feat(be): 주가·환율·벤치마크·뉴스 조회 API 분리

AGENTS.md

@@ -39,7 +39,7 @@ docker build --target test ./backtest_be_fast
 
 **DB schema:** `database/schema.sql` is the first-boot initdb script; schema changes go through Alembic (`backtest_be_fast/alembic/`). There are **no physical foreign keys** — cross-table references are logical, integrity is owned by the app (A-10). A DB created from schema.sql **before** revision `7b2e9c4f1a30` still has the old FK: baseline it with `alembic stamp d5c3763b29e6` then `alembic upgrade head` (plain `stamp head` would skip the FK drop). A DB created from the current schema.sql is already at head (`alembic stamp head`). Migrations are NOT run by the GitHub Actions pipeline or the runtime image — apply them manually. schema.sql and the Alembic head must stay identical (COMMENTs included): change both together and run `scripts/check-schema-parity.sh`. Keep the `SET NAMES utf8mb4;` at the top of schema.sql — the official image runs initdb with a latin1 client and would double-encode Korean COMMENTs. Upgrade/baseline procedures and the MySQL 8.0 → 8.4 notes live in `database/README.md`.
 
-**API:** POST `/api/v1/backtest` calculates results; GET `/api/v1/market-data` independently loads reference data. The console disables all POST `include_*` flags, renders the calculation first, then loads/retries reference data without rerunning it. Legacy POST defaults remain unchanged. See `docs/architecture/api-boundaries.md`. Errors: `@handle_portfolio_errors` decorator on the calculation path.
+**API:** POST `/api/v1/backtest` calculates results; four GETs (`/prices`, `/exchange-rates`, `/benchmarks`, `/news` under `/api/v1`) independently load reference data. The console disables all POST `include_*` flags, renders the calculation first, then loads/retries each resource without rerunning it. The old `/market-data` GET is replaced; legacy POST defaults remain unchanged. All reads share one bounded runner, not four independent capacity pools. See `docs/architecture/api-boundaries.md`. Errors: `@handle_portfolio_errors` decorator on the calculation path.
 
 ## Critical Constraints
 
@@ -84,15 +84,15 @@ docker build --target test ./backtest_be_fast
 - **Verify in Docker** (`docker compose exec`, or `docker build --target test`) before declaring work complete.
 - **BE markers:** `@pytest.mark.unit` (no DB), `@pytest.mark.integration` (DB), `@pytest.mark.external` (real API)
 - **FE:** Vitest + React Testing Library. Playwright E2E exists in `backtest-console` (`playwright.config.ts`, one smoke spec) and needs the dev stack running — it is deliberately NOT in the Docker CI test stage (no browser, no live backend there).
-- **Current baseline (2026-09-27):** BE 492 unit tests + 12 integration + 1 e2e golden master (`pytest tests/e2e -m e2e`, no network/DB), FE 344 tests — all green. Any failure is a regression, not pre-existing noise.
+- **Current Docker baseline (2026-10-25):** BE 529 unit tests + 1 e2e golden master (`pytest tests/e2e -m e2e`, no network/DB); the separate console has 356 tests. The older 12 DB integration tests are a historical baseline, not part of the Docker test target. Runtime HTTP verification for the four reads is `tests/integration/verify_reference_reads_http.py` with isolated synthetic fixtures. Any failure is a regression, not pre-existing noise.
 - **Test files are type-checked** via `tsconfig.test.json` / `npm run type-check:test`. `tsconfig.build.json` deliberately excludes them.
 - **Coverage (BE 2026-08-03 / FE 2026-09-27):** BE 71.6%, FE about 74% statements. Core financial modules are 82-98%; the remaining gaps are `data_fetcher` (41%), the `app/validators/` package (23-32%), and `currency_converter` (57%). The validators look dead but are reached via `backtest_engine.py` → `validation_service` — do not delete them.
 
