fix(performance): 启动回填 ResourceClosedError 修复（双 scalar 调用）+ 补提交拆表测试套件

- _migrate_legacy_executions() 中 existing.scalar() 双调用：Result.scalar() 取完首行即关闭 result，二次调用抛 ResourceClosedError——legacy 迁移一旦完成，此后每次启动回填必炸，导致拆表从表存量回填永远不执行；改为单次取值 legacy_count
- 补提交 test_performance_satellite_tables.py 14 用例（ea2a90cb 提交信息已声明但文件漏入库）
- HANDOFF_性能测试.md 记录拆表部署 5.60 闭环：12 从表建成 + 存量回填 101 行 + JSON 'null' 字面量校验口径 + 部署/验证脚本清单
Co-Authored-By: Claude <noreply@anthropic.com>

Docs/PRD/性能测试/HANDOFF_性能测试.md

@@ -2,13 +2,69 @@
 
 > **生成时间**: 2026-09-08
 > **当前分支**: `platform-auto-test`
-> **最近提交**: `b21f2fe9` fix(performance): 全库扫描 order_by 兜底排查——26 处两段式改造防 MySQL 1038（已推送 origin/platform-auto-test）
-> **会话窗口**: 性能测试 — 2026-09-08 P3 待办闭环：performance 大 JSON/TEXT 列垂直拆表（7 个从表 + 双写 + 读优先从表 + 启动回填 + 零回归）
-> **状态**: ✅ 代码开发与测试完成（3 文件修改 + 1 新增测试，全量 427 passed 零回归），待部署 5.60 与提交推送
+> **最近提交**: `ea2a90cb` perf(performance): 大JSON列垂直拆表闭环 MySQL 1038 + 登录汇总透出（已推送 origin/platform-auto-test）
+> **会话窗口**: 性能测试 — 2026-09-08 P3 待办闭环：performance 大 JSON/TEXT 列垂直拆表（7 个从表 + 双写 + 读优先从表 + 启动回填 + 零回归）+ **部署 5.60 验证闭环**
+> **状态**: ✅ 已提交推送 `ea2a90cb` + 已部署 5.60（12 张从表建成 + 存量回填 101 行 + 修复启动回填 ResourceClosedError，本地全量 437 passed）
 
