test: stabilize pytest setup for gateway and evaluator

pytest.ini

@@ -0,0 +1,5 @@
+[pytest]
+addopts = --import-mode=importlib
+testpaths =
+    services/gateway-api/tests
+    services/evaluator/tests

scripts/test-pytest.sh

@@ -0,0 +1,9 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
+
+cd "$ROOT_DIR"
+
+uv run --project services/gateway-api --group dev pytest services/gateway-api/tests "$@"
+uv run --project services/evaluator --group dev pytest services/evaluator/tests "$@"

services/evaluator/app/config.py

@@ -1,4 +1,6 @@
-from pydantic_settings import BaseSettings
+from functools import lru_cache
+
+from pydantic_settings import BaseSettings, SettingsConfigDict
 
 
 class Settings(BaseSettings):
@@ -7,11 +9,11 @@ class Settings(BaseSettings):
     log_level: str = "INFO"
 
     # DB
-    database_url: str
+    database_url: str | None = None
 
     # LLM (Judge 용)
     llm_api_base_url: str | None = None
-    llm_api_key: str
+    llm_api_key: str | None = None
     openai_model_judge: str = "gpt-5-mini"
 
     # Batch Evaluation Scheduler
@@ -33,9 +35,12 @@ class Settings(BaseSettings):
     smtp_from_email: str | None = None  # 발신자 이메일
     smtp_to_emails: str | None = None  # 수신자 이메일들 (쉼표로 구분)
 
-    class Config:
-        env_file = ".env"
-        env_file_encoding = "utf-8"
+    model_config = SettingsConfigDict(
+        env_file=".env",
+        env_file_encoding="utf-8",
+    )
 
 
-settings = Settings()
+@lru_cache
+def get_settings() -> Settings:
+    return Settings()

services/evaluator/app/db.py

@@ -1,16 +1,29 @@
+from functools import lru_cache
+
 from sqlalchemy import create_engine
 from sqlalchemy.orm import sessionmaker, declarative_base
 
-from .config import settings
-
-engine = create_engine(settings.database_url)
-SessionLocal = sessionmaker(autocommit=False, autoflush=False, bind=engine)
+from .config import get_settings
 
 Base = declarative_base()
 
 
+@lru_cache
+def get_engine():
+    settings = get_settings()
+    if not settings.database_url:
+        raise RuntimeError("DATABASE_URL is required to initialize the database engine.")
+
+    return create_engine(settings.database_url)
+
+
+@lru_cache
+def get_session_factory():
+    return sessionmaker(autocommit=False, autoflush=False, bind=get_engine())
+
+
 def get_db():
-    db = SessionLocal()
+    db = get_session_factory()()
     try:
         yield db
     finally:

services/evaluator/app/llm_judge.py

@@ -1,6 +1,7 @@
 import json
 import textwrap
 import time
+from functools import lru_cache
 from typing import TypedDict
 
 from fastapi import HTTPException
@@ -12,7 +13,7 @@
     AuthenticationError,
 )
 
-from .config import settings
+from .config import get_settings
 from .models import LLMLog
 
 
@@ -24,10 +25,19 @@ class EvaluationResult(TypedDict):
     raw_judge_response: str
 
 
-client = OpenAI(
-    api_key=settings.llm_api_key,
-    base_url=settings.llm_api_base_url or None,
-)
+@lru_cache
+def get_client() -> OpenAI:
+    settings = get_settings()
+    if not settings.llm_api_key:
+        raise HTTPException(
+            status_code=500,
+            detail="LLM_API_KEY is required for LLM judge evaluation.",
+        )
+
+    return OpenAI(
+        api_key=settings.llm_api_key,
+        base_url=settings.llm_api_base_url or None,
+    )
 
 
 def build_evaluation_prompt(log: LLMLog) -> str:
@@ -110,7 +120,9 @@ def run_judge(log: LLMLog) -> EvaluationResult:
     """
     하나의 LLMLog에 대해 Judge LLM을 호출하고 EvaluationResult 반환.
     """
+    settings = get_settings()
     prompt = build_evaluation_prompt(log)
+    client = get_client()
 
     try:
         start = time.perf_counter()

services/evaluator/app/main.py

@@ -6,16 +6,18 @@
 import time
 from prometheus_client import generate_latest, CONTENT_TYPE_LATEST
 
