注册函数到 ServerEnv
+  │
+  ├── pipeline_submit(tenant_id, pipeline_id, owner_id, title, params)
+  ├── pipeline_list(tenant_id, pipeline_id?)
+  ├── pipeline_detail(tenant_id, task_id)
+  ├── pipeline_node(tenant_id, task_id, step_name, version?)
+  ├── pipeline_modify(tenant_id, task_id, updates, rerun_from)
+  ├── pipeline_pause(tenant_id, task_id)
+  ├── pipeline_resume(tenant_id, task_id)
+  ├── pipeline_cancel(tenant_id, task_id)
+  └── pipeline_register_handler(step_type, fn)  ← 注册步骤处理器
 ```

-## 使用方式
+## 引擎工作原理

-```python
-from pipeline_service.engine import PipelineEngine
+1. **提交任务** → 读取 pipeline_steps 表的步骤定义 → 创建 pipeline_task_steps 记录 → 启动执行
+2. **执行循环** → 解析 DAG 依赖图 → 找到可执行步骤（所有前置完成）→ 调用 handler → 存 artifact → 继续下一步
+3. **多租户** → 所有查询按 tenant_id 隔离 → 同一产线多租户并发不冲突
+4. **人工干预** → 修改节点 artifact → 创建新版本 → BFS 计算受影响步骤 → 级联重跑

-engine = PipelineEngine(storage_dir="./tasks")
-task_id = engine.submit({
-    "name": "KTV视频制作",
-    "steps": [
-        {"type": "ai_video", "config": {...}},
-        {"type": "demucs", "config": {...}},
-        {"type": "asr", "config": {...}},
-        {"type": "ffmpeg_compose", "config": {...}}
-    ]
-})
-engine.run(task_id)
-```
+## 数据表

-## 当前状态
-
-本模块为早期原型，核心引擎已编写但：
- 未与 Sage 平台联调
- 未部署到生产环境
- 后续由 pipeline-app 替代为独立服务架构
-
-## 相关仓库
-
-| 仓库 | 说明 |
+| 表名 | 用途 |
 |------|------|
-| [pipeline-app](https://git.opencomputing.cn/yumoqing/pipeline-app) | 替代方案：独立 ahserver 后端服务 |
-| [pipeline](https://git.opencomputing.cn/yumoqing/pipeline) | Sage 前端桥接模块 |
-| [showcase](https://git.opencomputing.cn/yumoqing/showcase) | 展示平台（产线产出物展示） |
+| pipeline_tasks | 任务主表（tenant_id 隔离） |
+| pipeline_task_steps | 任务步骤执行记录 |
+| pipeline_artifacts | 步骤产物（input/output，支持版本） |
+| pipeline_steps | 产线步骤定义（由 pipeline_core 模块管理） |
+| pipelines | 产线定义（由 pipeline_core 模块管理） |
+
+## 步骤处理器
+
+可插拔注册，统一接口：
+
+```python
+async def handler(tenant_id, task_id, step_name, input_data, config) -> dict:
+    # 处理逻辑
+    return output_data
+
+# 注册
+from pipeline_service import register_handler
+register_handler("llm_generate", handler)
+```
+
+处理器按 `step_type` 匹配。步骤定义中的 `step_type` 对应 handler 注册名。
+
+## 宿主集成
+
+任何应用只需一行代码即可使用：
+
+```python
+from pipeline_service.init import load_pipeline_service
+load_pipeline_service()
+```
+
+宿主负责：
+- HTTP 路由（ahserver 管）
+- 用户认证（RBAC 管）
+- 前端交互（bricks 管）
+- 产线定义和定价（pipeline_core/ops/dist 管）
+
+pipeline-service 只做：调度 + 执行 + 存储。
+
+## 目录结构
+
+```
+pipeline-service/
+├── pipeline_service/
+│   ├── __init__.py        # 包导出
+│   ├── init.py            # load_pipeline_service() + ServerEnv 注册
+│   ├── state.py           # DAG 解析、步骤状态机
+│   ├── handler.py         # 步骤处理器注册表
+│   ├── storage.py         # MySQL 存储层（sqlor）
+│   └── executor.py        # 执行循环
+├── models/
+│   ├── pipeline_tasks.json
+│   ├── pipeline_task_steps.json
+│   └── pipeline_artifacts.json
+├── init/
+│   └── data.json          # appcodes 初始化数据
+├── scripts/
+│   └── load_path.py       # RBAC 权限注册
+├── pyproject.toml
+├── build.sh
+└── README.md
+```
+
+## 构建与部署
+
+```bash
+cd ~/repos/pipeline-service
+pip install .
+
+# 建表
+json2ddl mysql models/ > mysql.ddl.sql
+mysql -u root -p pipeline < mysql.ddl.sql
+
+# 加载 appcodes
+# (通过宿主应用的 build.sh 自动加载 init/data.json)
+```
--- a/build.sh
+++ b/build.sh
@ -0,0 +1,49 @@
+#!/usr/bin/env bash
+set -e
+cdir=$(cd "$(dirname "$0")" && pwd)
+cd "$cdir"
+
+echo "=== pipeline-service Build ==="
+
+# 1. Create venv if not exists
+if [ ! -d py3 ]; then
+    python3 -m venv py3
+fi
+source py3/bin/activate
+
+# 2. Install foundation packages
+mkdir -p pkgs
+for m in apppublic sqlor ahserver xls2ddl; do
+    echo "install $m ..."
+    cd "$cdir/pkgs"
+    if [ ! -d "$m" ]; then
+        git clone https://git.opencomputing.cn/yumoqing/$m 2>/dev/null || echo "SKIP: $m"
+    fi
+    if [ -d "$m" ]; then
+        cd "$m"
+        "$cdir/py3/bin/pip" install . 2>&1 | tail -1
+    fi
+    cd "$cdir"
+done
+
+# 3. Install self
+echo "install pipeline_service ..."
+"$cdir/py3/bin/pip" install . 2>&1 | tail -1
+
+# 4. Generate DDL from models
+if [ -d models ] && ls models/*.json >/dev/null 2>&1; then
+    echo "generating DDL ..."
+    "$cdir/py3/bin/json2ddl" mysql models/ > "$cdir/mysql.ddl.sql" 2>/dev/null || echo "  DDL generation skipped (json2ddl not available)"
+fi
+
+# 5. Create runtime dirs
+mkdir -p "$cdir/logs"
+
+chmod +x "$cdir/scripts/load_path.py" 2>/dev/null || true
+echo "=== Build complete ==="
+echo ""
+echo "Next steps:"
+echo "  1. CREATE DATABASE pipeline;"
+echo "  2. mysql -u root -p pipeline < mysql.ddl.sql"
+echo "  3. Load appcodes: init/data.json via host app build.sh"
+echo "  4. Register step handlers in host app"
--- a/init/data.json
+++ b/init/data.json
@ -0,0 +1,49 @@
+{
+    "appcodes": [
+        {
+            "parentid": "task_state",
+            "parentname": "任务状态",
+            "items": [
+                {"k": "submitted", "v": "已提交"},
+                {"k": "running", "v": "运行中"},
+                {"k": "completed", "v": "已完成"},
+                {"k": "failed", "v": "失败"},
+                {"k": "paused", "v": "已暂停"},
+                {"k": "cancelled", "v": "已取消"}
+            ]
+        },
+        {
+            "parentid": "step_state",
+            "parentname": "步骤状态",
+            "items": [
+                {"k": "pending", "v": "等待中"},
+                {"k": "running", "v": "执行中"},
+                {"k": "completed", "v": "已完成"},
+                {"k": "failed", "v": "失败"},
+                {"k": "skipped", "v": "已跳过"}
+            ]
+        },
+        {
+            "parentid": "io_type",
+            "parentname": "产物类型",
+            "items": [
+                {"k": "input", "v": "输入"},
+                {"k": "output", "v": "输出"}
+            ]
+        },
+        {
+            "parentid": "step_type",
+            "parentname": "步骤类型",
+            "items": [
+                {"k": "audio_process", "v": "音频处理"},
+                {"k": "video_process", "v": "视频处理"},
+                {"k": "llm_generate", "v": "LLM生成"},
+                {"k": "llm_evaluate", "v": "LLM评估"},
+                {"k": "api_call", "v": "外部API调用"},
+                {"k": "file_process", "v": "文件处理"},
+                {"k": "composite", "v": "合成"},
+                {"k": "custom", "v": "自定义"}
+            ]
+        }
+    ]
+}
--- a/models/pipeline_artifacts.json
+++ b/models/pipeline_artifacts.json
@ -0,0 +1,27 @@
+{
+    "summary": [
+        {
+            "name": "pipeline_artifacts",
+            "title": "步骤产物表",
+            "primary": ["id"],
+            "catelog": "entity"
+        }
+    ],
+    "fields": [
+        {"name": "id", "title": "主键ID", "type": "str", "length": 32, "nullable": "no"},
+        {"name": "task_id", "title": "任务ID", "type": "str", "length": 32, "nullable": "no"},
+        {"name": "version", "title": "版本号", "type": "int", "nullable": "no", "default": "1"},
+        {"name": "step_name", "title": "步骤名称", "type": "str", "length": 64, "nullable": "no"},
+        {"name": "io_type", "title": "产物类型(input/output)", "type": "str", "length": 16, "nullable": "no"},
+        {"name": "data", "title": "产物数据(JSON)", "type": "longtext"},
+        {"name": "saved_at", "title": "保存时间", "type": "timestamp", "nullable": "no"}
+    ],
+    "indexes": [
+        {"name": "idx_pa_task_ver", "idxtype": "index", "idxfields": ["task_id", "version"]},
+        {"name": "idx_pa_step", "idxtype": "index", "idxfields": ["step_name"]},
+        {"name": "idx_pa_lookup", "idxtype": "unique", "idxfields": ["task_id", "version", "step_name", "io_type"]}
+    ],
+    "codes": [
+        {"field": "io_type", "table": "appcodes_kv", "valuefield": "k", "textfield": "v", "cond": "parentid='io_type'"}
+    ]
+}
--- a/models/pipeline_task_steps.json
+++ b/models/pipeline_task_steps.json
@ -0,0 +1,31 @@
+{
+    "summary": [
+        {
+            "name": "pipeline_task_steps",
+            "title": "任务步骤执行表",
+            "primary": ["id"],
+            "catelog": "entity"
+        }
+    ],
+    "fields": [
+        {"name": "id", "title": "主键ID", "type": "str", "length": 32, "nullable": "no"},
+        {"name": "task_id", "title": "任务ID", "type": "str", "length": 32, "nullable": "no"},
+        {"name": "step_name", "title": "步骤名称", "type": "str", "length": 64, "nullable": "no"},
+        {"name": "step_type", "title": "步骤类型(用于handler匹配)", "type": "str", "length": 64, "nullable": "no"},
+        {"name": "display_name", "title": "显示名称", "type": "str", "length": 128},
+        {"name": "step_order", "title": "执行顺序", "type": "int", "nullable": "no", "default": "0"},
+        {"name": "deps", "title": "依赖步骤(JSON数组)", "type": "text", "default": "'[]'"},
+        {"name": "state", "title": "步骤状态", "type": "str", "length": 32, "nullable": "no", "default": "pending"},
+        {"name": "error_msg", "title": "错误信息", "type": "text"},
+        {"name": "started_at", "title": "开始执行时间", "type": "timestamp"},
+        {"name": "completed_at", "title": "完成时间", "type": "timestamp"}
+    ],
+    "indexes": [
+        {"name": "idx_pts_task", "idxtype": "index", "idxfields": ["task_id"]},
+        {"name": "idx_pts_state", "idxtype": "index", "idxfields": ["state"]}
+    ],
