awoooi/apps/api/tests/test_pre_decision_investigator.py

"""
PreDecisionInvestigator 測試
============================
ADR-081: Phase 1 決策前情報調查員

測試項目：
- 工具並行蒐集（多維度）
- 工具部分失敗不阻塞（Graceful Degradation）
- 工具逾時被丟棄
- EvidenceSnapshot 正確填入感官維度
- evidence_summary 組裝 + Token Budget 截斷
- fingerprint 計算一致性
- _fill_snapshot_dimension 正確路由

注意：不依賴真實 Redis / DB — 純邏輯測試

2026-04-15 Claude Sonnet 4.6 + ogt: Phase 1 初始建立
"""

from __future__ import annotations

import asyncio
import pytest

from src.plugins.mcp.interfaces import MCPTool, MCPToolProvider, MCPToolResult
from src.services.evidence_snapshot import EvidenceSnapshot
from src.services.mcp_tool_registry import (
    MCPToolRegistry,
    RegisteredTool,
    SensorDimension,
)
from src.services.pre_decision_investigator import (
    PreDecisionInvestigator,
    _compute_fingerprint,
    _fill_snapshot_dimension,
    _build_tool_params,
)


# ─────────────────────────────────────────────────────────────────────────────
# Stubs
# ─────────────────────────────────────────────────────────────────────────────

def _make_tool(name: str) -> MCPTool:
    return MCPTool(name=name, description="", input_schema={}, server_name="test")


class _SuccessProvider(MCPToolProvider):
    """永遠成功，回傳固定 output"""

    def __init__(self, output: dict | str | None = None) -> None:
        self._output = output if output is not None else {"status": "Running"}

    @property
    def name(self) -> str:
        return "success_provider"

    async def list_tools(self) -> list[MCPTool]:
        return []

    async def execute(self, tool_name: str, parameters: dict) -> MCPToolResult:
        return MCPToolResult(success=True, execution_id="ok", output=self._output)


class _FailProvider(MCPToolProvider):
    """永遠失敗"""

    @property
    def name(self) -> str:
        return "fail_provider"

    async def list_tools(self) -> list[MCPTool]:
        return []

    async def execute(self, tool_name: str, parameters: dict) -> MCPToolResult:
        return MCPToolResult(success=False, execution_id="fail", error="connection refused")


class _TimeoutProvider(MCPToolProvider):
    """永遠逾時"""

    @property
    def name(self) -> str:
        return "timeout_provider"

    async def list_tools(self) -> list[MCPTool]:
        return []

    async def execute(self, tool_name: str, parameters: dict) -> MCPToolResult:
        await asyncio.sleep(9999)
        return MCPToolResult(success=True, execution_id="never", output={})


def _stub_incident(
    alertname: str = "KubePodCrashLooping",
    namespace: str = "awoooi-prod",
    pod: str = "api-xyz",
    severity: str = "critical",
) -> object:
    """返回最小 Incident stub（僅需 .signals[0].labels）"""
    class _Signal:
        labels = {
            "alertname": alertname,
            "namespace": namespace,
            "pod": pod,
            "severity": severity,
        }

    class _Incident:
        incident_id = f"INC-{alertname[:4]}"
        signals = [_Signal()]

    return _Incident()


def _reg(tool_name: str, provider: MCPToolProvider, dim: SensorDimension) -> RegisteredTool:
    return RegisteredTool(
        tool=_make_tool(tool_name),
        provider=provider,
        dimensions=[dim],
        priority=5,
    )


# ─────────────────────────────────────────────────────────────────────────────
# _compute_fingerprint
# ─────────────────────────────────────────────────────────────────────────────

class TestComputeFingerprint:
    def test_same_labels_same_fingerprint(self):
        i1 = _stub_incident("Kube", "prod", "pod1", "critical")
        i2 = _stub_incident("Kube", "prod", "pod1", "critical")
        assert _compute_fingerprint(i1) == _compute_fingerprint(i2)

    def test_different_alertname_different_fingerprint(self):
        i1 = _stub_incident("Kube", "prod", "pod1", "critical")
        i2 = _stub_incident("Host", "prod", "pod1", "critical")
        assert _compute_fingerprint(i1) != _compute_fingerprint(i2)

    def test_fingerprint_length_16(self):
        i = _stub_incident()
        fp = _compute_fingerprint(i)
        assert len(fp) == 16

    def test_fingerprint_hex_chars_only(self):
        i = _stub_incident()
        fp = _compute_fingerprint(i)
        assert all(c in "0123456789abcdef" for c in fp)