-from .db import Base, engine, get_db
+from .db import Base, get_db, get_engine
 from .models import LLMLog, LLMEvaluation
 from .rules import basic_rule_evaluate
 from .llm_judge import run_judge
-from .config import settings
+from .config import get_settings
 from .scheduler import start_scheduler, stop_scheduler
 from .utils import get_pending_logs
 from .metrics import record_evaluation, update_pending_logs_count
 from .notifier import send_low_quality_alert
 
+settings = get_settings()
+
 # 로깅 설정
 logging.basicConfig(
     level=getattr(logging, settings.log_level.upper()),
@@ -24,29 +26,33 @@
 logger = logging.getLogger(__name__)
 
 
-@asynccontextmanager
-async def lifespan(app: FastAPI):
-    """
-    FastAPI 앱의 수명 주기 관리.
-    시작 시 테이블 생성 및 스케줄러 시작, 종료 시 스케줄러 중지.
-    """
-    # Startup
-    logger.info("Starting Evaluator Service...")
-    Base.metadata.create_all(bind=engine)
-    start_scheduler()
-    yield
-    # Shutdown
-    logger.info("Stopping Evaluator Service...")
-    stop_scheduler()
-
-
-# FastAPI 앱 생성
-app = FastAPI(
-    title="LLM Quality Observer - Evaluator Service",
-    description="룰 기반 및 LLM-as-a-judge 방식으로 LLM 응답 품질을 평가하는 서비스",
-    version="1.0.0",
-    lifespan=lifespan,
-)
+def create_app(*, testing: bool = False) -> FastAPI:
+    @asynccontextmanager
+    async def lifespan(app: FastAPI):
+        """
+        FastAPI 앱의 수명 주기 관리.
+        시작 시 테이블 생성 및 스케줄러 시작, 종료 시 스케줄러 중지.
+        """
+        if not getattr(app.state, "testing", False):
+            logger.info("Starting Evaluator Service...")
+            Base.metadata.create_all(bind=get_engine())
+            start_scheduler()
+        yield
+        if not getattr(app.state, "testing", False):
+            logger.info("Stopping Evaluator Service...")
+            stop_scheduler()
+
+    app = FastAPI(
+        title="LLM Quality Observer - Evaluator Service",
+        description="룰 기반 및 LLM-as-a-judge 방식으로 LLM 응답 품질을 평가하는 서비스",
+        version="1.0.0",
+        lifespan=lifespan,
+    )
+    app.state.testing = testing
+    return app
+
+
+app = create_app()
 
 
 @app.get("/health")
@@ -57,7 +63,7 @@ def health_check():
     """
     return {
         "status": "ok",
-        "env": settings.app_env,
+        "env": get_settings().app_env,
     }
 
 
@@ -93,6 +99,7 @@ def evaluate_once(
     pending_logs = get_pending_logs(db, limit=limit)
 
     if not pending_logs:
+        settings = get_settings()
         return {
             "evaluated": 0,
             "judge_type": judge_type,
@@ -102,6 +109,7 @@ def evaluate_once(
     # 2. 각 로그에 대해 평가 수행
     evaluated_count = 0
     judge_model_name = ""
+    settings = get_settings()
 
     for log in pending_logs:
         try:

services/evaluator/app/notifier.py

@@ -11,7 +11,7 @@
 from email.mime.text import MIMEText
 from email.mime.multipart import MIMEMultipart
 
-from .config import settings
+from .config import get_settings
 from .models import LLMLog, LLMEvaluation
 from .metrics import record_notification, record_low_quality_alert
 
