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 +1,200 @@
+# -*- coding: utf-8 -*-
+"""部署编排引擎测试（Issue #60）。
+
+覆盖：
+1. 默认 Helm 计划构造（4 步：pre_check/deploy/health_check/post_check）；
+2. 步骤顺序约束校验（违例 → DeployError）；
+3. 全成功执行（succeeded + 无需回滚）；
+4. deploy 步骤失败（needs_rollback + failed_step）；
+5. pre_check 失败（needs_rollback=False，非回滚检查点）；
+6. 非必需 post_check 失败（继续编排，succeeded）；
+7. action 异常（视为失败）；
+8. dry_run（全部 SKIPPED）；
+9. 空计划（succeeded=False）；
+10. HealthCheckContract 校验；
+11. 报告序列化。
+"""
+import unittest
+
+import _bootstrap  # noqa: F401  (sys.path 挂载)
+
+from deploy_plan import (
+    DEFAULT_AVAILABILITY_TARGET,
+    DeployError,
+    DeployOrchestrator,
+    DeployOutcome,
+    DeployPlan,
+    DeployStep,
+    DeployStepKind,
+    DeployStatus,
+    HealthCheckContract,
+)
+
+
+def _action(ok: bool, detail: str = ""):
+    """构造固定返回值的 action。"""
+    def _fn(ctx):
+        return ok, (detail or ("成功" if ok else "失败"))
+    return _fn
+
+
+def _raising_action(exc_type=RuntimeError):
+    def _fn(ctx):
+        raise exc_type("boom")
+    return _fn
+
+
+class TestDeployPlanModel(unittest.TestCase):
+    """部署计划模型与顺序约束。"""
+
+    def test_default_helm_release_has_four_steps(self):
+        plan = DeployPlan.default_helm_release()
+        kinds = [s.kind for s in plan.steps]
+        self.assertEqual(kinds, [
+            DeployStepKind.PRE_CHECK,
+            DeployStepKind.DEPLOY,
+            DeployStepKind.HEALTH_CHECK,
+            DeployStepKind.POST_CHECK,
+        ])
+        # post_check 默认非必需
+        self.assertFalse(plan.steps[-1].required)
+
+    def test_step_order_violation_rejected(self):
+        # health_check 排在 deploy 前 → 违例
+        with self.assertRaises(DeployError):
+            DeployPlan(
+                name="bad",
+                steps=[
+                    DeployStep(DeployStepKind.HEALTH_CHECK, "hc"),
+                    DeployStep(DeployStepKind.DEPLOY, "d"),
+                ])
+
+    def test_empty_step_name_rejected(self):
+        with self.assertRaises(DeployError):
+            DeployStep(DeployStepKind.DEPLOY, "")
+
+    def test_plan_name_required(self):
+        with self.assertRaises(DeployError):
+            DeployPlan(name="")
+
+
+class TestOrchestratorSuccess(unittest.TestCase):
+    """全成功编排。"""
+
+    def test_all_steps_succeed(self):
+        plan = DeployPlan.default_helm_release()
+        report = DeployOrchestrator(plan).execute()
+        self.assertTrue(report.succeeded)
+        self.assertFalse(report.outcome.needs_rollback)
+        self.assertEqual(len(report.results), 4)
+        self.assertEqual(
+            [r.status for r in report.results],
+            [DeployStatus.SUCCESS] * 4)
+        # 日志含 START/RUN/OK
+        self.assertTrue(any("[START]" in ln for ln in report.log_lines))
+        self.assertTrue(any("[OK]" in ln for ln in report.log_lines))
+
+    def test_context_populated(self):
+        plan = DeployPlan.default_helm_release()
+        report = DeployOrchestrator(plan).execute()
+        self.assertEqual(report.context["release"], "iaop")
+        self.assertEqual(report.context["namespace"], "iaop")
+        self.assertIn("health", report.context)
+
+
+class TestOrchestratorFailures(unittest.TestCase):
+    """失败场景 + 回滚标记。"""
+
+    def test_deploy_failure_triggers_rollback(self):
+        plan = DeployPlan.default_helm_release(
+            step_actions={DeployStepKind.DEPLOY: _action(False, "helm 失败")})
+        report = DeployOrchestrator(plan).execute()
+        self.assertFalse(report.succeeded)
+        self.assertTrue(report.outcome.needs_rollback)  # deploy 是回滚检查点