 ---
 
-## ⚡ 最新会话更新（2026-09-08）— P3 待办闭环：performance 大 JSON 列垂直拆表 ✅
+## ⚡ 最新会话更新（2026-09-08 续）— 拆表部署 5.60 验证闭环 + 启动回填 ResourceClosedError 修复 ✅
+
+### A. 部署状态核验（拆表代码此前未部署）
+
+- 提交 `ea2a90cb` 推送后**未走部署步骤**，远端 5.60 五个关键文件 md5 全部匹配 `b21f2fe9`（CRLF 变体），MySQL 无任何从表；
+- 本次补部署 5 个文件（`database.py` / `models/performance.py` / `services/performance_service.py` / `routers/performance.py` / `schemas/performance.py`），远端 .bak 备份 + 原子替换（.new → mv）+ md5 逐文件复核一致。
+
+### B. 意外发现并修复：启动回填 ResourceClosedError（`_migrate_legacy_executions`）
+
+首次部署重启后从表建成但回填 0 行，启动日志：`存量执行数据回填失败（不影响启动，下次重试）: This result object is closed.`
+
+| 项 | 说明 |
+|----|------|
+| 根因 | `performance_service.py` `_migrate_legacy_executions()` 中 `if existing.scalar() and existing.scalar() > 0:` 对**同一 Result 调用两次 `scalar()`**——SQLAlchemy 中 `scalar()`（内部 `first()`）取完第一行即关闭 result，第二次调用抛 `ResourceClosedError` |
+| 为何早不炸 | 5.60 首次迁移前无 legacy 记录 → `scalar()` 返回 0 falsy → `and` 短路不调用第二次；**迁移一旦完成，之后每次启动必炸**，导致后续拆表回填永远不执行 |
+| 修复 | 改为 `legacy_count = (...).scalar() or 0; if legacy_count > 0:`（单次取值）；全库扫描确认无其他同类双调用 |
+| 验证 | 本地最小复现（双 scalar 抛错）+ 修复后 init_db 正常；5.60 重启日志 `存量回填已执行过，跳过` + `大字段从表存量回填: 101 行` ✅ |
+
+### C. 部署验证结果 ✅
+
+| 项 | 结果 |
+|----|------|
+| 远端 5 文件 md5 | ✅ 全部 = HEAD `ea2a90cb` 工作区（含 scalar 修复） |
+| 容器 | ✅ `docker compose restart app` → Up (healthy)，`/health` 200 第 1 次探测 |
+| 代码标记 | ✅ 容器内 `_backfill_satellite_tables` / `_satellite_upsert` / `_TASK_SATELLITES` 全命中 |
+| 12 张从表建表 | ✅ MySQL 全部创建（create_all） |
+| 存量回填 | ✅ 101 行（任务 13 + 执行 79 + 合并报告 9），12/12 组「主表有效非空行数 = 从表行数」逐一吻合 |
+| API 冒烟 | ✅ `/api/performance/tasks`、`/api/performance/executions` 均 200 |
+| 本地全量测试 | ✅ **437 passed**（含拆表套件 14 用例），零回归 |
+| 启动日志 | ✅ 无 performance 相关 error/traceback |
+
+> 注：回填校验需排除 MySQL JSON 字面量 `'null'`（`JSON_TYPE=NULL`）行——主表部分历史执行的大列存的是 JSON null（无有效数据），从表无行是正确行为，读路径从表缺行回退主表列语义一致。
+
+### D. 剩余待办（更新）
+
+| # | 任务 | 优先级 | 说明 |
+|---|------|--------|------|
+| 1 | **scalar 修复提交推送** | P1 | `performance_service.py` `_migrate_legacy_executions()` 双 scalar 修复（本节 B）+ 本 HANDOFF 更新，待 /GitCommit |
+| 2 | 跟踪其他窗口 progress | — | recorder/projects 为其他窗口工作，合并时注意冲突 |
+
+### E. 本次会话工具与脚本（`backend/tmp/`，gitignored）
+
+| 脚本 | 用途 |
+|------|------|
+| `verify_satellite_deploy_560.py` | 部署核验：远端 5 文件 md5 + 容器健康 + 从表清单 + 启动日志（含双 scalar 修复前发现回填失败的过程） |
+| `deploy_satellite_560.py` | 部署执行：5 文件 .bak 备份 → .new 原子上传 → mv → md5 复核 → `docker compose restart app` → 健康探测 → 容器内 grep 代码标记 |
+| `verify_backfill_560.py` / `verify_backfill_final_560.py` | 回填对比：12 组「主表非空 vs 从表行数」（final 版排除 JSON 字面量 `'null'`，结论全部 OK） |
+| `diff_missing_backfill_560.py` | 定位缺行明细：LEFT JOIN 找主表非空但从表缺行的 execution（确认全部为 `JSON_TYPE=NULL` 的 `'null'` 字面量行） |
+| `check_tables_and_running_560.py` | 从表存在性 + running/pending 执行检查（部署前确认重启无中断风险） |
+
+**远端备份**：5.60 `/data/third_party/plat-auto-test/backend/app/` 下旧文件备份为 `*.bak_20260908_*`（部署脚本自动创建），确认稳定后可清理。
+
+**md5 对比方法论**：Windows 工作区文件为 CRLF，`git show` 输出为 LF，与 Linux 远端直接 md5 比对会全部误报不一致——需按「git blob 内容 × LF/CRLF 变体」逐一匹配远端 md5（本次远端 5 文件全部命中 `b21f2fe9` 的 CRLF 变体，由此断定未部署）。部署脚本上传本地文件原样字节（CRLF），Python 运行不受换行符影响，与历史部署习惯一致。
+
+---
+
+## ⚡ 会话更新（2026-09-08）— P3 待办闭环：performance 大 JSON 列垂直拆表 ✅
 
 ### A. 背景与目标
 
@@ -65,8 +121,8 @@ MySQL 1038（`Out of sort memory`）根因是整行加载大 JSON/TEXT 列消耗
 
 | # | 任务 | 优先级 | 说明 |
 |---|------|--------|------|
-| 1 | 部署 5.60 | P2 | 将 `database.py` / `models/performance.py` / `services/performance_service.py` 部署至 5.60 容器，并重启容器生效回填 |
-| 2 | 代码提交推送 | P2 | /GitCommit 规范提交并推送到 origin/platform-auto-test |
+| 1 | ~~部署 5.60~~ | ~~P2~~ | ✅ 已部署并验证（12 从表 + 101 行回填，见顶部 2026-09-08 续节 C） |
+| 2 | ~~代码提交推送~~ | ~~P2~~ | ✅ 已完成（`ea2a90cb` 已推送；后续 scalar 修复待提交，见顶部待办） |
 
