feat(#60): 一键部署编排与回滚引擎

纯标准库实现，对齐 PRD 5.6「⑥ 部署底座」（不真执行 k8s/helm，模拟编排）： - deploy_plan.py：DeployStep（4 类：pre_check/deploy/health_check/post_check） + DeployPlan（顺序约束自动校验，default_helm_release 默认 4 步计划）+ DeployOrchestrator（按序执行，必需步骤失败即中止并标记 needs_rollback， deploy/health_check 为回滚检查点，pre_check 失败无需回滚）+ HealthCheckContract（对齐 deploy/k8s/healthz 99.8% 目标）。 - rollback.py：RollbackPoint（部署前快照，values_hash SHA1 检测漂移）+ RollbackManager（快照栈 + rollback_to_latest/rollback(n)，失败压回栈顶保持一致性，FIFO 淘汰，支持回滚后健康检查）+ RollbackResult（4 状态）。 - tests：30 用例覆盖计划构造/顺序约束、全成功/各类失败、回滚检查点判定、 action 异常、dry_run、空计划、快照压栈/FIFO、回滚成功/失败/多版本/空栈/ 异常、健康检查契约、values 漂移、报告序列化。 - _sanity_check.py：冒烟验证快照→部署成功→部署失败→回滚全流程。
2026-08-05 05:32:27 +08:00
parent 2afea16005
commit 97fbbfa7b2
8 changed files with 1250 additions and 0 deletions
@@ -0,0 +1,185 @@
+# -*- coding: utf-8 -*-
+"""部署回滚引擎测试（Issue #60）。
+
+覆盖：
+1. 快照压栈 + latest 查询；
+2. FIFO 淘汰（超 max_snapshots）；
+3. rollback_to_latest 成功（弹出栈顶 + 回滚目标版本）；
+4. rollback(n) 多版本回溯；
+5. 空栈回滚（NO_TARGET）；
+6. restore 动作失败（FAILED + 压回栈顶）；
+7. restore 异常（FAILED）；
+8. post_rollback_health 失败（FAILED + 压回）；
+9. n<1 跳过；
+10. values_hash 漂移检测；
+11. 报告序列化。
+"""
+import unittest
+
+import _bootstrap  # noqa: F401  (sys.path 挂载)
+
+from deploy_plan import DeployPlan
+from rollback import (
+    DeployError,
+    RollbackManager,
+    RollbackPoint,
+    RollbackStatus,
+)
+
+
+def _restore(ok: bool, detail: str = ""):
+    def _fn(snapshot, ctx):
+        return ok, (detail or ("回滚成功" if ok else "回滚失败"))
+    return _fn
+
+
+def _raising_restore():
+    def _fn(snapshot, ctx):
+        raise RuntimeError("restore boom")
+    return _fn
+
+
+def _make_plan(version="v1.0.0", release="iaop"):
+    return DeployPlan.default_helm_release(release=release, version=version)
+
+
+class TestSnapshotStack(unittest.TestCase):
+    """快照压栈与查询。"""
+
+    def test_snapshot_push_and_latest(self):
+        mgr = RollbackManager()
+        # snapshot 即压栈（部署前快照），latest 即栈顶
+        mgr.snapshot(_make_plan("v1.0.0"), values={"k": 1}, reason="第一次")
+        self.assertEqual(mgr.latest().version, "v1.0.0")
+        self.assertEqual(len(mgr.snapshots), 1)
+        mgr.snapshot(_make_plan("v1.1.0"), values={"k": 2})
+        self.assertEqual(mgr.latest().version, "v1.1.0")
+        self.assertEqual(len(mgr.snapshots), 2)
+        # 空栈时 latest 返回 None
+        self.assertIsNotNone(mgr.latest())
+        mgr2 = RollbackManager()
+        self.assertIsNone(mgr2.latest())
+
+    def test_fifo_eviction(self):
+        mgr = RollbackManager(max_snapshots=2)
+        mgr.snapshot(_make_plan("v1.0.0"))
+        mgr.snapshot(_make_plan("v1.1.0"))
+        mgr.snapshot(_make_plan("v1.2.0"))
+        # 上限 2，淘汰最旧
+        self.assertEqual(len(mgr.snapshots), 2)
+        self.assertEqual([s.version for s in mgr.snapshots], ["v1.1.0", "v1.2.0"])
+
+    def test_max_snapshots_validation(self):
+        with self.assertRaises(DeployError):
+            RollbackManager(max_snapshots=0)
+
+    def test_values_hash_changes_with_config(self):
+        mgr = RollbackManager()
+        p1 = mgr.snapshot(_make_plan(), values={"backend": "gpu"})
+        p2 = mgr.snapshot(_make_plan(), values={"backend": "npu"})
+        self.assertNotEqual(p1.values_hash, p2.values_hash)
+
+
+class TestRollbackSuccess(unittest.TestCase):
+    """回滚成功场景。"""
+
+    def test_rollback_to_latest_success(self):
+        mgr = RollbackManager(restore_action=_restore(True))
+        mgr.snapshot(_make_plan("v1.0.0"))
+        mgr.snapshot(_make_plan("v1.1.0"))
+        result = mgr.rollback_to_latest(reason="部署 v1.2.0 失败")
+        self.assertTrue(result.succeeded)
+        self.assertEqual(result.status, RollbackStatus.SUCCESS)
