fix(api): 產生 MCP PlayBook 修復候選

apps/api/tests/test_repair_candidate_service.py

@@ -0,0 +1,202 @@
+from datetime import datetime, timezone
+from uuid import uuid4
+
+import pytest
+
+from src.models.approval import ApprovalRequestCreate, BlastRadius, DataImpact, RiskLevel
+from src.models.incident import Incident, Severity, Signal
+from src.models.playbook import (
+    ActionType,
+    Playbook,
+    PlaybookSource,
+    PlaybookStatus,
+    RepairStep,
+)
+from src.models.playbook import RiskLevel as PlaybookRiskLevel
+from src.services.approval_db import approval_request_to_record_data
+from src.services.evidence_snapshot import EvidenceSnapshot
+from src.services.repair_candidate_service import RepairCandidateService
+
+
+class FakeInvestigator:
+    def __init__(self, evidence: EvidenceSnapshot | None) -> None:
+        self.evidence = evidence
+
+    async def investigate(self, incident: Incident) -> EvidenceSnapshot | None:
+        return self.evidence
+
+
+class FakePlaybookRepository:
+    def __init__(self, playbook: Playbook | None) -> None:
+        self.playbook = playbook
+
+    async def get_by_id(self, playbook_id: str) -> Playbook | None:
+        if self.playbook and self.playbook.playbook_id == playbook_id:
+            return self.playbook
+        return None
+
+
+class FakeIncidentService:
+    async def get_from_working_memory(self, incident_id: str) -> None:
+        return None
+
+
+class FakeAutoRepairService:
+    def preview_write_ssh_mcp_route(self, incident: Incident, command: str) -> bool:
+        return True
+
+
+def _incident() -> Incident:
+    return Incident(
+        incident_id="INC-TEST-REPAIR",
+        severity=Severity.P2,
+        signals=[
+            Signal(
+                alert_name="NodeExporterDown",
+                severity=Severity.P2,
+                source="alertmanager",
+                fired_at=datetime.now(timezone.utc),
+                labels={"namespace": "awoooi-prod", "deployment": "awoooi-api"},
+                annotations={"summary": "node exporter down"},
+            )
+        ],
+        affected_services=["awoooi-api"],
+    )
+
+
+def _evidence(incident_id: str, *, sensors_succeeded: int = 2) -> EvidenceSnapshot:
+    return EvidenceSnapshot(
+        incident_id=incident_id,
+        sensors_attempted=3,
+        sensors_succeeded=sensors_succeeded,
+        mcp_health={"k8s": sensors_succeeded > 0, "prometheus": sensors_succeeded > 1},
+        evidence_summary="k8s events and metrics collected",
+    )
+
+
+def _playbook(command: str, *, risk_level: PlaybookRiskLevel = PlaybookRiskLevel.LOW) -> Playbook:
+    return Playbook(
+        playbook_id="PB-REPAIR-001",
+        name="重啟 API deployment",
+        description="以 approved PlayBook 修復 API 工作負載",
+        status=PlaybookStatus.APPROVED,
+        source=PlaybookSource.MANUAL,
+        trust_score=0.72,
+        estimated_duration_minutes=4,
+        repair_steps=[
+            RepairStep(
+                step_number=1,
+                action_type=ActionType.KUBECTL,
+                command=command,
+                expected_result="deployment restarted and rollout verified",
+                risk_level=risk_level,
+                requires_approval=True,
+            )
+        ],
+    )
+
+
+@pytest.mark.asyncio
+async def test_build_candidate_from_mcp_evidence_and_approved_playbook() -> None:
+    incident = _incident()
+    service = RepairCandidateService(
+        incident_service=FakeIncidentService(),
+        investigator=FakeInvestigator(_evidence(incident.incident_id)),
+        playbook_repository=FakePlaybookRepository(
+            _playbook("kubectl rollout restart deployment/awoooi-api -n awoooi-prod")
+        ),