 ### G. 多窗口并行开发注意（提交时必读）
 

backend/app/services/performance_service.py

@@ -1258,12 +1258,14 @@ class PerformanceService:
         Returns:
             int: 回填的执行记录数
         """
-        # 检查是否已有 legacy 回填记录
-        existing = await self.db.execute(
+        # 检查是否已有 legacy 回填记录（注意：Result.scalar() 只能调用一次，二次调用抛 ResourceClosedError）
+        legacy_count = (
+            await self.db.execute(
                 select(func.count()).select_from(PerformanceExecution)
                 .where(PerformanceExecution.triggered_by == "legacy")
             )
-        if existing.scalar() and existing.scalar() > 0:
+        ).scalar() or 0
+        if legacy_count > 0:
             logger.info("存量回填已执行过，跳过")
             return 0
 

backend/tests/test_performance_satellite_tables.py

+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+模块名称：test_performance_satellite_tables.py
+模块描述：性能测试大字段从表（P3 拆表）专项测试
+
+覆盖指标：
+- 从表 upsert 语义（插入/原地更新/空值删除）
+- 任务与执行记录的双写镜像（_sync_task_satellites / _sync_exec_satellites）
+- 读取从表优先、主表列回退（_load_task_satellite_fields / _load_exec_satellite_fields）
+- 启动存量回填幂等（_backfill_satellite_tables）
+- 合并报告 task_summaries 从表优先读（_get_report_task_summaries）
+- create_task / update_task CSV 从表联动
+
+作者：czj
+创建日期：2026-09-08
+"""
+
+import sys
+import uuid
+from pathlib import Path
+
+import pytest
+from sqlalchemy import select, func
+from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
+from sqlalchemy.pool import NullPool
+
+# 保证可导入 backend.app
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from app.database import Base  # noqa: E402
+from app.models.performance import (  # noqa: E402
+    PerformanceTask,
+    PerformanceExecution,
+    PerformanceProjectReport,
+    PerformanceTaskCsv,
+    PerformanceTaskResourceSummary,
+    PerformanceExecutionApiSummary,
+    PerformanceProjectReportSummary,
+)
+from app.services.performance_service import PerformanceService  # noqa: E402
+
+
+# ==================== 夹具 ====================
+
+@pytest.fixture()
+async def db_session(tmp_path):
+    """独立 SQLite 会话（每用例独立库文件，避免跨事件循环复用连接）"""
+    dbfile = tmp_path / f"sat_{uuid.uuid4().hex}.db"
+    engine = create_async_engine(
+        f"sqlite+aiosqlite:///{dbfile.as_posix()}",
+        poolclass=NullPool,
+    )
+    SessionLocal = async_sessionmaker(engine, expire_on_commit=False)
+    async with engine.begin() as conn:
+        await conn.run_sync(Base.metadata.create_all)
+
+    async with SessionLocal() as session:
+        yield session
+
+    await engine.dispose()
+
+
+def _make_task(**overrides) -> PerformanceTask:
+    """构造最小任务 ORM 对象（是否落库由调用方控制）"""
+    defaults = dict(
+        id=f"perf_{uuid.uuid4().hex[:12]}",
+        name="拆表测试任务",
+        target_url="https://192.168.5.44/api/login",
+        method="POST",
+        mode="concurrency",
+        concurrency=5,
+        duration=10,
+        csv_parameterization_enabled=True,
+        csv_content="username,password\nuser1,p1\nuser2,p2",
+    )
+    defaults.update(overrides)
+    return PerformanceTask(**defaults)
+
+
+def _make_execution(task: PerformanceTask, **overrides) -> PerformanceExecution:
+    """构造最小执行记录 ORM 对象（含执行机资源与接口汇总大 JSON）"""
+    defaults = dict(
+        id=f"exec_{uuid.uuid4().hex[:12]}",
+        task_id=task.id,
+        task_name=task.name,
+        status="completed",
+        resource_summary={"cpuSeries": [{"cpuPercent": 30.0}], "avgCpuPercent": 30.0},
+        api_summary={
+            "login": {"totalRequests": 100, "avgResponseTime": 12.5, "errorRate": 0.0}
+        },
+    )
+    defaults.update(overrides)
+    return PerformanceExecution(**defaults)
+
+
+async def _sat_count(db, sat_cls, owner_id: str) -> int:
+    """统计某从表下指定 owner 的行数"""
+    return (
+        await db.execute(
+            select(func.count()).select_from(sat_cls).where(sat_cls.owner_id == owner_id)
+        )
+    ).scalar()
+
+
+# ==================== 从表 upsert 语义 ====================
+
+class TestSatelliteUpsert:
+    """_satellite_upsert：插入 / 原地更新 / 空值删除"""
+
+    async def test_insert_then_update_no_duplicate(self, db_session):
+        svc = PerformanceService(db_session)
+        task = _make_task()
+        db_session.add(task)
+        await db_session.flush()
+
+        await svc._satellite_upsert(