+    "codes": [
+        {"field": "state", "table": "appcodes_kv", "valuefield": "k", "textfield": "v", "cond": "parentid='step_state'"},
+        {"field": "step_type", "table": "appcodes_kv", "valuefield": "k", "textfield": "v", "cond": "parentid='step_type'"}
+    ]
+}
--- a/models/pipeline_tasks.json
+++ b/models/pipeline_tasks.json
@ -0,0 +1,31 @@
+{
+    "summary": [
+        {
+            "name": "pipeline_tasks",
+            "title": "产线任务表",
+            "primary": ["id"],
+            "catelog": "entity"
+        }
+    ],
+    "fields": [
+        {"name": "id", "title": "主键ID", "type": "str", "length": 32, "nullable": "no"},
+        {"name": "tenant_id", "title": "租户ID", "type": "str", "length": 32, "nullable": "no"},
+        {"name": "pipeline_id", "title": "产线定义ID", "type": "str", "length": 32, "nullable": "no"},
+        {"name": "owner_id", "title": "提交人ID", "type": "str", "length": 32, "nullable": "no"},
+        {"name": "title", "title": "任务标题", "type": "str", "length": 255, "nullable": "no"},
+        {"name": "state", "title": "任务状态", "type": "str", "length": 32, "nullable": "no", "default": "submitted"},
+        {"name": "current_version", "title": "当前版本号", "type": "int", "nullable": "no", "default": "1"},
+        {"name": "params", "title": "提交参数", "type": "text"},
+        {"name": "created_at", "title": "创建时间", "type": "timestamp", "nullable": "no"},
+        {"name": "updated_at", "title": "更新时间", "type": "timestamp", "nullable": "no"}
+    ],
+    "indexes": [
+        {"name": "idx_pt_tenant", "idxtype": "index", "idxfields": ["tenant_id"]},
+        {"name": "idx_pt_pipeline", "idxtype": "index", "idxfields": ["pipeline_id"]},
+        {"name": "idx_pt_owner", "idxtype": "index", "idxfields": ["owner_id"]},
+        {"name": "idx_pt_state", "idxtype": "index", "idxfields": ["state"]}
+    ],
+    "codes": [
+        {"field": "state", "table": "appcodes_kv", "valuefield": "k", "textfield": "v", "cond": "parentid='task_state'"}
+    ]
+}
--- a/pipeline_service/init.py
+++ b/pipeline_service/init.py
@ -1,3 +1,21 @@
-"""Pipeline Service - Hermes backend for KTV/MV production pipelines"""
+"""pipeline_service - 通用产线执行引擎模块"""

-__version__ = "1.0.0"
+from .init import (
+    load_pipeline_service,
+    pipeline_submit,
+    pipeline_list,
+    pipeline_detail,
+    pipeline_node,
+    pipeline_modify,
+    pipeline_pause,
+    pipeline_resume,
+    pipeline_cancel,
+    pipeline_handlers,
+)
+from .handler import register_handler, list_handlers, get_handler
+from .state import (
+    STATE_PENDING, STATE_RUNNING, STATE_COMPLETED, STATE_FAILED, STATE_SKIPPED,
+    TASK_SUBMITTED, TASK_RUNNING, TASK_COMPLETED, TASK_FAILED, TASK_PAUSED, TASK_CANCELLED,
+)
+
+__version__ = "2.0.0"
--- a/pipeline_service/executor.py
+++ b/pipeline_service/executor.py
@ -1,182 +1,180 @@
-"""Pipeline step executor - runs steps asynchronously.
-
-Each step function receives (pipeline_id, manifest, params) and returns output data.
-The executor handles state transitions, artifact storage, and dependency resolution.
-"""
+"""Pipeline execution engine - schedules and runs steps."""

 import asyncio
 import logging
-from typing import Dict, Optional
+import traceback
+from typing import Dict

 from .state import (
    STATE_PENDING, STATE_RUNNING, STATE_COMPLETED, STATE_FAILED,
-    PIPELINE_RUNNING, PIPELINE_COMPLETED, PIPELINE_FAILED, PIPELINE_PAUSED,
-    build_dependency_map,
+    TASK_RUNNING, TASK_COMPLETED, TASK_FAILED, TASK_PAUSED,
+    build_step_graph, find_next_step,
+    check_all_completed, check_any_failed,
 )
 from .storage import (
-    get_manifest, save_manifest, save_artifact,
-    update_step_state, create_new_version, reset_steps,
+    get_pipeline_steps, get_step_states,
+    update_task_state, update_step_state,
+    save_artifact, get_artifact,
 )
+from .handler import get_handler

 logger = logging.getLogger("pipeline.executor")

-# Active pipeline execution tasks (pipeline_id -> asyncio.Task)
+# Active tasks: task_id -> asyncio.Task
 _active_tasks: Dict[str, asyncio.Task] = {}


-async def start_pipeline(pipeline_id: str):
-    """Start executing a pipeline from the beginning."""
-    task = asyncio.create_task(_run_pipeline(pipeline_id))
-    _active_tasks[pipeline_id] = task
+async def start_task(task_id: str):
+    """Start executing a pipeline task."""
+    task = asyncio.create_task(_run_task(task_id))
+    _active_tasks[task_id] = task
    return task


-async def resume_pipeline(pipeline_id: str):
-    """Resume a paused pipeline."""
-    task = asyncio.create_task(_run_pipeline(pipeline_id))
-    _active_tasks[pipeline_id] = task
+async def resume_task(task_id: str):
+    """Resume a paused task."""
+    task = asyncio.create_task(_run_task(task_id))
+    _active_tasks[task_id] = task
    return task


-async def stop_pipeline(pipeline_id: str):
-    """Cancel a running pipeline."""
-    task = _active_tasks.get(pipeline_id)
+async def stop_task(task_id: str):
+    """Cancel a running task."""
+    task = _active_tasks.get(task_id)
    if task and not task.done():
        task.cancel()
-        del _active_tasks[pipeline_id]
+        _active_tasks.pop(task_id, None)
        return True
    return False


-def is_running(pipeline_id: str) -> bool:
-    """Check if a pipeline is currently running."""
-    task = _active_tasks.get(pipeline_id)
+def is_running(task_id: str) -> bool:
+    """Check if a task is currently executing."""
+    task = _active_tasks.get(task_id)
    return task is not None and not task.done()


-async def _run_pipeline(pipeline_id: str):
-    """Main pipeline execution loop."""
+async def _run_task(task_id: str):
+    """Main execution loop for a pipeline task."""
    try:
+        # Get task info (no tenant_id needed here - internal execution)
+        db_info = await _get_task_raw(task_id)
+        if not db_info:
+            logger.error(f"Task {task_id} not found")
+            return
+
+        pipeline_id = db_info.get('pipeline_id', '')
+
+        # Load step definitions and build graph
+        step_records = await get_pipeline_steps(pipeline_id)
+        if not step_records:
+            logger.error(f"No steps defined for pipeline {pipeline_id}")
+            await update_task_state(task_id, TASK_FAILED)
+            return
+
+        step_graph = build_step_graph(step_records)
+
+        await update_task_state(task_id, TASK_RUNNING)
+
        while True:
-            manifest = get_manifest(pipeline_id)
-            if not manifest:
-                logger.error(f"Pipeline {pipeline_id} not found")
+            # Check if paused/cancelled
+            current = await _get_task_raw(task_id)
+            if not current:
+                break
+            state = current.get('state', current.get('State', ''))
+            if state in (TASK_PAUSED, TASK_FAILED, 'cancelled'):
+                logger.info(f"Task {task_id} stopped: {state}")
                break

-            if manifest["state"] == PIPELINE_PAUSED:
-                logger.info(f"Pipeline {pipeline_id} paused, waiting for user action")
+            # Get current step states
+            step_states = await get_step_states(task_id)
+
+            # Check completion
+            if check_all_completed(step_states):
+                await update_task_state(task_id, TASK_COMPLETED)
+                logger.info(f"Task {task_id} completed")
                break

-            if manifest["state"] == PIPELINE_COMPLETED:
-                logger.info(f"Pipeline {pipeline_id} already completed")
+            if check_any_failed(step_states):
+                await update_task_state(task_id, TASK_FAILED)
+                logger.warning(f"Task {task_id} failed (step failure)")
                break

-            # Find next pending step with all deps completed
-            next_step = _find_next_step(manifest)
+            # Find next executable step
+            next_step = find_next_step(step_graph, step_states)
            if not next_step:
-                # All steps done
-                manifest["state"] = PIPELINE_COMPLETED
-                save_manifest(pipeline_id, manifest)
-                logger.info(f"Pipeline {pipeline_id} completed")
+                # No executable step but not all completed - deadlock or waiting
+                logger.warning(f"Task {task_id}: no executable step, states={step_states}")
+                await update_task_state(task_id, TASK_FAILED)
                break

            # Execute the step
-            await _execute_step(pipeline_id, next_step)
+            await _execute_step(task_id, next_step, step_graph, current)

    except asyncio.CancelledError:
-        logger.info(f"Pipeline {pipeline_id} cancelled")
+        logger.info(f"Task {task_id} cancelled")
    except Exception as e:
-        logger.error(f"Pipeline {pipeline_id} error: {e}")
+        logger.error(f"Task {task_id} error: {e}\n{traceback.format_exc()}")
    finally:
-        _active_tasks.pop(pipeline_id, None)
+        _active_tasks.pop(task_id, None)


-def _find_next_step(manifest: dict) -> Optional[str]:
-    """Find the next step to execute (pending, all deps completed)."""