@@ -29,6 +29,8 @@ def send_slack_notification(message: str, notification_type: str = "alert") -> b
     Returns:
         bool: 전송 성공 여부
     """
+    settings = get_settings()
+
     if not settings.slack_webhook_url:
         logger.debug("Slack webhook URL이 설정되지 않았습니다.")
         return False
@@ -61,6 +63,8 @@ def send_discord_notification(message: str, notification_type: str = "alert") ->
     Returns:
         bool: 전송 성공 여부
     """
+    settings = get_settings()
+
     if not settings.discord_webhook_url:
         logger.debug("Discord webhook URL이 설정되지 않았습니다.")
         return False
@@ -95,6 +99,8 @@ async def send_email_notification(subject: str, message: str, notification_type:
     Returns:
         bool: 전송 성공 여부
     """
+    settings = get_settings()
+
     if not all([
         settings.smtp_host,
         settings.smtp_username,
@@ -154,6 +160,8 @@ def send_low_quality_alert(log: LLMLog, evaluation: LLMEvaluation):
         log: LLM 로그
         evaluation: 평가 결과
     """
+    settings = get_settings()
+
     if evaluation.overall_score >= settings.notification_score_threshold:
         # 임계값 이상이면 알림 안 보냄
         return

services/evaluator/app/scheduler.py

@@ -9,8 +9,8 @@
 from apscheduler.triggers.interval import IntervalTrigger
 from sqlalchemy.orm import Session
 
-from .config import settings
-from .db import SessionLocal
+from .config import get_settings
+from .db import get_session_factory
 from .utils import get_pending_logs
 from .models import LLMLog, LLMEvaluation
 from .rules import basic_rule_evaluate
@@ -36,7 +36,8 @@ def run_batch_evaluation():
     """
     logger.info("Starting batch evaluation...")
 
-    db: Session = SessionLocal()
+    settings = get_settings()
+    db: Session = get_session_factory()()
     try:
         # 1. 평가 대기 중인 로그 가져오기
         pending_logs = get_pending_logs(
@@ -148,6 +149,7 @@ def start_scheduler():
     스케줄러를 시작합니다.
     """
     global scheduler
+    settings = get_settings()
 
     if not settings.enable_auto_evaluation:
         logger.info("Auto evaluation is disabled")

services/evaluator/app/schemas.py

@@ -1,5 +1,5 @@
 from typing import Optional
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, ConfigDict, Field
 
 
 class EvaluationResult(BaseModel):
@@ -14,5 +14,4 @@ class EvaluationResult(BaseModel):
     judge_model: str = Field(default="rule-basic-v1", description="평가에 사용된 모델/룰 버전")
     comment: Optional[str] = Field(default=None, description="평가 근거 또는 코멘트")
 
-    class Config:
-        from_attributes = True  # Pydantic v2에서 ORM 모드 활성화
+    model_config = ConfigDict(from_attributes=True)

services/evaluator/pyproject.toml

@@ -23,5 +23,10 @@ dependencies = [
 requires = ["setuptools", "wheel"]
 build-backend = "setuptools.build_meta"
 
+[dependency-groups]
+dev = [
+    "pytest>=8.0",
+]
+
 [tool.setuptools]
 packages = ["app"]

services/evaluator/tests/test_health.py

@@ -2,9 +2,20 @@
 Health check endpoint tests
 """
 
+import sys
+from pathlib import Path
+
 from fastapi.testclient import TestClient
+
+SERVICE_ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(SERVICE_ROOT))
+
+for module_name in [name for name in list(sys.modules) if name == "app" or name.startswith("app.")]:
+    sys.modules.pop(module_name)
+
 from app.main import app
 
+app.state.testing = True
 client = TestClient(app)
 
 

services/gateway-api/app/config.py

@@ -1,10 +1,12 @@
-from pydantic_settings import BaseSettings
+from functools import lru_cache
+
+from pydantic_settings import BaseSettings, SettingsConfigDict
 
 
 class Settings(BaseSettings):
     app_env: str = "local"
 
-    database_url: str
+    database_url: str | None = None
     openai_model_main: str = "gpt-5-mini"
 
     llm_api_base_url: str | None = None
@@ -16,9 +18,12 @@ class Settings(BaseSettings):
 
     log_level: str = "INFO"
 
-    class Config:
-        env_file = ".env"
-        env_file_encoding = "utf-8"
+    model_config = SettingsConfigDict(
+        env_file=".env",
+        env_file_encoding="utf-8",
+    )
 
 
-settings = Settings()
+@lru_cache
+def get_settings() -> Settings:
+    return Settings()