 ## CI
 
 GitHub Actions lives in `.github/workflows/cicd.yml`. PRs to main/dev and pushes to either branch run Docker tests, dependency audit, runtime build, and schema parity checks. Only main publishes an ARM64 `ghcr.io/kyj0503/backtest` image and deploys its immutable digest through Tailscale and pinned SSH. dev has no OCI deployment. Runtime credentials stay in `/opt/backtest/production/.env` and are never committed.
 
-The dependency audit retains the documented bokeh exception in `scripts/audit-deps.sh`; do not weaken it. Deployment checks `/health/ready` after Docker liveness succeeds and restores the previous release on failure. CI does not automatically migrate a deployed DB. Shared gateway configuration is managed outside this repository at `/opt/gateway`.
+The dependency audit has no exceptions in `scripts/audit-deps.sh`; do not weaken it. Deployment checks `/health/ready` after Docker liveness succeeds and restores the previous release on failure. CI does not automatically migrate a deployed DB. Shared gateway configuration is managed outside this repository at `/opt/gateway`.
 
 ## Commit Convention
 

HISTORY.md

@@ -9,6 +9,28 @@
 
 ---
 
+## 2026-10-25 네 종류의 독립 조회
+
+- **A-39**: 통합 GET `/api/v1/market-data`를 `/prices`, `/exchange-rates`,
+  `/benchmarks`, `/news` 네 GET으로 교체했다. 각 응답은 소유한 자료와 수집 상태만
+  반환한다. 기존 POST 요청·기본값·계산 필드는 유지한다.
+- 기존 수집기를 재사용하되 뉴스·환율·벤치마크의 불필요한 종목 메타데이터 조회를
+  생략한다. 주가와 급등락은 동일 시세를 공유한다. 뉴스에는 기간을 받지 않고,
+  취소 후 다음 종목을 조회하거나 캐시에 쓰지 않도록 취소 확인을 추가했다.
+- 네 GET은 하나의 파일 슬롯 실행기를 공유하며 기본 전체·IP 상한을 각각 4로 맞췄다.
+  API마다 독립 풀을 만들어 총 조회량이 네 배로 늘어나는 구성을 피한다.
+- 콘솔의 로딩·오류·빈 결과·재시도를 종류별로 분리했다. 함수형 상태 병합과 필드
+  허용 목록으로 병렬 응답끼리 덮어쓰지 않으며 재시도는 다른 요청을 취소하지 않는다.
+  새 계산·화면 이탈 때 모든 이전 요청을 취소한다. 뉴스 설명도 최신 뉴스로 바로잡았다.
+- Docker BE 529개·골든 E2E 1개, 콘솔 356개·lint·두 TypeScript 검사,
+  예외 없는 두 의존성 감사와 runtime 빌드를 통과했다. 배포용 이미지·격리 MySQL의
+  합성 시세·뉴스로 네 병렬 조회, 기존 계산값·현금 1,500 보존을 확인했다.
+- 실제 브라우저에서 네 GET에 503을 주고 하나씩 복구했다. 각 재시도는 그 GET만
+  1회 증가시키며 계산 POST는 네 번의 재시도 전후 4건으로 같았다. 운영 자격증명과
+  운영 DB는 사용하지 않았다. 이 검증은 계산 정확성 전반이나 운영 부하 보장이 아니다.
+- 옛 GET을 호출하는 콘솔은 새 API와 호환되지 않으므로 운영 릴리스는 두 저장소를
+  함께 계획하고 이전 탭을 새로고침해야 한다. 현재 계약은 [API 경계](docs/architecture/api-boundaries.md)에 있다.
+
 ## 2026-10-25 API 경계·게이트웨이 보완
 
 - **A-38**: 콘솔의 계산 POST와 참고 자료 GET을 분리했다. 기존 POST 본문·기본값·응답

README.md

@@ -67,8 +67,8 @@ CA 파일은 `/opt/backtest/production/tls/oracle-mysql-ca.pem`에 두고 컨테
 
 `POST /api/v1/backtest`가 주요 API입니다. 최소 백테스트 기간은 30일입니다.
 