-    steps = manifest.get("steps", {})
-    for name, info in sorted(steps.items(), key=lambda x: x[1].get("order", 0)):
-        if info["state"] != STATE_PENDING:
-            continue
-        # Check all dependencies are completed
-        deps_ok = all(
-            steps.get(dep, {}).get("state") == STATE_COMPLETED
-            for dep in info.get("deps", [])
-        )
-        if deps_ok:
-            return name
+async def _get_task_raw(task_id: str) -> dict:
+    """Get task record without tenant filtering (internal use only)."""
+    from sqlor.dbpools import DBPools
+    db, dbname = DBPools(), 'pipeline'
+    async with db.sqlorContext(dbname) as sor:
+        recs = await sor.R('pipeline_tasks', {'id': task_id})
+        if not recs:
            return None
+        rec = recs[0]
+        if hasattr(rec, '__dict__'):
+            return {k: getattr(rec, k) for k in dir(rec) if not k.startswith('_')}
+        return dict(rec)


-async def _execute_step(pipeline_id: str, step_name: str):
-    """Execute a single pipeline step."""
-    update_step_state(pipeline_id, step_name, STATE_RUNNING)
+async def _execute_step(task_id: str, step_name: str, step_graph: dict, task_info: dict):
+    """Execute a single step."""
+    step_info = step_graph[step_name]
+    step_type = step_info["step_type"]
+    version = task_info.get('current_version', task_info.get('current_Version', 1))
+    tenant_id = task_info.get('tenant_id', task_info.get('tenant_Id', ''))
+
+    await update_step_state(task_id, step_name, STATE_RUNNING)

    try:
-        # Load input artifacts from dependencies
-        manifest = get_manifest(pipeline_id)
-        version = manifest["current_version"]
-        input_data = await _gather_step_inputs(pipeline_id, step_name, version, manifest)
+        # Gather inputs from dependency outputs
+        input_data = await _gather_inputs(task_id, version, step_info["deps"])

        # Save input artifact
-        save_artifact(pipeline_id, version, step_name, "input", input_data)
+        await save_artifact(task_id, version, step_name, "input", input_data)

-        # Execute step handler
-        handler = STEP_HANDLERS.get(step_name, _default_handler)
-        output_data = await handler(pipeline_id, step_name, input_data, manifest)
+        # Look up handler by step_type
+        handler = get_handler(step_type)
+        if not handler:
+            handler = get_handler("__default__")
+        if not handler:
+            raise ValueError(f"No handler for step_type '{step_type}' and no default handler")
+
+        # Execute handler
+        output_data = await handler(tenant_id, task_id, step_name, input_data, {})

        # Save output artifact
-        save_artifact(pipeline_id, version, step_name, "output", output_data)
-        update_step_state(pipeline_id, step_name, STATE_COMPLETED)
-
-        logger.info(f"Step {step_name} completed for {pipeline_id}")
+        await save_artifact(task_id, version, step_name, "output", output_data)
+        await update_step_state(task_id, step_name, STATE_COMPLETED)
+        logger.info(f"Step {step_name} completed for task {task_id}")

    except Exception as e:
-        logger.error(f"Step {step_name} failed for {pipeline_id}: {e}")
-        update_step_state(pipeline_id, step_name, STATE_FAILED, str(e))
+        error_msg = str(e)
+        logger.error(f"Step {step_name} failed for task {task_id}: {error_msg}")
+        await update_step_state(task_id, step_name, STATE_FAILED, error_msg)


-async def _gather_step_inputs(pipeline_id: str, step_name: str, version: int, manifest: dict) -> dict:
-    """Gather input data for a step from its dependencies' outputs."""
-    from .storage import get_artifact
-    steps = manifest.get("steps", {})
-    step_info = steps.get(step_name, {})
-    deps = step_info.get("deps", [])
-
+async def _gather_inputs(task_id: str, version: int, deps: list) -> dict:
+    """Gather input data from dependency step outputs."""
    inputs = {}
    for dep in deps:
-        dep_artifact = get_artifact(pipeline_id, version, dep, "output")
-        if dep_artifact:
-            inputs[dep] = dep_artifact.get("data", {})
-
-    # Also include initial params
-    if "_params" in inputs:
-        inputs["params"] = inputs.pop("_params")
-
-    # Check if there's a user-modified input for this step
-    user_input = get_artifact(pipeline_id, version, step_name, "input")
-    if user_input and user_input.get("data"):
-        # User modified input exists, use it
-        return user_input["data"]
-
+        art = await get_artifact(task_id, version, dep, "output")
+        if art:
+            inputs[dep] = art
    return inputs
-
-
-# === Step Handlers ===
-
-async def _default_handler(pipeline_id: str, step_name: str, input_data: dict, manifest: dict) -> dict:
-    """Default handler - stub that returns input as output (for testing)."""
-    logger.info(f"Default handler for {step_name}, pipeline {pipeline_id}")
-    # Simulate some processing time
-    await asyncio.sleep(0.5)
-    return {"step": step_name, "status": "completed", "input_summary": str(input_data)[:200]}
-
-
-# Real handlers will be registered here
-STEP_HANDLERS = {
-    # Will be populated by actual KTV pipeline step implementations
-}
-
-
-def register_handler(step_name: str, handler):
-    """Register a step handler function."""
-    STEP_HANDLERS[step_name] = handler
--- a/pipeline_service/handler.py
+++ b/pipeline_service/handler.py
@ -0,0 +1,52 @@
+"""Pipeline step handler registry.
+
+Handlers are async functions: async def handler(tenant_id, task_id, step_name, input_data, config) -> output_data
+Register via register_handler(step_type, fn).
+Lookup via get_handler(step_type).
+"""
+
+import asyncio
+import logging
+from typing import Callable, Dict, Optional
+
+logger = logging.getLogger("pipeline.handlers")
+
+# step_type -> handler function
+_HANDLERS: Dict[str, Callable] = {}
+
+
+def register_handler(step_type: str, handler: Callable):
+    """Register a step handler function.
+    
+    Args:
+        step_type: matches pipeline_steps.step_type
+        handler: async def handler(tenant_id, task_id, step_name, input_data, config) -> dict
+    """
+    _HANDLERS[step_type] = handler
+    logger.info(f"Registered handler: {step_type}")
+
+
+def get_handler(step_type: str) -> Optional[Callable]:
+    """Get handler for a step type. Returns None if not registered."""
+    return _HANDLERS.get(step_type)
+
+
+def list_handlers() -> Dict[str, str]:
+    """List all registered handlers. Returns {step_type: function_name}."""
+    return {k: v.__name__ for k, v in _HANDLERS.items()}
+
+
+async def default_handler(tenant_id: str, task_id: str, step_name: str, input_data: dict, config: dict) -> dict:
+    """Default handler - passthrough for testing. Returns input as output."""
+    logger.info(f"Default handler: step={step_name}, task={task_id}, tenant={tenant_id}")
+    await asyncio.sleep(0.1)
+    return {
+        "step": step_name,
+        "status": "completed",
+        "input_summary": str(input_data)[:500],
+    }
+
+
+def register_default_handler():
+    """Register the default handler as fallback for '__default__'."""