+        self.assertEqual(result.target.version, "v1.1.0")
+        # 弹出栈顶后栈剩 1 个
+        self.assertEqual(len(mgr.snapshots), 1)
+        self.assertEqual(mgr.latest().version, "v1.0.0")
+        self.assertIn("v1.2.0 失败", result.reason)
+
+    def test_rollback_n_versions(self):
+        mgr = RollbackManager(restore_action=_restore(True))
+        for v in ["v1.0.0", "v1.1.0", "v1.2.0"]:
+            mgr.snapshot(_make_plan(v))
+        # 回滚 2 个版本 → 目标 v1.1.0（弹出 v1.2.0 和 v1.1.0）
+        result = mgr.rollback(n=2)
+        self.assertTrue(result.succeeded)
+        self.assertEqual(result.target.version, "v1.1.0")
+        self.assertEqual(len(mgr.snapshots), 1)
+
+
+class TestRollbackFailures(unittest.TestCase):
+    """回滚失败与边界。"""
+
+    def test_empty_stack_no_target(self):
+        mgr = RollbackManager()
+        result = mgr.rollback_to_latest()
+        self.assertEqual(result.status, RollbackStatus.NO_TARGET)
+        self.assertFalse(result.succeeded)
+
+    def test_insufficient_stack_for_n(self):
+        mgr = RollbackManager(restore_action=_restore(True))
+        mgr.snapshot(_make_plan("v1.0.0"))
+        # 栈仅 1 个，回滚 2 个 → NO_TARGET
+        result = mgr.rollback(n=2)
+        self.assertEqual(result.status, RollbackStatus.NO_TARGET)
+
+    def test_restore_failure_pushes_target_back(self):
+        mgr = RollbackManager(restore_action=_restore(False, "helm rollback 失败"))
+        mgr.snapshot(_make_plan("v1.0.0"))
+        mgr.snapshot(_make_plan("v1.1.0"))
+        result = mgr.rollback_to_latest()
+        self.assertEqual(result.status, RollbackStatus.FAILED)
+        # 失败时目标压回栈顶，栈仍 2 个
+        self.assertEqual(len(mgr.snapshots), 2)
+        self.assertEqual(mgr.latest().version, "v1.1.0")
+
+    def test_restore_exception_failure(self):
+        mgr = RollbackManager(restore_action=_raising_restore())
+        mgr.snapshot(_make_plan("v1.0.0"))
+        result = mgr.rollback_to_latest()
+        self.assertEqual(result.status, RollbackStatus.FAILED)
+        self.assertIn("异常", result.detail)
+
+    def test_post_rollback_health_failure(self):
+        # restore 成功但健康检查失败（注入失败的健康检查）
+        def _restore_then_health_fail(snapshot, ctx):
+            return True, "回滚动作成功"
+        mgr = RollbackManager(
+            restore_action=_restore_then_health_fail, post_rollback_health=True)
+        # 覆盖 _health_check 为失败
+        mgr._health_check = lambda target: (False, "/health 仍不可用")
+        mgr.snapshot(_make_plan("v1.0.0"))
+        result = mgr.rollback_to_latest()
+        self.assertEqual(result.status, RollbackStatus.FAILED)
+        self.assertIn("健康检查", result.detail)
+        # 压回栈顶
+        self.assertEqual(len(mgr.snapshots), 1)
+
+    def test_n_less_than_one_skipped(self):
+        mgr = RollbackManager()
+        result = mgr.rollback(n=0)
+        self.assertEqual(result.status, RollbackStatus.SKIPPED)
+
+
+class TestResultExport(unittest.TestCase):
+    """结果序列化。"""
+
+    def test_result_to_dict(self):
+        mgr = RollbackManager(restore_action=_restore(True))
+        mgr.snapshot(_make_plan("v1.0.0"))
+        result = mgr.rollback_to_latest()
+        d = result.to_dict()
+        self.assertEqual(d["status"], "success")
+        self.assertTrue(d["succeeded"])
+        self.assertEqual(d["target"]["version"], "v1.0.0")
+
+    def test_point_to_dict(self):
+        p = RollbackPoint(release="r", namespace="ns", version="v1",
+                          chart="c", values_hash="abc123", reason="x")
+        d = p.to_dict()
+        self.assertEqual(d["version"], "v1")
+        self.assertEqual(d["values_hash"], "abc123")
+
+
+if __name__ == "__main__":
+    unittest.main()