-콘솔은 `include_*`를 끄고 계산 결과를 먼저 받은 뒤 `GET /api/v1/market-data`로
-참고 데이터를 조회합니다. 기존 통합 응답은 기본 옵션으로 계속 지원합니다.
+콘솔은 `include_*`를 끄고 계산 결과를 먼저 받은 뒤 `/api/v1/prices`, `/exchange-rates`,
+`/benchmarks`, `/news`를 각각 조회·재시도합니다. 기존 POST 통합 응답은 기본 옵션으로 계속 지원합니다.
 조회·취소·실행 한도와 이전 절차는 [API 경계](docs/architecture/api-boundaries.md)를 참고하세요.
 프론트엔드의 `MIN_BACKTEST_PERIOD_DAYS`와 백엔드 설정을 함께 유지합니다.
 CORS는 공통 Nginx에서 처리합니다. 서버 비밀값을 프론트엔드 `VITE_*` 변수에 넣지 않습니다.

backtest_be_fast/app/api/v1/api.py

@@ -7,7 +7,7 @@
 
 **엔드포인트**:
 - POST /api/v1/backtest: 계산 실행 (기존 include_* 통합 응답도 지원)
-- GET /api/v1/market-data: 계산과 독립적인 주가·환율·벤치마크·뉴스 조회
+- GET /api/v1/prices, /exchange-rates, /benchmarks, /news: 독립 조회·재시도
 - 전략 목록은 프론트엔드에서 관리
 - 콘솔은 계산을 먼저 표시하고 참고 데이터를 별도로 조회한다.
 """
@@ -23,5 +23,5 @@
     tags=["백테스팅"]
 )
 
-api_router.include_router(market_data.router, prefix="/market-data", tags=["시장 데이터"])
+api_router.include_router(market_data.router, tags=["시장 데이터"])
 

backtest_be_fast/app/api/v1/endpoints/market_data.py

@@ -1,4 +1,4 @@
-"""Reference market data, independently retriable without another simulation."""
+"""Independently retriable reads sharing one bounded reference-data capacity."""
 import asyncio
 from datetime import date
 from functools import lru_cache
@@ -19,23 +19,36 @@
 _running = Gauge('market_data_jobs_running', 'Active reference-data jobs', multiprocess_mode='livesum')
 _waiting = Gauge('market_data_jobs_waiting', 'Queued reference-data jobs', multiprocess_mode='livesum')
 
+# Both price views share one fetch. A response can only update its own fields.
+READ_FIELDS = {
+    'prices': ('ticker_info', 'stock_data', 'volatility_events'),
+    'exchange-rates': ('exchange_rates', 'exchange_stats'),
+    'benchmarks': ('sp500_benchmark', 'nasdaq_benchmark'),
+    'news': ('latest_news',),
+}
+READ_STATUSES = {
+    'prices': ('ticker_info', 'stock_data', 'volatility_events'),
+    'exchange-rates': ('exchange_rates',),
+    'benchmarks': ('benchmarks',),
+    'news': ('news',),
+}
 
-class MarketDataQuery(BaseModel):
+
+class SymbolsQuery(BaseModel):
     model_config = {'extra': 'forbid'}
-    symbols: list[str] = Field(default_factory=list, max_length=settings.max_portfolio_items)
-    start_date: date
-    end_date: date
-    include_stock_data: bool = True
-    include_volatility_events: bool = True
-    include_exchange_rates: bool = True
-    include_benchmarks: bool = True
-    include_news: bool = True
+    symbols: list[str] = Field(min_length=1, max_length=settings.max_portfolio_items)
 
     @field_validator('symbols')
     @classmethod
     def validate_symbols(cls, values):
         return list(dict.fromkeys(PortfolioStock(symbol=value).symbol for value in values))
 
+
+class PeriodQuery(BaseModel):
+    model_config = {'extra': 'forbid'}
+    start_date: date
+    end_date: date
+
     @model_validator(mode='after')
     def validate_dates(self):
         if self.end_date < self.start_date:
@@ -47,48 +60,80 @@ def validate_dates(self):
         return self
 