# ─────────────────────────────────────────────────────────────────────────────
# _build_tool_params
# ─────────────────────────────────────────────────────────────────────────────

class TestBuildToolParams:
    def test_namespace_extracted(self):
        p = _build_tool_params(_stub_incident(namespace="mynamespace"))
        assert p["namespace"] == "mynamespace"

    def test_pod_name_extracted(self):
        p = _build_tool_params(_stub_incident(pod="mypod"))
        assert p["pod_name"] == "mypod"

    def test_alertname_extracted(self):
        p = _build_tool_params(_stub_incident(alertname="MyAlert"))
        assert p["alertname"] == "MyAlert"

    def test_default_namespace_fallback(self):
        class _Signal:
            labels = {}
        class _Inc:
            incident_id = "x"
            signals = [_Signal()]
        p = _build_tool_params(_Inc())
        assert p["namespace"] == "awoooi-prod"


# ─────────────────────────────────────────────────────────────────────────────
# _fill_snapshot_dimension
# ─────────────────────────────────────────────────────────────────────────────

class TestFillSnapshotDimension:
    def _reg(self, dim: SensorDimension) -> RegisteredTool:
        return RegisteredTool(
            tool=_make_tool("t"),
            provider=_SuccessProvider(),
            dimensions=[dim],
            priority=5,
        )

    def test_d1_dict_fills_k8s_state(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D1_K8S_STATE), {"status": "Running"})
        assert snap.k8s_state == {"status": "Running"}

    def test_d1_str_wraps_in_raw(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D1_K8S_STATE), "raw k8s output")
        assert snap.k8s_state == {"raw": "raw k8s output"}

    def test_d2_fills_recent_logs(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D2_LOGS), "log line 1\nlog line 2")
        assert "log line 1" in snap.recent_logs

    def test_d2_dict_serialized_to_string(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D2_LOGS), {"msg": "hello"})
        assert "hello" in snap.recent_logs

    def test_d3_fills_metrics(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D3_METRICS), {"cpu": 95.2})
        assert snap.metrics_snapshot == {"cpu": 95.2}

    def test_d4_list_fills_deployments(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D4_CHANGES), [{"rev": "abc"}])
        assert snap.recent_deployments == [{"rev": "abc"}]

    def test_d4_dict_wrapped_in_list(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D4_CHANGES), {"rev": "abc"})
        assert snap.recent_deployments == [{"rev": "abc"}]

    def test_d5_fills_business_metrics(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D5_BUSINESS), {"sli": 0.99})
        assert snap.business_metrics == {"sli": 0.99}

    def test_d6_truncated_at_2000(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D6_HISTORY), "X" * 5000)
        assert len(snap.historical_context) <= 2100  # 2000 + possible truncation note

    def test_d7_fills_peer_health(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D7_PEERS), {"replica_0": "ok"})
        assert snap.peer_health == {"replica_0": "ok"}

    def test_d8_fills_topology(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D8_TOPOLOGY), {"upstream": "db"})
        assert snap.dependency_topology == {"upstream": "db"}

    def test_none_raw_is_noop(self):
        snap = EvidenceSnapshot(incident_id="x")
        _fill_snapshot_dimension(snap, self._reg(SensorDimension.D1_K8S_STATE), None)
        assert snap.k8s_state is None  # 未被修改


# ─────────────────────────────────────────────────────────────────────────────
# PreDecisionInvestigator._collect_one
# ─────────────────────────────────────────────────────────────────────────────

class TestCollectOne:
    """單工具蒐集行為（不需要 DB / Redis）"""

