muse-agent-example/tests/集成/test_维护任务恢复.py

"""维护流程使用真实S02状态与owner回执，关闭会话不会取消已登记工作。"""

from dataclasses import replace
from datetime import timedelta
from uuid import uuid4

import pytest
import test_原文归档与恢复 as 原文测试

from muse.任务运行.接口 import 任务状态
from muse.共享.调用身份 import 用途
from muse.共享.错误 import 配置错误
from muse.启动 import 构建
from muse.基础设施.备份恢复 import 列出备份, 创建备份, 备份恢复错误
from muse.编排.数据维护 import 发起维护任务, 推进维护任务
from muse.资源加载 import 加载清单
from muse.配置 import 应用配置

pytestmark = pytest.mark.数据库
原文环境 = 原文测试.原文环境


@pytest.mark.case_id(
    "NC-w29-29c001",
    environment="隔离PG",
    given="显式隔离配置、合成数据与已知发布构建",
    when="经真实维护、业务或浏览器入口执行并读取持久结果",
    then=["备份核对计划跨装配继续且冻结目录与命令范围"],
    contract="docs/系统架构/新版设计/迁移与验收.md#备份与恢复入口",
)
def test_备份核对计划跨装配继续且冻结目录与命令范围__29c001(应用测试库, tmp_path):
    root = tmp_path / "backups"
    root.mkdir(mode=0o700)
    manifest = 创建备份(
        应用配置(应用测试库[用途.维护].引用, 加载清单()["构建身份"], 运行用途=用途.维护),
        root / "one",
    )
    row = 列出备份(str(root))["items"][0]
    config = 应用配置(应用测试库[用途.生产].引用, 加载清单()["构建身份"], 备份目录=str(root))
    app = 构建(config)
    with app.生命周期():
        request = {
            "action": "verify_backup",
            "entry_id": row["entry_id"],
            "backup_id": manifest["backup_id"],
        }
        created = 发起维护任务(app, "author", "check-one", request)
        task_id = created["task_id"]
        assert 发起维护任务(app, "author", "check-one", request) == created
        with pytest.raises(配置错误, match="重放"):
            发起维护任务(app, "author", "check-one", {**request, "backup_id": "another"})
        claim = app.任务运行.领取步骤(
            "before-restart", ["maintenance.verify"], 任务ID=task_id, 作者="author"
        )
        assert claim is not None
        app.任务运行.执行一步(claim)
        checkpoint = app.任务运行.读取任务(task_id).步骤[0]["checkpoint"]
        assert checkpoint["scope_hash"]
        # 每个装配各有一条常驻连接池，跨装配继续必须各自进入生命周期。
        wrong = 构建(replace(config, 备份目录=str(tmp_path / "other")))
        with wrong.生命周期(), pytest.raises(配置错误, match="范围已变化"):
            推进维护任务(wrong, "author", task_id)
        assert app.任务运行.读取任务(task_id).状态 == 任务状态.已失败
        fresh = 构建(config)
        with fresh.生命周期():
            fresh.任务运行.控制任务(task_id, "author", 任务状态.已失败, "恢复", 命令ID="resume")
            with pytest.raises(配置错误, match="当前作者"):
                推进维护任务(fresh, "another", task_id)
            finished = 推进维护任务(fresh, "author", task_id)
        assert finished["state"] == "completed"
        assert finished["steps"][0]["checkpoint"] == checkpoint
        result = finished["steps"][-1]["result"]
        assert result["backup_id"] == manifest["backup_id"]
        assert result["archive_verified"] and not result["restore_verified"]


@pytest.mark.case_id(
    "NC-w29-29c002",
    environment="隔离PG",
    given="显式隔离配置、合成数据与已知发布构建",
    when="经真实维护、业务或浏览器入口执行并读取持久结果",
    then=["归档损坏不伪称通过且修复后原任务可恢复"],
    contract="docs/系统架构/新版设计/迁移与验收.md#备份与恢复入口",
)
def test_归档损坏不伪称通过且修复后原任务可恢复__29c002(应用测试库, tmp_path):
    root = tmp_path / "backups"
    root.mkdir(mode=0o700)
    manifest = 创建备份(
        应用配置(应用测试库[用途.维护].引用, 加载清单()["构建身份"], 运行用途=用途.维护),
        root / "one",
    )
    row = 列出备份(str(root))["items"][0]
    app = 构建(应用配置(应用测试库[用途.生产].引用, 加载清单()["构建身份"], 备份目录=str(root)))
    with app.生命周期():
        task_id = 发起维护任务(
            app,
            "author",
            "check-bad",
            {
                "action": "verify_backup",
                "entry_id": row["entry_id"],
                "backup_id": manifest["backup_id"],
            },
        )["task_id"]
        dump = root / "one/database.dump"
        original = dump.read_bytes()
        dump.write_bytes(original + b"bad")
        with pytest.raises(备份恢复错误):
            推进维护任务(app, "author", task_id)
        failed = app.任务运行.读取任务(task_id)
        assert failed.状态 == 任务状态.已失败
        assert not failed.步骤[-1]["result"]
        dump.write_bytes(original)
        app.任务运行.控制任务(task_id, "author", 任务状态.已失败, "恢复", 命令ID="repaired")
        assert 推进维护任务(app, "author", task_id)["state"] == "completed"


@pytest.mark.case_id(
    "NC-w29-29c003",
    environment="隔离PG",
    given="显式隔离配置、合成数据与已知发布构建",
    when="经真实维护、业务或浏览器入口执行并读取持久结果",
    then=["原文维护保留有效租约并保留孤儿清理真实回执"],
    contract="docs/系统架构/新版设计/迁移与验收.md#备份与恢复入口",
)
def test_原文维护保留有效租约并保留孤儿清理真实回执__29c003(原文环境):
    raw, original_task, authorization, data, digest, clock, pools = 原文环境
    lease = raw.创建租约(
        original_task, authorization, "keep", clock.现在() + timedelta(hours=1), 最少剩余秒=0
    )
    raw.写入(original_task, lease, digest, data)
    orphan = str(uuid4())
    raw.文件.写入(orphan, digest, data)
    app = 构建(
        应用配置(pools[用途.生产].引用, 加载清单()["构建身份"], 原文暂存=str(raw.文件.根)),
        时钟=clock,
    )
    with app.生命周期():
        recover = 发起维护任务(
            app,
            "author",
            "raw-recover",
            {
                "action": "recover_raw",
                "target_task_id": original_task,
            },
        )["task_id"]
        done = 推进维护任务(app, "author", recover)
        assert done["steps"][-1]["result"]["items"] == [{"lease_id": lease, "action": "retained"}]
        with pytest.raises(配置错误, match="不属于"):
            发起维护任务(
                app,
                "other",
                "wrong-owner",
                {
                    "action": "recover_raw",
                    "target_task_id": original_task,
                },
            )
        clean = 发起维护任务(app, "author", "orphan-clean", {"action": "clean_orphans"})["task_id"]
        # 无租约只证明缺登记，不证明可删：维护入口盘点并列出待人工核对对象。
        with pytest.raises(配置错误, match="部分原文尚待恢复"):
            推进维护任务(app, "author", clean)
        checkpoint = app.任务运行.读取任务(clean).步骤[-1]["checkpoint"]
        assert checkpoint["preview_only"] is True and checkpoint["needs_recovery"] == [orphan]
        assert checkpoint["cleaned_orphan_ids"] == []
        assert (raw.文件.根 / orphan).exists()
        assert raw.读取(original_task, lease, digest) == data