+        auto_repair_service=FakeAutoRepairService(),
+    )
+
+    result = await service.build_from_incident(
+        incident=incident,
+        alertname="NodeExporterDown",
+        target_resource="awoooi-api",
+        namespace="awoooi-prod",
+        message="node exporter is down",
+        fallback_action="NO_ACTION - REPAIR_CANDIDATE_MISSING",
+        matched_playbook_id="PB-REPAIR-001",
+        severity="medium",
+    )
+
+    assert result.candidate_found is True
+    assert result.approval_request is not None
+    assert result.approval_request.action == "kubectl rollout restart deployment/awoooi-api -n awoooi-prod"
+    assert result.approval_request.risk_level == RiskLevel.MEDIUM
+    assert result.approval_request.matched_playbook_id == "PB-REPAIR-001"
+    assert result.metadata["repair_candidate_status"] == "candidate_ready_for_approval"
+    assert result.metadata["mcp_evidence"]["sensors_succeeded"] == 2
+    assert result.metadata["playbook_trust"]["trust_score"] == 0.72
+    assert result.metadata["verifier_plan"]
+
+
+@pytest.mark.asyncio
+async def test_candidate_blocked_when_mcp_evidence_missing() -> None:
+    incident = _incident()
+    service = RepairCandidateService(
+        incident_service=FakeIncidentService(),
+        investigator=FakeInvestigator(_evidence(incident.incident_id, sensors_succeeded=0)),
+        playbook_repository=FakePlaybookRepository(
+            _playbook("kubectl rollout restart deployment/awoooi-api -n awoooi-prod")
+        ),
+        auto_repair_service=FakeAutoRepairService(),
+    )
+
+    result = await service.build_from_incident(
+        incident=incident,
+        alertname="NodeExporterDown",
+        target_resource="awoooi-api",
+        namespace="awoooi-prod",
+        message="node exporter is down",
+        fallback_action="NO_ACTION - REPAIR_CANDIDATE_MISSING",
+        matched_playbook_id="PB-REPAIR-001",
+        severity="medium",
+    )
+
+    assert result.candidate_found is False
+    assert "mcp_evidence_missing" in result.blockers
+    assert result.metadata["repair_candidate_status"] == "blocked"
+
+
+@pytest.mark.asyncio
+async def test_candidate_blocked_when_playbook_is_observe_only() -> None:
+    incident = _incident()
+    service = RepairCandidateService(
+        incident_service=FakeIncidentService(),
+        investigator=FakeInvestigator(_evidence(incident.incident_id)),
+        playbook_repository=FakePlaybookRepository(
+            _playbook("kubectl get pods -n awoooi-prod")
+        ),
+        auto_repair_service=FakeAutoRepairService(),
+    )
+
+    result = await service.build_from_incident(
+        incident=incident,
+        alertname="NodeExporterDown",
+        target_resource="awoooi-api",
+        namespace="awoooi-prod",
+        message="node exporter is down",
+        fallback_action="NO_ACTION - REPAIR_CANDIDATE_MISSING",
+        matched_playbook_id="PB-REPAIR-001",
+        severity="medium",
+    )
+
+    assert result.candidate_found is False
+    assert "playbook_observe_only" in result.blockers
+
+
+def test_approval_record_data_uses_preallocated_id_without_leaking_metadata() -> None:
+    approval_id = str(uuid4())
+    request = ApprovalRequestCreate(
+        action="kubectl rollout restart deployment/awoooi-api -n awoooi-prod",
+        description="candidate",
+        risk_level=RiskLevel.MEDIUM,
+        requested_by="repair-candidate-test",
+        blast_radius=BlastRadius(data_impact=DataImpact.WRITE),
+        metadata={"preallocated_approval_id": approval_id, "source": "test"},
+    )
+
+    data = approval_request_to_record_data(request, RiskLevel.MEDIUM, required_sigs=1)
+
+    assert data["id"] == approval_id
+    assert data["extra_metadata"] == {"source": "test"}