+    _HANDLERS["__default__"] = default_handler
--- a/pipeline_service/init.py
+++ b/pipeline_service/init.py
@ -0,0 +1,282 @@
+"""pipeline_service - 通用产线执行引擎模块。
+
+把 Hermes Agent 验证过的业务流程固化为可重复、可并发的产线业务环境。
+支持多租户隔离、DAG 步骤调度、可插拔步骤处理器、artifact 版本管理。
+
+任何宿主应用都可以通过 load_pipeline_service() 加载本模块。
+"""
+
+import json
+from ahserver.serverenv import ServerEnv
+from appPublic.uniqueID import getID
+from appPublic.log import debug
+
+from .state import (
+    TASK_SUBMITTED, TASK_RUNNING, TASK_COMPLETED, TASK_FAILED, TASK_PAUSED, TASK_CANCELLED,
+    build_step_graph, get_cascade_rerun_steps, get_rerun_from_next,
+)
+from .storage import (
+    create_task, init_task_steps, get_task, get_task_steps,
+    get_artifact, get_all_artifacts, list_tasks,
+    update_task_state, update_task_version,
+    get_pipeline_steps, reset_steps, save_artifact,
+)
+from .executor import start_task, resume_task, stop_task, is_running
+from .handler import register_handler, list_handlers, register_default_handler
+
+MODULE_NAME = "pipeline_service"
+MODULE_VERSION = "2.0.0"
+
+
+async def pipeline_submit(tenant_id, pipeline_id, owner_id, title, params=None):
+    """提交新产线任务。
+
+    Args:
+        tenant_id: 租户ID（由宿主应用提供，可以是 org_id、user_id 等）
+        pipeline_id: 产线定义ID（来自 pipelines 表）
+        owner_id: 提交人ID
+        title: 任务标题
+        params: 提交参数（dict）
+
+    Returns:
+        JSON string with status, task_id
+    """
+    result = {"success": False}
+    try:
+        if not tenant_id or not pipeline_id:
+            result["message"] = "缺少 tenant_id 或 pipeline_id"
+            return json.dumps(result, ensure_ascii=False)
+
+        params = params or {}
+        task_id = await create_task(tenant_id, pipeline_id, owner_id, title, params)
+
+        # Read step definitions from pipeline_steps table
+        step_records = await get_pipeline_steps(pipeline_id)
+        if not step_records:
+            result["message"] = f"产线 {pipeline_id} 没有步骤定义"
+            return json.dumps(result, ensure_ascii=False)
+
+        # Create step execution records
+        await init_task_steps(task_id, step_records)
+
+        # Start execution
+        await start_task(task_id)
+
+        result["success"] = True
+        result["task_id"] = task_id
+        result["message"] = "任务已提交并开始执行"
+    except Exception as e:
+        result["message"] = str(e)
+    return json.dumps(result, ensure_ascii=False)
+
+
+async def pipeline_list(tenant_id, pipeline_id=None, limit=100):
+    """查询租户的任务列表。"""
+    result = {"success": False}
+    try:
+        tasks = await list_tasks(tenant_id, pipeline_id, limit)
+        result["success"] = True
+        result["tasks"] = tasks
+        result["total"] = len(tasks)
+    except Exception as e:
+        result["message"] = str(e)
+    return json.dumps(result, ensure_ascii=False, default=str)
+
+
+async def pipeline_detail(tenant_id, task_id):
+    """获取任务详情 + 步骤状态树。"""
+    result = {"success": False}
+    try:
+        task = await get_task(tenant_id, task_id)
+        if not task:
+            result["message"] = "任务不存在"
+            return json.dumps(result, ensure_ascii=False)
+
+        steps = await get_task_steps(task_id)
+        task["steps"] = steps
+        task["is_running"] = is_running(task_id)
+
+        result["success"] = True
+        result["task"] = task
+    except Exception as e:
+        result["message"] = str(e)
+    return json.dumps(result, ensure_ascii=False, default=str)
+
+
+async def pipeline_node(tenant_id, task_id, step_name, version=None):
+    """获取某节点某版本的 input/output artifact。"""
+    result = {"success": False}
+    try:
+        task = await get_task(tenant_id, task_id)
+        if not task:
+            result["message"] = "任务不存在"
+            return json.dumps(result, ensure_ascii=False)
+
+        v = version or task.get("current_version", 1)
+        if isinstance(v, str):
+            v = int(v)
+
+        input_data = await get_artifact(task_id, v, step_name, "input")
+        output_data = await get_artifact(task_id, v, step_name, "output")
+
+        result["success"] = True
+        result["step_name"] = step_name
+        result["version"] = v
+        result["input"] = input_data
+        result["output"] = output_data
+    except Exception as e:
+        result["message"] = str(e)
+    return json.dumps(result, ensure_ascii=False, default=str)
+
+
+async def pipeline_modify(tenant_id, task_id, updates, rerun_from="node"):
+    """修改节点 artifact 并触发级联重跑。
+
+    Args:
+        updates: {step_name: {content: {...}}, ...}
+        rerun_from: "node" = 从该节点重跑, "next" = 从下游节点重跑
+    """
+    result = {"success": False}
+    try:
+        task = await get_task(tenant_id, task_id)
+        if not task:
+            result["message"] = "任务不存在"
+            return json.dumps(result, ensure_ascii=False)
+
+        if is_running(task_id):
+            result["message"] = "任务正在执行中，请先暂停"
+            return json.dumps(result, ensure_ascii=False)
+
+        pipeline_id = task.get("pipeline_id", task.get("Pipeline_id", ""))
+        current_version = task.get("current_version", task.get("current_Version", 1))
+        if isinstance(current_version, str):
+            current_version = int(current_version)
+
+        # Load step graph
+        step_records = await get_pipeline_steps(pipeline_id)
+        step_graph = build_step_graph(step_records)
+
+        # Calculate affected steps
+        all_rerun = set()
+        for step_name in updates:
+            if step_name not in step_graph:
+                result["message"] = f"未知步骤: {step_name}"
+                return json.dumps(result, ensure_ascii=False)
+            if rerun_from == "node":
+                affected = get_cascade_rerun_steps(step_graph, step_name)
+            else:
+                affected = get_rerun_from_next(step_graph, step_name)
+            all_rerun.update(affected)
+
+        # Create new version
+        new_version = current_version + 1
+        await update_task_version(task_id, new_version)
+
+        # Save modified artifacts
+        for step_name, step_update in updates.items():
+            content = step_update.get("content", step_update)
+            io_type = "input" if rerun_from == "node" else "output"
+            await save_artifact(task_id, new_version, step_name, io_type, content)
+
+        # Reset affected steps
+        all_rerun_list = sorted(all_rerun, key=lambda s: step_graph.get(s, {}).get("order", 999))
+        await reset_steps(task_id, all_rerun_list)
+
+        # Update task state and resume
+        await update_task_state(task_id, TASK_RUNNING)
+        await resume_task(task_id)
+
+        result["success"] = True
+        result["new_version"] = new_version
+        result["rerun_steps"] = all_rerun_list
+        result["message"] = f"创建 v{new_version}，重跑 {len(all_rerun_list)} 个步骤"
+    except Exception as e:
+        result["message"] = str(e)
+    return json.dumps(result, ensure_ascii=False, default=str)
+
+
+async def pipeline_pause(tenant_id, task_id):
+    """暂停任务。"""
+    result = {"success": False}
+    try:
+        task = await get_task(tenant_id, task_id)
+        if not task:
+            result["message"] = "任务不存在"
+            return json.dumps(result, ensure_ascii=False)
+
+        await stop_task(task_id)
+        await update_task_state(task_id, TASK_PAUSED)
+
+        result["success"] = True
+        result["message"] = "任务已暂停"
+    except Exception as e:
+        result["message"] = str(e)
+    return json.dumps(result, ensure_ascii=False)
+
+
+async def pipeline_resume(tenant_id, task_id):
+    """恢复任务。"""
+    result = {"success": False}
+    try:
+        task = await get_task(tenant_id, task_id)
+        if not task:
+            result["message"] = "任务不存在"
+            return json.dumps(result, ensure_ascii=False)
+
+        await update_task_state(task_id, TASK_RUNNING)
+        await resume_task(task_id)
+
+        result["success"] = True
+        result["message"] = "任务已恢复"
+    except Exception as e:
+        result["message"] = str(e)
+    return json.dumps(result, ensure_ascii=False)
+
+
+async def pipeline_cancel(tenant_id, task_id):
+    """取消任务。"""
+    result = {"success": False}
+    try:
+        task = await get_task(tenant_id, task_id)
+        if not task:
+            result["message"] = "任务不存在"
+            return json.dumps(result, ensure_ascii=False)
+
+        await stop_task(task_id)
+        await update_task_state(task_id, TASK_CANCELLED)
+
+        result["success"] = True
+        result["message"] = "任务已取消"
+    except Exception as e:
+        result["message"] = str(e)
+    return json.dumps(result, ensure_ascii=False)
+
+
+def pipeline_handlers():
+    """查看已注册的步骤处理器。"""
+    return json.dumps(list_handlers(), ensure_ascii=False)
+
+
+def load_pipeline_service():
+    """注册所有函数到 ServerEnv。任何宿主应用调用此函数即可使用产线引擎。"""
+    env = ServerEnv()
+
+    # Task lifecycle
+    env.pipeline_submit = pipeline_submit
+    env.pipeline_list = pipeline_list
+    env.pipeline_detail = pipeline_detail
+    env.pipeline_node = pipeline_node
+    env.pipeline_modify = pipeline_modify
+    env.pipeline_pause = pipeline_pause
+    env.pipeline_resume = pipeline_resume
+    env.pipeline_cancel = pipeline_cancel
+
+    # Handler management
+    env.pipeline_register_handler = register_handler
+    env.pipeline_handlers = pipeline_handlers
+
+    # Register default handler
+    register_default_handler()
+
+    debug(f"[{MODULE_NAME}] v{MODULE_VERSION} loaded — generic pipeline execution engine")
+    return True
--- a/pipeline_service/server.py
+++ b/pipeline_service/server.py
@ -1,361 +0,0 @@
-"""Pipeline Service - aiohttp HTTP API server.