 
+class PricesQuery(PeriodQuery, SymbolsQuery):
+    pass
+
+
+class NewsQuery(SymbolsQuery):
+    limit: int = Field(default=15, ge=1, le=20)
+
+
 @lru_cache(maxsize=1)
 def get_market_data_runner():
     return BacktestJobRunner(
         slot_dir=settings.backtest_slot_dir + '-market-data',
         max_concurrent=settings.max_concurrent_market_data,
-        per_client=1,
+        per_client=4,
         queue_timeout=settings.market_data_queue_timeout_seconds,
         exec_timeout=settings.market_data_timeout_seconds,
         cancel_grace=settings.backtest_cancel_grace_seconds,
-        running_metric=_running,
-        waiting_metric=_waiting,
-        job_name='참고 데이터 조회',
+        running_metric=_running, waiting_metric=_waiting, job_name='참고 데이터 조회',
     )
 
 
-@router.get('', summary='주가·환율·벤치마크·뉴스 조회')
-async def get_market_data(query: Annotated[MarketDataQuery, Query()], request: Request):
+async def _read(section: str, request: Request, *, symbols=None, start_date=None, end_date=None, limit=15):
     async def collect():
         data = await asyncio.to_thread(
             unified_data_service.collect_all_unified_data,
-            symbols=query.symbols,
-            start_date=query.start_date.isoformat(),
-            end_date=query.end_date.isoformat(),
-            include_news=query.include_news,
-            news_display_count=15,
-            include_stock_data=query.include_stock_data,
-            include_volatility_events=query.include_volatility_events,
-            include_exchange_rates=query.include_exchange_rates,
-            include_benchmarks=query.include_benchmarks,
+            symbols=symbols or [],
+            start_date=start_date.isoformat() if start_date else None,
+            end_date=end_date.isoformat() if end_date else None,
+            include_ticker_info=section == 'prices',
+            include_stock_data=section == 'prices',
+            include_volatility_events=section == 'prices',
+            include_exchange_rates=section == 'exchange-rates',
+            include_benchmarks=section == 'benchmarks',
+            include_news=section == 'news',
+            news_display_count=limit,
             timeout_seconds=settings.supplemental_data_timeout_seconds,
         )
-        benchmark = data.get('sp500_benchmark', [])
-        data['sp500_total_return_pct'] = (
-            unified_data_service.calculate_benchmark_return(benchmark)
-            if len(benchmark) >= 2 else None
-        )
-        return {'status': 'success', 'data': recursive_serialize(data)}
+        result = {field: data[field] for field in READ_FIELDS[section]}
+        result['supplemental_status'] = {
+            name: data['supplemental_status'][name] for name in READ_STATUSES[section]
+        }
+        if section == 'benchmarks':
+            benchmark = result['sp500_benchmark']
+            result['sp500_total_return_pct'] = (
+                unified_data_service.calculate_benchmark_return(benchmark)
+                if len(benchmark) >= 2 else None
+            )
+        return {'status': 'success', 'data': recursive_serialize(result)}
 
     try:
         return await get_market_data_runner().run(
-            collect, client_key=_client_key(request),
-            is_disconnected=request.is_disconnected, label='market-data',
+            collect, client_key=_client_key(request), is_disconnected=request.is_disconnected,
+            label=section,
         )
     except BacktestRejected as exc:
         return _rejection_response(exc)
+
+
+@router.get('/prices', summary='종목 정보·주가·급등락 이벤트 조회')
+async def get_prices(query: Annotated[PricesQuery, Query()], request: Request):
+    return await _read('prices', request, symbols=query.symbols,
+                       start_date=query.start_date, end_date=query.end_date)
+
+
+@router.get('/exchange-rates', summary='기간별 원달러 환율·통계 조회')
+async def get_exchange_rates(query: Annotated[PeriodQuery, Query()], request: Request):
+    return await _read('exchange-rates', request, start_date=query.start_date, end_date=query.end_date)
+
+
+@router.get('/benchmarks', summary='기간별 S&P 500·NASDAQ 조회')
+async def get_benchmarks(query: Annotated[PeriodQuery, Query()], request: Request):
+    return await _read('benchmarks', request, start_date=query.start_date, end_date=query.end_date)
+
+
+@router.get('/news', summary='종목별 최신 뉴스 조회 — 백테스트 기간과 무관')
+async def get_news(query: Annotated[NewsQuery, Query()], request: Request):
+    return await _read('news', request, symbols=query.symbols, limit=query.limit)

