awoooi/apps/api/tests/test_global_repair_cooldown.py

"""
Global Repair Cooldown 測試
===========================
ADR-039: 全域修復熔斷機制

測試項目：
- 有狀態服務黑名單檢查
- 全域計數閾值
- Redis 故障降級

注意：需要 Redis 環境，測試會使用獨立的 key 前綴
"""

import pytest

from src.services.global_repair_cooldown import (
    GLOBAL_COOLDOWN_KEY,
    GLOBAL_COOLDOWN_THRESHOLD,
    STATEFUL_SERVICE_BLACKLIST,
    check_global_repair_cooldown,
    get_global_repair_status,
    record_global_repair_action,
)


class TestStatefulServiceBlacklist:
    """有狀態服務黑名單測試"""

    @pytest.mark.asyncio
    async def test_postgres_blocked(self):
        """PostgreSQL 服務應該被阻擋"""
        can_repair, reason = await check_global_repair_cooldown(
            incident_id="test-001",
            affected_services=["awoooi-postgres"],
        )
        assert not can_repair
        assert "有狀態服務" in reason
        assert "禁止自動重啟" in reason

    @pytest.mark.asyncio
    async def test_redis_blocked(self):
        """Redis 服務應該被阻擋"""
        can_repair, reason = await check_global_repair_cooldown(
            incident_id="test-002",
            affected_services=["redis-stack"],
        )
        assert not can_repair
        assert "有狀態服務" in reason

    @pytest.mark.asyncio
    async def test_clickhouse_blocked(self):
        """ClickHouse 服務應該被阻擋"""
        can_repair, reason = await check_global_repair_cooldown(
            incident_id="test-003",
            affected_services=["signoz-clickhouse-0"],
        )
        assert not can_repair
        assert "有狀態服務" in reason

    @pytest.mark.asyncio
    async def test_stateless_service_allowed(self):
        """無狀態服務應該被允許"""
        can_repair, reason = await check_global_repair_cooldown(
            incident_id="test-004",
            affected_services=["awoooi-api-deployment"],
        )
        assert can_repair
        assert "允許" in reason

    @pytest.mark.asyncio
    async def test_empty_services_allowed(self):
        """空服務列表應該被允許"""
        can_repair, reason = await check_global_repair_cooldown(
            incident_id="test-005",
            affected_services=[],
        )
        assert can_repair

    @pytest.mark.asyncio
    async def test_none_services_allowed(self):
        """None 服務列表應該被允許"""
        can_repair, reason = await check_global_repair_cooldown(
            incident_id="test-006",
            affected_services=None,
        )
        assert can_repair

    def test_blacklist_contains_common_stateful_services(self):
        """黑名單應該包含常見有狀態服務"""
        assert "postgres" in STATEFUL_SERVICE_BLACKLIST
        assert "redis" in STATEFUL_SERVICE_BLACKLIST
        assert "clickhouse" in STATEFUL_SERVICE_BLACKLIST
        assert "elasticsearch" in STATEFUL_SERVICE_BLACKLIST
        assert "etcd" in STATEFUL_SERVICE_BLACKLIST
        assert "minio" in STATEFUL_SERVICE_BLACKLIST


class TestGlobalCooldown:
    """全域冷卻期測試 - 需要 Redis"""

    @pytest.fixture
    async def clean_redis(self):
        """清理測試用 Redis key"""
        from src.core.redis_client import get_redis

        redis = get_redis()
        await redis.delete(GLOBAL_COOLDOWN_KEY)
        yield
        await redis.delete(GLOBAL_COOLDOWN_KEY)

    @pytest.mark.asyncio
    async def test_record_increments_counter(self, clean_redis):
        """記錄應該增加計數"""
        from src.core.redis_client import get_redis

        redis = get_redis()

        # 記錄一次
        await record_global_repair_action()

        count = await redis.get(GLOBAL_COOLDOWN_KEY)
        assert int(count) == 1

    @pytest.mark.asyncio
    async def test_record_sets_ttl(self, clean_redis):
        """第一次記錄應該設定 TTL"""
        from src.core.redis_client import get_redis

        redis = get_redis()

        await record_global_repair_action()

        ttl = await redis.ttl(GLOBAL_COOLDOWN_KEY)
        assert ttl > 0
        assert ttl <= 900  # 15 分鐘

    @pytest.mark.asyncio
    async def test_cooldown_triggers_after_threshold(self, clean_redis):
        """超過閾值後應該觸發凍結"""
        # 記錄 5 次（達到閾值）
        for _ in range(GLOBAL_COOLDOWN_THRESHOLD):
            await record_global_repair_action()

        can_repair, reason = await check_global_repair_cooldown(
            incident_id="test-threshold",
            affected_services=["awoooi-api"],
        )

        assert not can_repair
        assert "超出安全閾值" in reason
        assert str(GLOBAL_COOLDOWN_THRESHOLD) in reason

    @pytest.mark.asyncio
    async def test_get_status_returns_correct_info(self, clean_redis):
        """狀態 API 應該返回正確資訊"""
        # 記錄 2 次
        await record_global_repair_action()
        await record_global_repair_action()

        status = await get_global_repair_status()

        assert status["current_count"] == 2
        assert status["threshold"] == GLOBAL_COOLDOWN_THRESHOLD
        assert not status["is_frozen"]
        assert status["ttl_remaining"] is not None
        assert status["ttl_remaining"] > 0