-
-Endpoints:
-  POST /v1/task/submit        - Create and start a new pipeline
-  GET  /v1/tasks              - List pipelines for a user
-  GET  /v1/task/{id}          - Get pipeline detail with node tree
-  GET  /v1/task/{id}/node/{step} - Get node input/output
-  POST /v1/task/update        - Modify node artifact + cascade rerun
-  POST /v1/task/{id}/pause    - Pause a running pipeline
-  POST /v1/task/{id}/resume   - Resume a paused pipeline
-  GET  /v1/health             - Health check
-"""
-
-import json
-import logging
-import os
-from aiohttp import web
-
-from .state import (
-    PIPELINE_SUBMITTED, PIPELINE_RUNNING, PIPELINE_COMPLETED, PIPELINE_FAILED, PIPELINE_PAUSED,
-    get_cascade_rerun_steps, get_rerun_from_next, build_dependency_map,
-)
-from .storage import (
-    create_pipeline, get_manifest, save_manifest,
-    get_artifact, save_artifact, get_all_artifacts,
-    create_new_version, reset_steps, get_user_pipelines,
-)
-from .executor import start_pipeline, resume_pipeline, stop_pipeline, is_running
-
-logger = logging.getLogger("pipeline.api")
-
-JSON_HEADERS = {"Content-Type": "application/json; charset=utf-8"}
-
-
-def json_response(data, status=200):
-    return web.Response(
-        text=json.dumps(data, ensure_ascii=False),
-        status=status,
-        headers=JSON_HEADERS,
-    )
-
-
-def error_response(message, status=400):
-    return json_response({"status": "error", "message": message}, status=status)
-
-
-# === Health ===
-
-async def handle_health(request):
-    return json_response({"status": "ok", "service": "pipeline-service", "version": "1.0.0"})
-
-
-# === Submit New Pipeline ===
-
-async def handle_submit(request):
-    """POST /v1/task/submit
-    Body: {mode, title, lyrics?, input_audio?, input_video?, user_id, ...}
-    """
-    try:
-        body = await request.json()
-    except Exception:
-        return error_response("Invalid JSON body")
-
-    mode = body.get("mode")
-    title = body.get("title")
-    user_id = body.get("user_id")
-
-    if not mode:
-        return error_response("Missing required field: mode")
-    if not title:
-        return error_response("Missing required field: title")
-    if not user_id:
-        return error_response("Missing required field: user_id")
-
-    # Validate mode
-    valid_modes = ["audio_lyrics", "video_lyrics", "lyrics_only"]
-    if mode not in valid_modes:
-        return error_response(f"Invalid mode: {mode}. Must be one of: {valid_modes}")
-
-    # Validate mode-specific requirements
-    if mode == "audio_lyrics" and not body.get("input_audio"):
-        return error_response("Mode audio_lyrics requires input_audio")
-    if mode == "video_lyrics" and not body.get("input_video"):
-        return error_response("Mode video_lyrics requires input_video")
-
-    # Extract params (everything except mode/title/user_id)
-    params = {k: v for k, v in body.items() if k not in ("mode", "title", "user_id")}
-
-    try:
-        manifest = create_pipeline(user_id, mode, title, params)
-    except Exception as e:
-        return error_response(f"Failed to create pipeline: {e}", 500)
-
-    pipeline_id = manifest["pipeline_id"]
-
-    # Start execution in background
-    await start_pipeline(pipeline_id)
-
-    return json_response({
-        "status": "ok",
-        "pipeline_id": pipeline_id,
-        "mode": mode,
-        "title": title,
-        "message": f"Pipeline created and started",
-    })
-
-
-# === List User Pipelines ===
-
-async def handle_list(request):
-    """GET /v1/tasks?user_id=xxx"""
-    user_id = request.query.get("user_id")
-    if not user_id:
-        return error_response("Missing query param: user_id")
-
-    pipeline_ids = get_user_pipelines(user_id)
-    tasks = []
-    for pid in pipeline_ids:
-        manifest = get_manifest(pid)
-        if manifest:
-            tasks.append({
-                "pipeline_id": manifest["pipeline_id"],
-                "mode": manifest["mode"],
-                "title": manifest["title"],
-                "state": manifest["state"],
-                "current_version": manifest["current_version"],
-                "created_at": manifest["created_at"],
-                "updated_at": manifest["updated_at"],
-            })
-
-    return json_response({"status": "ok", "tasks": tasks, "total": len(tasks)})
-
-
-# === Pipeline Detail ===
-
-async def handle_detail(request):
-    """GET /v1/task/{id}"""
-    pipeline_id = request.match_info["id"]
-    manifest = get_manifest(pipeline_id)
-    if not manifest:
-        return error_response(f"Pipeline not found: {pipeline_id}", 404)
-
-    version = manifest["current_version"]
-    artifacts = get_all_artifacts(pipeline_id, version)
-
-    # Build artifact summary
-    artifact_summary = {}
-    for key, art in artifacts.items():
-        artifact_summary[key] = {
-            "step": art.get("step"),
-            "type": art.get("type"),
-            "version": art.get("version"),
-            "saved_at": art.get("saved_at"),
-        }
-
-    return json_response({
-        "status": "ok",
-        "pipeline_id": manifest["pipeline_id"],
-        "user_id": manifest["user_id"],
-        "mode": manifest["mode"],
-        "title": manifest["title"],
-        "state": manifest["state"],
-        "current_version": version,
-        "created_at": manifest["created_at"],
-        "updated_at": manifest["updated_at"],
-        "steps": manifest["steps"],
-        "versions": manifest["versions"],
-        "artifacts": artifact_summary,
-        "is_running": is_running(pipeline_id),
-    })
-
-
-# === Node Input/Output ===
-
-async def handle_node(request):
-    """GET /v1/task/{id}/node/{step}?version=N"""
-    pipeline_id = request.match_info["id"]
-    step = request.match_info["step"]
-    version = request.query.get("version")
-
-    manifest = get_manifest(pipeline_id)
-    if not manifest:
-        return error_response(f"Pipeline not found: {pipeline_id}", 404)
-
-    if step not in manifest["steps"]:
-        return error_response(f"Step not found: {step}")
-
-    v = int(version) if version else manifest["current_version"]
-
-    input_art = get_artifact(pipeline_id, v, step, "input")
-    output_art = get_artifact(pipeline_id, v, step, "output")
-
-    step_info = manifest["steps"][step]
-
-    return json_response({
-        "status": "ok",
-        "pipeline_id": pipeline_id,
-        "step": step,
-        "display_name": step_info.get("display_name", step),
-        "version": v,
-        "state": step_info["state"],
-        "input": input_art["data"] if input_art else None,
-        "output": output_art["data"] if output_art else None,
-        "input_version": input_art.get("version") if input_art else None,
-        "output_version": output_art.get("version") if output_art else None,
-    })
-
-
-# === Modify Node + Cascade Rerun ===
-
-async def handle_update(request):
-    """POST /v1/task/update
-    Body: {pipeline_id, updates: {step: {content: ...}}, rerun_from: "node"|"next"}
-    
-    rerun_from:
-      - "node": input changed, rerun from this step
-      - "next": output changed, rerun from next step(s)
-    """
-    try:
-        body = await request.json()
-    except Exception:
-        return error_response("Invalid JSON body")
-
-    pipeline_id = body.get("pipeline_id")
-    updates = body.get("updates", {})
-    rerun_from = body.get("rerun_from", "node")
-
-    if not pipeline_id:
-        return error_response("Missing pipeline_id")
-    if not updates:
-        return error_response("Missing or empty updates")
-
-    manifest = get_manifest(pipeline_id)
-    if not manifest:
-        return error_response(f"Pipeline not found: {pipeline_id}", 404)
-
-    if manifest["state"] in (PIPELINE_RUNNING,):
-        if is_running(pipeline_id):
-            return error_response("Pipeline is currently running. Pause it first.")
-
-    mode = manifest["mode"]
-
-    # Collect all steps that need rerunning
-    all_rerun = set()
-    for step_name, step_update in updates.items():
-        if step_name not in manifest["steps"]:
-            return error_response(f"Unknown step: {step_name}")
-
-        content = step_update.get("content", {})
-
-        if rerun_from == "node":
-            # Input modified - save as new input, rerun from this step
-            affected = get_cascade_rerun_steps(mode, step_name)
-        else:
-            # Output modified - save as new output, rerun from next steps
-            affected = get_rerun_from_next(mode, step_name)
-
-        all_rerun.update(affected)
-
-    # Create new version
-    change_desc = f"修改步骤: {', '.join(updates.keys())} (rerun_from={rerun_from})"
-    new_version = create_new_version(pipeline_id, change_desc)
-
-    # Save modified artifacts
-    for step_name, step_update in updates.items():
-        content = step_update.get("content", {})
-        io_type = "input" if rerun_from == "node" else "output"
-        save_artifact(pipeline_id, new_version, step_name, io_type, content)
-
-    # Reset affected steps to pending
-    all_rerun_list = sorted(all_rerun, key=lambda s: manifest["steps"].get(s, {}).get("order", 0))
-    reset_steps(pipeline_id, all_rerun_list)
-
-    # Update manifest
-    manifest = get_manifest(pipeline_id)
-    manifest["state"] = PIPELINE_RUNNING
-    save_manifest(pipeline_id, manifest)
-
-    # Resume execution
-    await resume_pipeline(pipeline_id)
-
-    return json_response({
-        "status": "ok",
-        "pipeline_id": pipeline_id,
-        "new_version": new_version,
-        "rerun_steps": all_rerun_list,
-        "rerun_from": rerun_from,
-        "message": f"Created v{new_version}, rerunning {len(all_rerun_list)} steps",
-    })
-
-
-# === Pause/Resume ===
-
-async def handle_pause(request):
-    """POST /v1/task/{id}/pause"""
-    pipeline_id = request.match_info["id"]
-    manifest = get_manifest(pipeline_id)
-    if not manifest:
-        return error_response(f"Pipeline not found: {pipeline_id}", 404)
-
-    await stop_pipeline(pipeline_id)
-    manifest["state"] = PIPELINE_PAUSED
-    save_manifest(pipeline_id, manifest)
-
-    return json_response({"status": "ok", "pipeline_id": pipeline_id, "state": PIPELINE_PAUSED})
-
-
-async def handle_resume(request):
-    """POST /v1/task/{id}/resume"""
-    pipeline_id = request.match_info["id"]
-    manifest = get_manifest(pipeline_id)
-    if not manifest:
-        return error_response(f"Pipeline not found: {pipeline_id}", 404)
-
-    manifest["state"] = PIPELINE_RUNNING
-    save_manifest(pipeline_id, manifest)
-    await resume_pipeline(pipeline_id)
-
-    return json_response({"status": "ok", "pipeline_id": pipeline_id, "state": PIPELINE_RUNNING})
-
-
-# === CORS Middleware ===
-
-@web.middleware
-async def cors_middleware(request, handler):
-    response = await handler(request)
-    response.headers["Access-Control-Allow-Origin"] = "*"
-    response.headers["Access-Control-Allow-Methods"] = "GET, POST, PUT, DELETE, OPTIONS"
-    response.headers["Access-Control-Allow-Headers"] = "Content-Type, Authorization"
-    return response
-
-
-# === App Setup ===
-
-def create_app():
-    app = web.Application(middlewares=[cors_middleware])
-
-    app.router.add_get("/v1/health", handle_health)
-    app.router.add_post("/v1/task/submit", handle_submit)
-    app.router.add_get("/v1/tasks", handle_list)
-    app.router.add_get("/v1/task/{id}", handle_detail)
-    app.router.add_get("/v1/task/{id}/node/{step}", handle_node)
-    app.router.add_post("/v1/task/update", handle_update)
-    app.router.add_post("/v1/task/{id}/pause", handle_pause)
-    app.router.add_post("/v1/task/{id}/resume", handle_resume)
-
-    return app
-
-
-def main():
-    logging.basicConfig(level=logging.INFO, format="%(asctime)s [%(name)s] %(levelname)s: %(message)s")
-    port = int(os.environ.get("PIPELINE_PORT", "8190"))
-    host = os.environ.get("PIPELINE_HOST", "0.0.0.0")
-
-    app = create_app()
-    logger.info(f"Starting Pipeline Service on {host}:{port}")
-    web.run_app(app, host=host, port=port)
-
-
-if __name__ == "__main__":
-    main()
--- a/pipeline_service/state.py
+++ b/pipeline_service/state.py
@ -1,62 +1,7 @@
-"""Pipeline state machine - step definitions, dependency graph, and ordering."""