+        self.assertIsNotNone(report.outcome.failed_step)
+        # deploy 之后的 health_check/post_check 未执行
+        executed_kinds = [r.step.kind for r in report.results]
+        self.assertNotIn(DeployStepKind.HEALTH_CHECK, executed_kinds)
+        self.assertNotIn(DeployStepKind.POST_CHECK, executed_kinds)
+
+    def test_pre_check_failure_no_rollback(self):
+        plan = DeployPlan.default_helm_release(
+            step_actions={DeployStepKind.PRE_CHECK: _action(False, "集群不可达")})
+        report = DeployOrchestrator(plan).execute()
+        self.assertFalse(report.succeeded)
+        # pre_check 失败：尚未部署，无需回滚
+        self.assertFalse(report.outcome.needs_rollback)
+        self.assertIn("集群不可达", report.outcome.reason)
+
+    def test_non_required_post_check_failure_continues(self):
+        plan = DeployPlan.default_helm_release(
+            step_actions={DeployStepKind.POST_CHECK: _action(False, "烟测失败")})
+        report = DeployOrchestrator(plan).execute()
+        # post_check 非必需，失败仍判成功（核心步骤都过了）
+        self.assertTrue(report.succeeded)
+        self.assertFalse(report.outcome.needs_rollback)
+        # 但有 WARN 日志
+        self.assertTrue(any("[WARN]" in ln for ln in report.log_lines))
+
+    def test_action_exception_treated_as_failure(self):
+        plan = DeployPlan.default_helm_release(
+            step_actions={DeployStepKind.DEPLOY: _raising_action()})
+        report = DeployOrchestrator(plan).execute()
+        self.assertFalse(report.succeeded)
+        self.assertTrue(report.outcome.needs_rollback)
+        self.assertIn("action 异常", report.outcome.reason)
+
+    def test_health_check_failure_triggers_rollback(self):
+        plan = DeployPlan.default_helm_release(
+            step_actions={DeployStepKind.HEALTH_CHECK: _action(False, "/health 不可用")})
+        report = DeployOrchestrator(plan).execute()
+        self.assertFalse(report.succeeded)
+        self.assertTrue(report.outcome.needs_rollback)
+
+
+class TestDryRunAndEmpty(unittest.TestCase):
+    """dry_run + 空计划。"""
+
+    def test_dry_run_skips_all(self):
+        plan = DeployPlan.default_helm_release()
+        report = DeployOrchestrator(plan, dry_run=True).execute()
+        self.assertTrue(report.succeeded)
+        self.assertEqual(
+            [r.status for r in report.results],
+            [DeployStatus.SKIPPED] * 4)
+
+    def test_empty_plan_fails(self):
+        plan = DeployPlan(name="empty", steps=[])
+        report = DeployOrchestrator(plan).execute()
+        self.assertFalse(report.succeeded)
+        self.assertIn("无步骤", report.outcome.reason)
+
+
+class TestHealthContractAndReport(unittest.TestCase):
+    """HealthCheckContract 校验 + 报告序列化。"""
+
+    def test_health_contract_defaults(self):
+        h = HealthCheckContract()
+        self.assertEqual(h.endpoint, "http://iaop:8000/health")
+        self.assertEqual(h.retries, 3)
+        self.assertAlmostEqual(h.availability_target, DEFAULT_AVAILABILITY_TARGET)
+
+    def test_health_contract_validation(self):
+        with self.assertRaises(DeployError):
+            HealthCheckContract(timeout_s=0)
+        with self.assertRaises(DeployError):
+            HealthCheckContract(retries=0)
+        with self.assertRaises(DeployError):
+            HealthCheckContract(availability_target=1.5)
+
+    def test_report_to_dict(self):
+        plan = DeployPlan.default_helm_release()
+        report = DeployOrchestrator(plan).execute()
+        d = report.to_dict()
+        self.assertTrue(d["succeeded"])
+        self.assertEqual(len(d["steps"]), 4)
+        self.assertEqual(d["steps"][0]["kind"], "pre_check")
+
+
+if __name__ == "__main__":
+    unittest.main()