+            PerformanceTaskResourceSummary, task.id, {"v": 1}
+        )
+        await svc._satellite_upsert(
+            PerformanceTaskResourceSummary, task.id, {"v": 2}
+        )
+        await db_session.commit()
+
+        assert await _sat_count(db_session, PerformanceTaskResourceSummary, task.id) == 1
+        data = (
+            await db_session.execute(
+                select(PerformanceTaskResourceSummary.data).where(
+                    PerformanceTaskResourceSummary.owner_id == task.id
+                )
+            )
+        ).scalar_one_or_none()
+        assert data == {"v": 2}
+
+    async def test_empty_value_deletes_row(self, db_session):
+        """None 值同步应删除从表行（避免读到陈旧数据）"""
+        svc = PerformanceService(db_session)
+        task = _make_task()
+        db_session.add(task)
+        await db_session.flush()
+
+        await svc._satellite_upsert(
+            PerformanceTaskResourceSummary, task.id, {"v": 1}
+        )
+        await db_session.flush()
+        assert await _sat_count(db_session, PerformanceTaskResourceSummary, task.id) == 1
+
+        await svc._satellite_upsert(PerformanceTaskResourceSummary, task.id, None)
+        await db_session.commit()
+        assert await _sat_count(db_session, PerformanceTaskResourceSummary, task.id) == 0
+
+    async def test_blank_string_deletes_csv_row(self, db_session):
+        """空字符串 CSV 视为无数据 → 删除从表行"""
+        svc = PerformanceService(db_session)
+        task = _make_task(csv_content="")
+        db_session.add(task)
+        await db_session.flush()
+
+        # 先写入再清空
+        await svc._satellite_upsert(PerformanceTaskCsv, task.id, "a,b\n1,2")
+        await db_session.flush()
+        assert await _sat_count(db_session, PerformanceTaskCsv, task.id) == 1
+
+        await svc._satellite_upsert(PerformanceTaskCsv, task.id, "")
+        await db_session.commit()
+        assert await _sat_count(db_session, PerformanceTaskCsv, task.id) == 0
+
+
+# ==================== 双写镜像 ====================
+
+class TestDualWriteMirror:
+    """_sync_task_satellites / _sync_exec_satellites：主表 → 从表镜像"""
+
+    async def test_task_full_mirror(self, db_session):
+        """全字段镜像：csv + 5 个汇总 JSON 全部落从表"""
+        svc = PerformanceService(db_session)
+        task = _make_task(
+            resource_summary={"cpu": [1, 2]},
+            target_resource_summary={"cpuPercent": 40.0},
+            transaction_summary={"steps": [{"name": "s1", "avg": 5.0}]},
+            api_summary={"login": {"totalRequests": 1}},
+            login_summary={"total": 5, "success": 5, "failed": 0},
+        )
+        db_session.add(task)
+        await db_session.flush()
+        await svc._sync_task_satellites(task)
+        await db_session.commit()
+
+        for sat_cls in (
+            PerformanceTaskCsv,
+            PerformanceTaskResourceSummary,
+        ):
+            assert await _sat_count(db_session, sat_cls, task.id) == 1
+
+        csv_data = (
+            await db_session.execute(
+                select(PerformanceTaskCsv.data).where(
+                    PerformanceTaskCsv.owner_id == task.id
+                )
+            )
+        ).scalar_one_or_none()
+        assert "user1,p1" in csv_data
+
+    async def test_task_partial_mirror_keeps_other_fields(self, db_session):
+        """fields 限定镜像范围时，未指定字段的从表行不受影响"""
+        svc = PerformanceService(db_session)
+        task = _make_task(
+            resource_summary={"cpu": [1]},
+            login_summary={"total": 3, "success": 2, "failed": 1},
+        )
+        db_session.add(task)
+        await db_session.flush()
+        await svc._sync_task_satellites(task)
+        await db_session.commit()
+
+        # 主表清空 resource_summary，仅同步该字段
+        task.resource_summary = None
+        await svc._sync_task_satellites(task, fields=["resource_summary"])
+        await db_session.commit()
+
+        assert await _sat_count(db_session, PerformanceTaskResourceSummary, task.id) == 0
+        # login_summary 从表行未被波及
+        assert await _sat_count(db_session, PerformanceExecutionApiSummary, task.id) == 0
+        from app.models.performance import PerformanceTaskLoginSummary
+        assert await _sat_count(db_session, PerformanceTaskLoginSummary, task.id) == 1