+"""Pipeline state machine - dynamic DAG resolution from pipeline_steps table."""

-from typing import Dict, List, Optional, Tuple
-
-# Step definitions per mode: (step_name, dependencies, display_name)
-MODE_STEPS: Dict[str, List[Tuple[str, List[str], str]]] = {
-    "audio_lyrics": [  # Mode A
-        ("audio_preparing",            [],                         "音频准备"),
-        ("demucs_separating",          ["audio_preparing"],        "人声分离"),
-        ("lyric_generating",           ["demucs_separating"],      "歌词生成"),
-        ("lyric_evaluating",           ["lyric_generating"],       "歌词评估"),
-        ("music_generating",           ["lyric_evaluating"],       "音乐生成"),
-        ("music_polling",              ["music_generating"],       "音乐轮询"),
-        ("lyric_calibrating",          ["music_polling", "demucs_separating"], "歌词校准"),
-        ("subtitle_rendering",         ["lyric_calibrating"],      "字幕渲染"),
-        ("subtitle_exporting",         ["subtitle_rendering"],     "字幕导出"),
-        ("character_designing",        ["lyric_calibrating"],      "角色设计"),
-        ("character_image_generating", ["character_designing"],    "角色图生成"),
-        ("storyboard_generating",      ["character_designing", "music_polling"], "分镜剧本"),
-        ("scene_video_generating",     ["storyboard_generating", "character_image_generating"], "分镜视频生成"),
-        ("scene_video_evaluating",     ["scene_video_generating"], "分镜视频评估"),
-        ("scene_video_concatenating",  ["scene_video_evaluating"], "分镜视频拼接"),
-        ("ktv_synthesizing",           ["scene_video_concatenating", "subtitle_rendering", "music_polling"], "KTV合成"),
-    ],
-    "video_lyrics": [  # Mode B
-        ("video_preparing",            [],                         "视频准备"),
-        ("demucs_separating",          ["video_preparing"],        "人声分离"),
-        ("lyric_generating",           ["demucs_separating"],      "歌词生成"),
-        ("lyric_evaluating",           ["lyric_generating"],       "歌词评估"),
-        ("music_generating",           ["lyric_evaluating"],       "音乐生成"),
-        ("music_polling",              ["music_generating"],       "音乐轮询"),
-        ("lyric_calibrating",          ["music_polling", "demucs_separating"], "歌词校准"),
-        ("subtitle_rendering",         ["lyric_calibrating"],      "字幕渲染"),
-        ("subtitle_exporting",         ["subtitle_rendering"],     "字幕导出"),
-        ("character_designing",        ["lyric_calibrating"],      "角色设计"),
-        ("character_image_generating", ["character_designing"],    "角色图生成"),
-        ("storyboard_generating",      ["character_designing", "music_polling"], "分镜剧本"),
-        ("scene_video_generating",     ["storyboard_generating", "character_image_generating"], "分镜视频生成"),
-        ("scene_video_evaluating",     ["scene_video_generating"], "分镜视频评估"),
-        ("scene_video_concatenating",  ["scene_video_evaluating"], "分镜视频拼接"),
-        ("ktv_synthesizing",           ["scene_video_concatenating", "subtitle_rendering", "music_polling"], "KTV合成"),
-    ],
-    "lyrics_only": [  # Mode C
-        ("lyric_generating",           [],                         "歌词生成"),
-        ("lyric_evaluating",           ["lyric_generating"],       "歌词评估"),
-        ("music_generating",           ["lyric_evaluating"],       "音乐生成"),
-        ("music_polling",              ["music_generating"],       "音乐轮询"),
-        ("lyric_calibrating",          ["music_polling"],          "歌词校准"),
-        ("subtitle_rendering",         ["lyric_calibrating"],      "字幕渲染"),
-        ("subtitle_exporting",         ["subtitle_rendering"],     "字幕导出"),
-        ("character_designing",        ["lyric_calibrating"],      "角色设计"),
-        ("character_image_generating", ["character_designing"],    "角色图生成"),
-        ("storyboard_generating",      ["character_designing", "music_polling"], "分镜剧本"),
-        ("scene_video_generating",     ["storyboard_generating", "character_image_generating"], "分镜视频生成"),
-        ("scene_video_evaluating",     ["scene_video_generating"], "分镜视频评估"),
-        ("scene_video_concatenating",  ["scene_video_evaluating"], "分镜视频拼接"),
-        ("ktv_synthesizing",           ["scene_video_concatenating", "subtitle_rendering", "music_polling"], "KTV合成"),
-    ],
-}
+import json
+from typing import Dict, List, Optional

 # Step states
 STATE_PENDING = "pending"
@ -65,49 +10,84 @@ STATE_COMPLETED = "completed"
 STATE_FAILED = "failed"
 STATE_SKIPPED = "skipped"

-# Pipeline states
-PIPELINE_SUBMITTED = "submitted"
-PIPELINE_RUNNING = "running"
-PIPELINE_COMPLETED = "completed"
-PIPELINE_FAILED = "failed"
-PIPELINE_PAUSED = "paused"  # waiting for user modification
+# Pipeline task states
+TASK_SUBMITTED = "submitted"
+TASK_RUNNING = "running"
+TASK_COMPLETED = "completed"
+TASK_FAILED = "failed"
+TASK_PAUSED = "paused"
+TASK_CANCELLED = "cancelled"


-def get_step_graph(mode: str) -> List[Tuple[str, List[str], str]]:
-    """Get step definitions for a mode. Returns [(name, deps, display_name), ...]"""
-    if mode not in MODE_STEPS:
-        raise ValueError(f"Unknown mode: {mode}. Available: {list(MODE_STEPS.keys())}")
-    return MODE_STEPS[mode]
+def build_step_graph(step_records: list) -> Dict[str, dict]:
+    """Build dependency map from pipeline_steps table records.
    
+    Args:
+        step_records: list of dicts from pipeline_steps table, each with:
+            - step_name, step_type, display_name, step_order, deps (JSON string)
    
-def build_dependency_map(mode: str) -> Dict[str, dict]:
-    """Build a dependency map for a mode.
-    Returns: {step_name: {"deps": [...], "dependents": [...], "display_name": "...", "order": int}}
+    Returns:
+        {step_name: {"deps": [...], "dependents": [...], "step_type": "...", 
+                      "display_name": "...", "order": int}}
    """
-    steps = get_step_graph(mode)
    dep_map = {}
-    for i, (name, deps, display) in enumerate(steps):
+    for rec in step_records:
+        name = rec['step_name']
+        deps_raw = rec.get('deps', '[]')
+        if isinstance(deps_raw, str):
+            deps = json.loads(deps_raw) if deps_raw else []
+        else:
+            deps = deps_raw if deps_raw else []
+
        dep_map[name] = {
-            "deps": list(deps),
+            "deps": deps,
            "dependents": [],
-            "display_name": display,
-            "order": i + 1,
+            "step_type": rec.get('step_type', name),
+            "display_name": rec.get('display_name', name),
+            "order": rec.get('step_order', 0),
        }
-    # Build reverse mapping
+
+    # Build reverse mapping (dependents)
    for name, info in dep_map.items():
        for dep in info["deps"]:
            if dep in dep_map:
                dep_map[dep]["dependents"].append(name)
+
    return dep_map


-def get_cascade_rerun_steps(mode: str, from_step: str) -> List[str]:
-    """Get all steps that need to be rerun when a step is modified.
-    BFS from the modified step through dependents.
-    Returns ordered list of step names.
+def find_next_step(step_graph: Dict[str, dict], step_states: Dict[str, str]) -> Optional[str]:
+    """Find the next step to execute: pending + all deps completed.
+    
+    Args:
+        step_graph: from build_step_graph()
+        step_states: {step_name: current_state}
+    
+    Returns:
+        step_name or None if all done / blocked
    """
-    dep_map = build_dependency_map(mode)
-    if from_step not in dep_map:
+    candidates = []
+    for name, info in step_graph.items():
+        if step_states.get(name) != STATE_PENDING:
+            continue
+        deps_ok = all(
+            step_states.get(dep) == STATE_COMPLETED
+            for dep in info["deps"]
+        )
+        if deps_ok:
+            candidates.append(name)
+
+    if not candidates:
+        return None
+
+    # Sort by order, return first
+    candidates.sort(key=lambda s: step_graph[s]["order"])
+    return candidates[0]
+
+
+def get_cascade_rerun_steps(step_graph: Dict[str, dict], from_step: str) -> List[str]:
+    """BFS from modified step through dependents. Returns ordered list."""
+    if from_step not in step_graph:
        return []

    visited = set()
@ -120,26 +100,37 @@ def get_cascade_rerun_steps(mode: str, from_step: str) -> List[str]:
            continue
        visited.add(current)
        result.append(current)
-        for dep in dep_map.get(current, {}).get("dependents", []):
+        for dep in step_graph.get(current, {}).get("dependents", []):
            if dep not in visited:
                queue.append(dep)

-    # Sort by order
-    result.sort(key=lambda s: dep_map.get(s, {}).get("order", 999))
+    result.sort(key=lambda s: step_graph.get(s, {}).get("order", 999))
    return result


-def get_rerun_from_next(mode: str, from_step: str) -> List[str]:
-    """When output is modified, rerun from the NEXT steps (dependents only, not the step itself)."""
