fix(aiops): block k8s playbooks for host repair

apps/api/src/services/auto_repair_service.py

@@ -337,6 +337,23 @@ class AutoRepairService:
                 blocked_by="NOT_APPROVED",
             )
 
+        if self._is_host_or_backup_incident(incident) and self._playbook_has_k8s_steps(best_match.playbook):
+            logger.warning(
+                "auto_repair_blocked_host_backup_k8s_playbook",
+                incident_id=incident.incident_id,
+                playbook_id=best_match.playbook.playbook_id,
+                alert_category=getattr(incident, "alert_category", None),
+            )
+            return AutoRepairDecision(
+                can_auto_repair=False,
+                playbook=best_match.playbook,
+                reason=(
+                    "主機/備份類告警禁止執行 K8s Playbook；"
+                    "需改走 SSH 診斷或緊急介入"
+                ),
+                blocked_by="HOST_BACKUP_K8S_PLAYBOOK",
+            )
+
         # 5. 可以自動修復
         logger.info(
             "auto_repair_approved",
@@ -676,6 +693,29 @@ class AutoRepairService:
 
         return max_risk
 
+    def _is_host_or_backup_incident(self, incident: Incident) -> bool:
+        """主機/備份類事件只能走 SSH/只讀診斷，不允許 K8s rollout 類修復。"""
+
+        category = (getattr(incident, "alert_category", None) or "").lower()
+        if category in {"host_resource", "backup_failure"}:
+            return True
+
+        for signal in incident.signals or []:
+            labels = signal.labels or {}
+            alertname = str(labels.get("alertname") or signal.alert_name or "")
+            if alertname.startswith("HostBackup") or alertname.startswith("Host"):
+                return True
+        return False
+
+    def _playbook_has_k8s_steps(self, playbook: Playbook) -> bool:
+        """檢查 Playbook 是否包含 K8s 指令，避免主機告警誤執行 deployment 操作。"""
+
+        for step in playbook.repair_steps:
+            command = (step.command or "").strip().lower()
+            if step.action_type == ActionType.KUBECTL or command.startswith("kubectl "):
+                return True
+        return False
+
     def _risk_exceeds_threshold(self, risk: RiskLevel) -> bool:
         """檢查風險是否超過自動修復門檻"""
         high_risks = {RiskLevel.HIGH, RiskLevel.CRITICAL}

apps/api/tests/test_auto_repair_service.py

@@ -55,6 +55,8 @@ class MockPlaybookService:
 def create_test_incident(
     incident_id: str = "INC-TEST-001",
     severity: Severity = Severity.P2,
+    alert_category: str | None = None,
+    alert_name: str = "HighCPU",
 ) -> Incident:
     """Create a test incident"""
     now = now_taipei()
@@ -63,13 +65,14 @@ def create_test_incident(
         status=IncidentStatus.INVESTIGATING,
         severity=severity,
         affected_services=["test-service"],
+        alert_category=alert_category,
         signals=[
             Signal(
-                alert_name="HighCPU",
+                alert_name=alert_name,
                 severity=severity,
                 source="prometheus",
                 fired_at=now,
-                labels={"namespace": "prod"},
+                labels={"namespace": "prod", "alertname": alert_name},
             ),
         ],
     )
@@ -274,6 +277,64 @@ class TestAutoRepairService:
         assert decision.playbook.playbook_id == playbook.playbook_id
         assert decision.blocked_by is None
 
+    @pytest.mark.asyncio
+    async def test_backup_failure_blocks_k8s_playbook(self, service, mock_playbook_service):
+        """Backup/host incidents must not execute K8s rollout playbooks."""
+        playbook = create_high_quality_playbook(risk_level=RiskLevel.MEDIUM)
+        mock_playbook_service.add_playbook(playbook)
+        mock_playbook_service.set_recommendations([
+            MockPlaybookRecommendation(playbook, similarity_score=0.85)
+        ])
+
+        incident = create_test_incident(
+            severity=Severity.P2,
+            alert_category="backup_failure",
+            alert_name="HostBackupFailed",
+        )
+        decision = await service.evaluate_auto_repair(incident)
+
+        assert decision.can_auto_repair is False
+        assert decision.blocked_by == "HOST_BACKUP_K8S_PLAYBOOK"
+
+    @pytest.mark.asyncio
+    async def test_backup_failure_allows_ssh_playbook(self, service, mock_playbook_service):
+        """Backup/host incidents may still use SSH playbooks."""
+        playbook = Playbook(
+            playbook_id="PB-BACKUP-SSH",
+            name="Backup SSH diagnostics",
+            description="Read-only backup diagnosis",
+            status=PlaybookStatus.APPROVED,
+            symptom_pattern=SymptomPattern(
+                alert_names=["HostBackupFailed"],
+                affected_services=["test-service"],
+            ),
+            repair_steps=[
+                RepairStep(
+                    step_number=1,
+                    action_type=ActionType.SSH_COMMAND,
+                    command="ssh {host} 'tail -80 /var/log/backup.log'",
+                    risk_level=RiskLevel.LOW,
+                    description="collect backup logs",
+                ),
+            ],
+            success_count=20,
+            failure_count=1,
+        )
+        mock_playbook_service.add_playbook(playbook)
+        mock_playbook_service.set_recommendations([
+            MockPlaybookRecommendation(playbook, similarity_score=0.85)
+        ])
+
+        incident = create_test_incident(
+            severity=Severity.P2,
+            alert_category="backup_failure",
+            alert_name="HostBackupFailed",
+        )
+        decision = await service.evaluate_auto_repair(incident)
+
+        assert decision.can_auto_repair is True
+        assert decision.blocked_by is None
+
     @pytest.mark.asyncio
     async def test_evaluate_low_risk_allowed(self, service, mock_playbook_service):
         """Test that LOW risk actions are allowed"""

docs/LOGBOOK.md

@@ -12,12 +12,14 @@ Live e2e 用 `HostBackupFailed` 打 Alertmanager 後發現 aged backup 告警會
 
 ### 完成
 - `_should_use_alertmanager_rule_first()` / `_should_bypass_alertmanager_llm()` 納入 `backup_failure`，備份失敗 YAML `SSH_DIAGNOSE` 不再被 LLM 覆蓋成 K8s 動作。
+- `AutoRepairService` 追加 host/backup Playbook guard：主機/備份 incident 若匹配到 K8s rollout 類 Playbook，阻擋為 `HOST_BACKUP_K8S_PLAYBOOK`，改走緊急介入。
 - `NodeExporterDown` Prometheus rule `auto_repair` 改為 `true`，與 YAML rule catalog 的 exporter restart 策略一致。
 - 補 `backup_failure` NO_ACTION / SSH_DIAGNOSE 單元測試。
 
 ### 驗證
 - `python3 -m py_compile apps/api/src/api/v1/webhooks.py` 通過。
 - `cd apps/api && pytest tests/test_alertmanager_rule_bypass.py tests/test_telegram_ai_automation_block.py tests/test_ai_router_diagnose_fallback.py -q` → 24 passed。
+- `cd apps/api && pytest tests/test_auto_repair_service.py tests/test_alertmanager_rule_bypass.py -q` → 27 passed。
 - YAML parse `ops/monitoring/alerts-unified.yml`、`apps/api/alert_rules.yaml` 通過。
 
 ## 2026-04-30 | ADR-104 Playbook 版本化 lineage