+
+    async def test_exec_mirror(self, db_session):
+        """执行记录镜像：resource_summary / api_summary 落 exec 版从表"""
+        svc = PerformanceService(db_session)
+        task = _make_task()
+        db_session.add(task)
+        await db_session.flush()
+        execution = _make_execution(task)
+        db_session.add(execution)
+        await db_session.flush()
+
+        await svc._sync_exec_satellites(execution)
+        await db_session.commit()
+
+        assert await _sat_count(db_session, PerformanceExecutionApiSummary, execution.id) == 1
+        data = (
+            await db_session.execute(
+                select(PerformanceExecutionApiSummary.data).where(
+                    PerformanceExecutionApiSummary.owner_id == execution.id
+                )
+            )
+        ).scalar_one_or_none()
+        assert data["login"]["totalRequests"] == 100
+
+
+# ==================== 读优先级 ====================
+
+class TestReadPriority:
+    """从表优先，主表列回退"""
+
+    async def test_satellite_wins_over_stale_main_column(self, db_session):
+        """从表有值时以其为准（即使主表列陈旧/不同）"""
+        svc = PerformanceService(db_session)
+        task = _make_task()
+        db_session.add(task)
+        await db_session.flush()
+
+        await svc._satellite_upsert(
+            PerformanceTaskResourceSummary, task.id, {"from": "satellite"}
+        )
+        task.resource_summary = {"from": "main"}
+        await db_session.flush()
+
+        fields = await svc._load_task_satellite_fields(task)
+        assert fields["resource_summary"] == {"from": "satellite"}
+
+    async def test_fallback_to_main_column_when_no_satellite(self, db_session):
+        """从表无行时回退主表列（存量未回填场景）"""
+        svc = PerformanceService(db_session)
+        task = _make_task(
+            csv_content="x,y",
+            login_summary={"total": 1, "success": 1, "failed": 0},
+        )
+        db_session.add(task)
+        await db_session.flush()
+
+        fields = await svc._load_task_satellite_fields(task)
+        assert fields["csv_content"] == "x,y"
+        assert fields["login_summary"] == {"total": 1, "success": 1, "failed": 0}
+        # 主表无值 → None
+        assert fields["resource_summary"] is None
+
+    async def test_exec_fallback_and_priority(self, db_session):
+        svc = PerformanceService(db_session)
+        task = _make_task()
+        db_session.add(task)
+        await db_session.flush()
+        execution = _make_execution(task)
+        db_session.add(execution)
+        await db_session.flush()
+
+        # 无从表行 → 回退主表列
+        fields = await svc._load_exec_satellite_fields(execution)
+        assert fields["resource_summary"] == execution.resource_summary
+
+        # 从表行存在 → 从表优先
+        await svc._satellite_upsert(
+            PerformanceExecutionApiSummary, execution.id, {"sat": True}
+        )
+        fields = await svc._load_exec_satellite_fields(execution)
+        assert fields["api_summary"] == {"sat": True}
+        assert fields["resource_summary"] == execution.resource_summary
+
+
+# ==================== 存量回填幂等 ====================
+
+class TestBackfill:
+    async def test_backfill_creates_rows_once(self, db_session):
+        """存量行回填从表；二次执行不重复写入"""
+        svc = PerformanceService(db_session)
+        task = _make_task(
+            transaction_summary={"steps": [{"name": "t", "avg": 1.0}]},
+        )
+        db_session.add(task)
+        await db_session.flush()
+        execution = _make_execution(task)
+        db_session.add(execution)
+        await db_session.flush()
+        report = PerformanceProjectReport(
+            id=f"rpt_{uuid.uuid4().hex[:12]}",
+            project_id="proj_x",
+            task_ids="[]",
+            task_summaries='[{"task_id": "t1", "avg": 1.0}]',
+        )
+        db_session.add(report)
+        await db_session.flush()
+
+        first = await svc._backfill_satellite_tables()
+        await db_session.commit()
+        assert first >= 3  # task.csv + task.transaction + exec.resource + exec.api + report
+
+        second = await svc._backfill_satellite_tables()
+        await db_session.commit()
+        assert second == 0
+
+        assert await _sat_count(db_session, PerformanceTaskCsv, task.id) == 1