-    dep_map = build_dependency_map(mode)
-    if from_step not in dep_map:
+def get_rerun_from_next(step_graph: Dict[str, dict], from_step: str) -> List[str]:
+    """When output is modified, rerun from downstream dependents only."""
+    if from_step not in step_graph:
        return []

-    direct_dependents = dep_map[from_step]["dependents"]
+    direct_dependents = step_graph[from_step]["dependents"]
    all_steps = set()
    for d in direct_dependents:
-        all_steps.update(get_cascade_rerun_steps(mode, d))
+        all_steps.update(get_cascade_rerun_steps(step_graph, d))

    result = list(all_steps)
-    result.sort(key=lambda s: dep_map.get(s, {}).get("order", 999))
+    result.sort(key=lambda s: step_graph.get(s, {}).get("order", 999))
    return result
+
+
+def check_all_completed(step_states: Dict[str, str]) -> bool:
+    """Check if all steps are completed or skipped."""
+    for state in step_states.values():
+        if state not in (STATE_COMPLETED, STATE_SKIPPED):
+            return False
+    return True
+
+
+def check_any_failed(step_states: Dict[str, str]) -> bool:
+    """Check if any step has failed."""
+    return any(s == STATE_FAILED for s in step_states.values())
--- a/pipeline_service/storage.py
+++ b/pipeline_service/storage.py
@ -1,258 +1,214 @@
-"""File-based storage for pipeline tasks, artifacts, and versions."""
+"""Pipeline storage layer - MySQL via sqlor."""

 import json
-import os
-import shutil
-import uuid
-from datetime import datetime
-from typing import Any, Dict, List, Optional
+from typing import Dict, List, Optional
+from sqlor.dbpools import DBPools
+from appPublic.uniqueID import getID
+from appPublic.log import debug

-from .state import (
-    STATE_PENDING, STATE_RUNNING, STATE_COMPLETED, STATE_FAILED,
-    PIPELINE_SUBMITTED, PIPELINE_RUNNING,
-    build_dependency_map,
-)
-
-# Base data directory
-DATA_DIR = os.environ.get("PIPELINE_DATA_DIR", os.path.expanduser("~/pipeline_data"))
+DBNAME = "pipeline"


-def _ensure_dir(path: str):
-    os.makedirs(path, exist_ok=True)
+def _get_db():
+    return DBPools(), DBNAME


-def _pipeline_dir(pipeline_id: str) -> str:
-    return os.path.join(DATA_DIR, pipeline_id)
+async def get_pipeline_steps(pipeline_id: str) -> list:
+    """Read step definitions from pipeline_steps table (defined by pipeline_core)."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        recs = await sor.R('pipeline_steps', {'pipeline_id': pipeline_id}, sort='step_order')
+        return list(recs) if recs else []


-def _version_dir(pipeline_id: str, version: int) -> str:
-    return os.path.join(_pipeline_dir(pipeline_id), f"v{version}")
-
-
-def _manifest_path(pipeline_id: str) -> str:
-    return os.path.join(_pipeline_dir(pipeline_id), "manifest.json")
-
-
-def _read_json(path: str) -> dict:
-    with open(path, "r", encoding="utf-8") as f:
-        return json.load(f)
-
-
-def _write_json(path: str, data: dict):
-    _ensure_dir(os.path.dirname(path))
-    with open(path, "w", encoding="utf-8") as f:
-        json.dump(data, f, ensure_ascii=False, indent=2)
-
-
-def generate_pipeline_id() -> str:
-    return f"ktv_{uuid.uuid4().hex[:12]}"
-
-
-def create_pipeline(user_id: str, mode: str, title: str, params: dict) -> dict:
-    """Create a new pipeline task. Returns the manifest."""
-    pipeline_id = generate_pipeline_id()
-    dep_map = build_dependency_map(mode)
-    now = datetime.now().isoformat()
-
-    # Build steps
-    steps = {}
-    for name, info in dep_map.items():
-        steps[name] = {
-            "order": info["order"],
-            "display_name": info["display_name"],
-            "deps": info["deps"],
-            "dependents": info["dependents"],
-            "state": STATE_PENDING,
-            "version": 1,
-            "started_at": None,
-            "completed_at": None,
-            "error": None,
-        }
-
-    manifest = {
+async def create_task(tenant_id: str, pipeline_id: str, owner_id: str, title: str, params: dict) -> str:
+    """Create a new pipeline task. Returns task_id."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        task_id = getID()
+        data = {
+            "id": task_id,
+            "tenant_id": tenant_id,
            "pipeline_id": pipeline_id,
-        "user_id": user_id,
-        "mode": mode,
+            "owner_id": owner_id,
            "title": title,
-        "params": params,
-        "created_at": now,
-        "updated_at": now,
+            "state": "submitted",
            "current_version": 1,
-        "state": PIPELINE_SUBMITTED,
-        "steps": steps,
-        "versions": {
-            "1": {
-                "created_at": now,
-                "changes": "初始版本",
+            "params": json.dumps(params, ensure_ascii=False, default=str),
        }
-        },
+        await sor.C('pipeline_tasks', data)
+        return task_id
+
+
+async def init_task_steps(task_id: str, step_records: list):
+    """Create step execution records from pipeline step definitions."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        for rec in step_records:
+            step_id = getID()
+            data = {
+                "id": step_id,
+                "task_id": task_id,
+                "step_name": rec['step_name'],
+                "step_type": rec.get('step_type', rec['step_name']),
+                "display_name": rec.get('display_name', rec['step_name']),
+                "step_order": rec.get('step_order', 0),
+                "deps": rec.get('deps', '[]') if isinstance(rec.get('deps'), str) else json.dumps(rec.get('deps', [])),
+                "state": "pending",
            }
-
-    # Write to disk
-    pdir = _pipeline_dir(pipeline_id)
-    _ensure_dir(pdir)
-    _ensure_dir(_version_dir(pipeline_id, 1))
-    _write_json(_manifest_path(pipeline_id), manifest)
-
-    # Store initial params as artifact
-    save_artifact(pipeline_id, 1, "_params", "input", params)
-
-    # Index by user
-    _add_to_user_index(user_id, pipeline_id)
-
-    return manifest
+            await sor.C('pipeline_task_steps', data)


-def get_manifest(pipeline_id: str) -> Optional[dict]:
-    """Read pipeline manifest."""
-    path = _manifest_path(pipeline_id)
-    if not os.path.exists(path):
+async def get_task(tenant_id: str, task_id: str) -> Optional[dict]:
+    """Get task record, filtered by tenant."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        recs = await sor.R('pipeline_tasks', {'id': task_id, 'tenant_id': tenant_id})
+        if not recs:
            return None
-    return _read_json(path)
+        rec = recs[0]
+        if hasattr(rec, '__dict__'):
+            return {k: getattr(rec, k) for k in dir(rec) if not k.startswith('_')}
+        return dict(rec)


-def save_manifest(pipeline_id: str, manifest: dict):
-    """Save pipeline manifest."""
-    manifest["updated_at"] = datetime.now().isoformat()
-    _write_json(_manifest_path(pipeline_id), manifest)
-
-
-def save_artifact(pipeline_id: str, version: int, step: str, io_type: str, data: Any):
-    """Save artifact data for a step.
-    io_type: 'input' or 'output'
-    """
-    vdir = _version_dir(pipeline_id, version)
-    _ensure_dir(vdir)
-    path = os.path.join(vdir, f"{step}.{io_type}.json")
-    _write_json(path, {"step": step, "version": version, "type": io_type, "data": data,
-                        "saved_at": datetime.now().isoformat()})
-
-
-def get_artifact(pipeline_id: str, version: int, step: str, io_type: str) -> Optional[dict]:
-    """Read artifact data for a step."""
-    vdir = _version_dir(pipeline_id, version)
-    path = os.path.join(vdir, f"{step}.{io_type}.json")
-    if not os.path.exists(path):
-        # Try previous versions
-        for v in range(version, 0, -1):
-            path = os.path.join(_version_dir(pipeline_id, v), f"{step}.{io_type}.json")
-            if os.path.exists(path):
-                return _read_json(path)
-        return None
-    return _read_json(path)
-
-
-def get_all_artifacts(pipeline_id: str, version: int) -> Dict[str, dict]:
-    """Get all artifacts for a specific version."""
-    vdir = _version_dir(pipeline_id, version)
-    if not os.path.exists(vdir):
-        return {}
-    artifacts = {}
-    for fname in os.listdir(vdir):
-        if fname.endswith(".json"):
-            fpath = os.path.join(vdir, fname)
-            try:
-                data = _read_json(fpath)
-                key = fname.replace(".json", "")
-                artifacts[key] = data
-            except Exception:
-                pass
-    return artifacts
-
-
-def create_new_version(pipeline_id: str, changes: str) -> int:
-    """Create a new version directory. Returns new version number."""
-    manifest = get_manifest(pipeline_id)
-    if not manifest:
-        raise ValueError(f"Pipeline not found: {pipeline_id}")
-
-    new_version = manifest["current_version"] + 1
-    manifest["current_version"] = new_version
-    manifest["versions"][str(new_version)] = {
-        "created_at": datetime.now().isoformat(),
-        "changes": changes,
-    }
-
-    # Copy previous version artifacts to new version (hard links)
-    prev_vdir = _version_dir(pipeline_id, new_version - 1)
-    new_vdir = _version_dir(pipeline_id, new_version)
-    _ensure_dir(new_vdir)
-
-    if os.path.exists(prev_vdir):
-        for fname in os.listdir(prev_vdir):
-            src = os.path.join(prev_vdir, fname)
-            dst = os.path.join(new_vdir, fname)
-            if os.path.isfile(src) and not os.path.exists(dst):
-                try:
-                    os.link(src, dst)  # hard link
-                except OSError:
-                    shutil.copy2(src, dst)
-
-    save_manifest(pipeline_id, manifest)
-    return new_version
-
-
-def reset_steps(pipeline_id: str, step_names: List[str]):
-    """Reset specified steps to pending state."""