    @pytest.mark.asyncio
    async def test_success_fills_snapshot(self):
        investigator = PreDecisionInvestigator()
        snap = EvidenceSnapshot(incident_id="x")
        reg = _reg("kubectl_describe", _SuccessProvider({"status": "Running"}), SensorDimension.D1_K8S_STATE)

        await investigator._collect_one(snap, reg, {"namespace": "prod"})

        assert snap.mcp_health["kubectl_describe"] is True
        assert snap.sensors_succeeded == 1
        assert snap.k8s_state is not None

    @pytest.mark.asyncio
    async def test_failed_tool_marks_health_false(self):
        investigator = PreDecisionInvestigator()
        snap = EvidenceSnapshot(incident_id="x")
        reg = _reg("kubectl_logs", _FailProvider(), SensorDimension.D2_LOGS)

        await investigator._collect_one(snap, reg, {})

        assert snap.mcp_health["kubectl_logs"] is False
        assert snap.sensors_succeeded == 0

    @pytest.mark.asyncio
    async def test_timeout_marks_health_false(self):
        """工具逾時必須被丟棄，不阻塞主路徑"""
        investigator = PreDecisionInvestigator()
        snap = EvidenceSnapshot(incident_id="x")
        reg = _reg("slow_tool", _TimeoutProvider(), SensorDimension.D3_METRICS)

        # _collect_one 本身有 MCP_TOOL_TIMEOUT_SEC=5 的 wait_for 保護
        # 但在測試中我們直接驗證它不會 raise，只是設 health=False
        # 用一個足夠短的超時替代（patch不做，因為這是純邏輯驗證）
        with pytest.raises(Exception):
            # TimeoutProvider 會永久阻塞，直接觸發 asyncio.TimeoutError
            await asyncio.wait_for(
                investigator._collect_one(snap, reg, {}),
                timeout=0.1,
            )
        # 超時後 health 預設 False（在 _collect_one 開頭設定）
        assert snap.mcp_health.get("slow_tool") is False


# ─────────────────────────────────────────────────────────────────────────────
# PreDecisionInvestigator._collect_all
# ─────────────────────────────────────────────────────────────────────────────

class TestCollectAll:
    """多工具並行蒐集 — Graceful Degradation"""

    @pytest.mark.asyncio
    async def test_partial_failure_does_not_block(self):
        """失敗工具不阻塞成功工具"""
        investigator = PreDecisionInvestigator()
        snap = EvidenceSnapshot(incident_id="x")
        snap.sensors_attempted = 2

        tools = [
            _reg("kubectl_describe", _SuccessProvider({"status": "Running"}), SensorDimension.D1_K8S_STATE),
            _reg("kubectl_logs", _FailProvider(), SensorDimension.D2_LOGS),
        ]
        incident = _stub_incident()

        await investigator._collect_all(snap, tools, incident)

        assert snap.mcp_health["kubectl_describe"] is True
        assert snap.mcp_health["kubectl_logs"] is False
        assert snap.sensors_succeeded == 1
        assert snap.k8s_state is not None

    @pytest.mark.asyncio
    async def test_all_success_fills_multiple_dimensions(self):
        investigator = PreDecisionInvestigator()
        snap = EvidenceSnapshot(incident_id="x")

        tools = [
            _reg("kubectl_describe", _SuccessProvider({"status": "Running"}), SensorDimension.D1_K8S_STATE),
            _reg("prometheus_query", _SuccessProvider({"cpu": 95.0}), SensorDimension.D3_METRICS),
        ]
        incident = _stub_incident()

        await investigator._collect_all(snap, tools, incident)

        assert snap.k8s_state is not None
        assert snap.metrics_snapshot is not None
        assert snap.sensors_succeeded == 2

    @pytest.mark.asyncio
    async def test_empty_tools_produces_empty_snapshot(self):
        investigator = PreDecisionInvestigator()
        snap = EvidenceSnapshot(incident_id="x")

        await investigator._collect_all(snap, [], _stub_incident())

        assert snap.sensors_succeeded == 0
        assert snap.mcp_health == {}