+        assert await _sat_count(db_session, PerformanceProjectReportSummary, report.id) == 1
+
+    async def test_backfill_skips_empty_values(self, db_session):
+        """主表列为空的行不产生从表行"""
+        svc = PerformanceService(db_session)
+        task = _make_task(csv_content="")
+        db_session.add(task)
+        await db_session.flush()
+
+        await svc._backfill_satellite_tables()
+        await db_session.commit()
+        assert await _sat_count(db_session, PerformanceTaskCsv, task.id) == 0
+
+
+# ==================== 合并报告从表读 ====================
+
+class TestReportSummaries:
+    async def test_report_satellite_priority(self, db_session):
+        """task_summaries 从表优先，无行回退主表列"""
+        svc = PerformanceService(db_session)
+        report = PerformanceProjectReport(
+            id=f"rpt_{uuid.uuid4().hex[:12]}",
+            project_id="proj_x",
+            task_ids="[]",
+            task_summaries='[{"from": "main"}]',
+        )
+        db_session.add(report)
+        await db_session.flush()
+
+        # 无从表行 → 回退主表列
+        text = await svc._get_report_task_summaries(report)
+        assert '"from": "main"' in text
+
+        # 从表行存在 → 从表优先
+        await svc._satellite_upsert(
+            PerformanceProjectReportSummary, report.id, '[{"from": "satellite"}]'
+        )
+        text = await svc._get_report_task_summaries(report)
+        assert '"from": "satellite"' in text
+
+        # _report_to_dict 使用 override
+        result = svc._report_to_dict(report, text)
+        assert result["task_summaries"] == [{"from": "satellite"}]
+        # 不传 override 时回退主表列
+        result_main = svc._report_to_dict(report)
+        assert result_main["task_summaries"] == [{"from": "main"}]
+
+
+# ==================== CRUD 联动 ====================
+
+class TestCrudIntegration:
+    async def test_create_and_update_task_csv_satellite(self, db_session):
+        """create_task 写从表；update_task 变更同步；清空删除从表行"""
+        from app.schemas.performance import PerformanceTaskCreate
+
+        svc = PerformanceService(db_session)
+        task = await svc.create_task(
+            PerformanceTaskCreate(
+                name="CRUD联动任务",
+                target_url="https://192.168.5.44/api/login",
+                method="POST",
+                mode="concurrency",
+                concurrency=2,
+                duration=5,
+                csv_parameterization_enabled=True,
+                csv_content="u,p\na,b",
+            )
+        )
+        await db_session.commit()
+        assert await _sat_count(db_session, PerformanceTaskCsv, task.id) == 1
+
+        # 更新 CSV → 从表同步更新
+        await svc.update_task(task.id, {"csv_content": "u,p\nc,d"})
+        await db_session.commit()
+        data = (
+            await db_session.execute(
+                select(PerformanceTaskCsv.data).where(
+                    PerformanceTaskCsv.owner_id == task.id
+                )
+            )
+        ).scalar_one_or_none()
+        assert data == "u,p\nc,d"
+
+        # 清空 CSV → 从表行删除
+        await svc.update_task(task.id, {"csv_content": ""})
+        await db_session.commit()
+        assert await _sat_count(db_session, PerformanceTaskCsv, task.id) == 0
+        # 主表列同步为空（双写一致性）
+        refreshed = await db_session.get(PerformanceTask, task.id)
+        assert not (refreshed.csv_content or "").strip()
+
+    async def test_get_report_mode3_reads_satellite(self, db_session):
+        """get_report 模式3（任务直构）读取从表数据"""
+        svc = PerformanceService(db_session)
+        task = _make_task()
+        db_session.add(task)
+        await db_session.flush()
+        await svc._sync_task_satellites(task)
+        await db_session.commit()
+
+        report = await svc.get_report(task_id=task.id)
+        assert report is not None
+        assert report["resource_summary"] is None  # 未设置
+        assert "user1,p1" not in str(report)  # csv 不在报告体中
+        # 主表列与从表一致时报告能正常取到汇总
+        task.resource_summary = {"cpu": [1]}
+        await svc._sync_task_satellites(task, fields=["resource_summary"])
+        await db_session.commit()
+        report = await svc.get_report(task_id=task.id)
+        assert report["resource_summary"] == {"cpu": [1]}