-    manifest = get_manifest(pipeline_id)
-    if not manifest:
-        return
-    for name in step_names:
-        if name in manifest["steps"]:
-            manifest["steps"][name]["state"] = STATE_PENDING
-            manifest["steps"][name]["error"] = None
-            manifest["steps"][name]["started_at"] = None
-            manifest["steps"][name]["completed_at"] = None
-    save_manifest(pipeline_id, manifest)
-
-
-def update_step_state(pipeline_id: str, step: str, state: str, error: str = None):
-    """Update a step's state."""
-    manifest = get_manifest(pipeline_id)
-    if not manifest or step not in manifest["steps"]:
-        return
-    now = datetime.now().isoformat()
-    manifest["steps"][step]["state"] = state
-    if state == STATE_RUNNING:
-        manifest["steps"][step]["started_at"] = now
-    elif state in (STATE_COMPLETED, STATE_FAILED):
-        manifest["steps"][step]["completed_at"] = now
-    if error:
-        manifest["steps"][step]["error"] = error
-
-    # Update pipeline state
-    all_states = [s["state"] for s in manifest["steps"].values()]
-    if all(s == STATE_COMPLETED for s in all_states):
-        manifest["state"] = "completed"
-    elif any(s == STATE_FAILED for s in all_states):
-        manifest["state"] = "failed"
-    elif any(s == STATE_RUNNING for s in all_states):
-        manifest["state"] = PIPELINE_RUNNING
-
-    save_manifest(pipeline_id, manifest)
-
-
-# === User Index ===
-
-def _user_index_path(user_id: str) -> str:
-    return os.path.join(DATA_DIR, f"_user_{user_id}.json")
-
-
-def _add_to_user_index(user_id: str, pipeline_id: str):
-    path = _user_index_path(user_id)
-    if os.path.exists(path):
-        data = _read_json(path)
+async def get_task_steps(task_id: str) -> list:
+    """Get all step records for a task."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        recs = await sor.R('pipeline_task_steps', {'task_id': task_id}, sort='step_order')
+        result = []
+        for rec in (recs or []):
+            if hasattr(rec, '__dict__'):
+                result.append({k: getattr(rec, k) for k in dir(rec) if not k.startswith('_')})
            else:
-        data = {"user_id": user_id, "pipelines": []}
-    if pipeline_id not in data["pipelines"]:
-        data["pipelines"].append(pipeline_id)
-    _write_json(path, data)
+                result.append(dict(rec))
+        return result


-def get_user_pipelines(user_id: str) -> List[str]:
-    """Get all pipeline IDs for a user."""
-    path = _user_index_path(user_id)
-    if not os.path.exists(path):
-        return []
-    data = _read_json(path)
-    return data.get("pipelines", [])
+async def get_step_states(task_id: str) -> Dict[str, str]:
+    """Get {step_name: state} for all steps of a task."""
+    steps = await get_task_steps(task_id)
+    return {s['step_name']: s['state'] for s in steps}
+
+
+async def update_task_state(task_id: str, state: str):
+    """Update task state."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        await sor.U('pipeline_tasks', {"id": task_id, "state": state})
+
+
+async def update_task_version(task_id: str, version: int):
+    """Update task current_version."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        await sor.U('pipeline_tasks', {"id": task_id, "current_version": version})
+
+
+async def update_step_state(task_id: str, step_name: str, state: str, error_msg: str = None):
+    """Update step state."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        recs = await sor.R('pipeline_task_steps', {'task_id': task_id, 'step_name': step_name})
+        if not recs:
+            return
+        rec = recs[0]
+        rec_id = rec.id if hasattr(rec, 'id') else rec['id']
+        data = {"id": rec_id, "state": state}
+        if state == "running":
+            data["started_at"] = "NOW()"
+        elif state in ("completed", "failed"):
+            data["completed_at"] = "NOW()"
+        if error_msg:
+            data["error_msg"] = error_msg
+        await sor.U('pipeline_task_steps', data)
+
+
+async def save_artifact(task_id: str, version: int, step_name: str, io_type: str, data: dict):
+    """Save or update an artifact."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        # Check if exists
+        existing = await sor.R('pipeline_artifacts', {
+            'task_id': task_id, 'version': version,
+            'step_name': step_name, 'io_type': io_type
+        })
+        data_json = json.dumps(data, ensure_ascii=False, default=str)
+        if existing:
+            rec = existing[0]
+            rec_id = rec.id if hasattr(rec, 'id') else rec['id']
+            await sor.U('pipeline_artifacts', {"id": rec_id, "data": data_json})
+        else:
+            await sor.C('pipeline_artifacts', {
+                "id": getID(),
+                "task_id": task_id,
+                "version": version,
+                "step_name": step_name,
+                "io_type": io_type,
+                "data": data_json,
+            })
+
+
+async def get_artifact(task_id: str, version: int, step_name: str, io_type: str) -> Optional[dict]:
+    """Get artifact data."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        recs = await sor.R('pipeline_artifacts', {
+            'task_id': task_id, 'version': version,
+            'step_name': step_name, 'io_type': io_type
+        })
+        if not recs:
+            return None
+        rec = recs[0]
+        raw = rec.data if hasattr(rec, 'data') else rec['data']
+        if isinstance(raw, str):
+            return json.loads(raw)
+        return raw
+
+
+async def get_all_artifacts(task_id: str, version: int) -> Dict[str, dict]:
+    """Get all artifacts for a task version. Returns {step_name_io_type: data}."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        recs = await sor.R('pipeline_artifacts', {'task_id': task_id, 'version': version})
+        result = {}
+        for rec in (recs or []):
+            sn = rec.step_name if hasattr(rec, 'step_name') else rec['step_name']
+            io = rec.io_type if hasattr(rec, 'io_type') else rec['io_type']
+            raw = rec.data if hasattr(rec, 'data') else rec['data']
+            key = f"{sn}_{io}"
+            result[key] = json.loads(raw) if isinstance(raw, str) else raw
+        return result
+
+
+async def list_tasks(tenant_id: str, pipeline_id: str = None, limit: int = 100) -> list:
+    """List tasks for a tenant."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        filters = {'tenant_id': tenant_id}
+        if pipeline_id:
+            filters['pipeline_id'] = pipeline_id
+        recs = await sor.R('pipeline_tasks', filters, sort='created_at desc')
+        result = []
+        for rec in (recs or [])[:limit]:
+            if hasattr(rec, '__dict__'):
+                result.append({k: getattr(rec, k) for k in dir(rec) if not k.startswith('_')})
+            else:
+                result.append(dict(rec))
+        return result
+
+
+async def reset_steps(task_id: str, step_names: list):
+    """Reset specified steps to pending state."""
+    db, dbname = _get_db()
+    async with db.sqlorContext(dbname) as sor:
+        for sn in step_names:
+            recs = await sor.R('pipeline_task_steps', {'task_id': task_id, 'step_name': sn})
+            if recs:
+                rec = recs[0]
+                rec_id = rec.id if hasattr(rec, 'id') else rec['id']
+                await sor.U('pipeline_task_steps', {
+                    "id": rec_id, "state": "pending",
+                    "error_msg": None, "started_at": None, "completed_at": None
+                })
--- a/pyproject.toml
+++ b/pyproject.toml
@ -3,12 +3,12 @@ requires = ["setuptools>=45", "wheel"]
 build-backend = "setuptools.build_meta"

 [project]
-name = "pipeline-service"
-version = "1.0.0"
-description = "Hermes Pipeline Backend - manages KTV/MV production pipeline tasks"
+name = "pipeline_service"
+version = "2.0.0"
+description = "通用产线执行引擎模块 — DAG调度、多租户隔离、可插拔步骤处理器、artifact版本管理"
 requires-python = ">=3.8"
 dependencies = [
-    "aiohttp>=3.8",
+    "sqlor",
 ]

 [tool.setuptools.packages.find]
--- a/scripts/load_path.py
+++ b/scripts/load_path.py
@ -0,0 +1,58 @@
+#!/usr/bin/env python3
+"""pipeline_service 模块 RBAC 权限注册。
+
+注意: pipeline_service 是纯后端引擎模块，无 wwwroot 前端文件。
+此脚本仅注册模块 API 可能被引用的路径。
+宿主应用的 load_path.py 负责注册实际的 dspy 路径。
+"""
+
+import os
+import sys
+import subprocess
+
+SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
+
+# Find Sage root
+SAGE_ROOT = None
+for candidate in [
+    os.path.join(SCRIPT_DIR, "..", ".."),
+    os.path.expanduser("~/repos/sage"),
+    os.path.expanduser("~/sage"),
+]:
+    if os.path.isdir(os.path.join(candidate, "wwwroot")) and os.path.isdir(os.path.join(candidate, "py3")):
+        SAGE_ROOT = os.path.abspath(candidate)
+        break
+
+if not SAGE_ROOT:
+    print("ERROR: Cannot find Sage root directory")
+    sys.exit(1)
+
+SET_ROLE_PERM = os.path.join(SAGE_ROOT, "set_role_perm.py")
+PYTHON = os.path.join(SAGE_ROOT, "py3", "bin", "python")
+
+# pipeline_service is a backend-only module, no wwwroot paths needed.
+# The host app's load_path.py handles the dspy paths that call pipeline_submit etc.
+# This script exists for consistency and future extensibility.
+
+PATHS_LOGINED = []
+PATHS_ANY = []
+
+
+def register_paths():
+    for path in PATHS_ANY:
+        subprocess.run([PYTHON, SET_ROLE_PERM, "any", path], cwd=SAGE_ROOT)
+        print(f"  any: {path}")
+
+    for path in PATHS_LOGINED:
+        subprocess.run([PYTHON, SET_ROLE_PERM, "logined", path], cwd=SAGE_ROOT)
+        print(f"  logined: {path}")
+
+
+if __name__ == "__main__":
+    print(f"=== pipeline_service RBAC registration ===")
+    print(f"Sage root: {SAGE_ROOT}")
+    if not PATHS_LOGINED and not PATHS_ANY:
+        print("No paths to register (backend-only module)")
+    else:
+        register_paths()
+    print("Done.")