backtest_be_fast/app/core/config.py

@@ -159,7 +159,7 @@ def backend_cors_origins(self) -> list[str]:
 
     # Read-only market data has its own capacity; slow feeds must not occupy
     # the simulation slots or keep a completed calculation waiting.
-    max_concurrent_market_data: int = 2
+    max_concurrent_market_data: int = 4
     market_data_queue_timeout_seconds: float = 2.0
     market_data_timeout_seconds: float = 20.0
 

backtest_be_fast/app/services/unified_data_service.py

@@ -18,7 +18,7 @@
 )
 from app.repositories.stock_repository import get_stock_repository
 from ..core.config import settings
-from ..core.cancellation import CancelToken, bind_token, current_token, submit_with_context
+from ..core.cancellation import CancelToken, bind_token, current_token, submit_with_context, check_cancelled
 
 logger = logging.getLogger(__name__)
 
@@ -273,6 +273,7 @@ def collect_latest_news(
 
         latest_news = {}
         for symbol in symbols:
+            check_cancelled()
             try:
                 # 1. DB에서 먼저 조회 (3시간 이내)
                 cached_news = self.stock_repo.load_ticker_news(symbol, max_age_hours=max_cache_hours)
@@ -286,6 +287,7 @@ def collect_latest_news(
                     logger.info(f"{symbol} 뉴스 캐시가 없거나 오래됨 - API 호출")
                     search_query = self.news_service.get_ticker_query(symbol)
                     news_list = self.news_service.search_news(search_query, display=display)
+                    check_cancelled()
 
                     # 3. API 결과를 DB에 저장
                     if news_list:
@@ -350,15 +352,16 @@ def _fetch_price_histories(
     def collect_all_unified_data(
         self,
         symbols: List[str],
-        start_date: str,
-        end_date: str,
+        start_date: Optional[str] = None,
+        end_date: Optional[str] = None,
         include_news: bool = True,
         news_display_count: int = 20,
         include_stock_data: bool = True,
         include_volatility_events: bool = True,
         include_exchange_rates: bool = True,
         include_benchmarks: bool = True,
         timeout_seconds: Optional[float] = None,
+        include_ticker_info: bool = True,
     ) -> Dict[str, Any]:
         """
         모든 통합 데이터를 한 번에 수집
@@ -374,6 +377,7 @@ def collect_all_unified_data(
             include_exchange_rates: 환율·환율 통계 포함 여부 (A-08)
             include_benchmarks: S&P 500/NASDAQ 포함 여부 (A-08)
             timeout_seconds: 병렬 수집 구간의 시간 예산. None이면 무제한 (A-08)
+            include_ticker_info: 종목 메타데이터 포함. 독립 뉴스·환율·벤치마크 조회는 끈다.
 
         Returns:
             모든 통합 데이터를 포함하는 딕셔너리. 끈 섹션·시간 초과 섹션도 키는
@@ -405,6 +409,8 @@ def collect_all_unified_data(
             타임아웃까지 응답을 기다린 뒤 끝난다(CPU를 쓰지 않는 대기라 동시 계산
             상한의 목적은 유지된다).
         """
+        if (include_stock_data or include_volatility_events or include_exchange_rates or include_benchmarks) and not (start_date and end_date):
+            raise ValueError('가격·환율·벤치마크 조회에는 시작일과 종료일이 필요합니다.')
         total_start = time.perf_counter()
         timings: Dict[str, float] = {}
 
@@ -431,9 +437,10 @@ def timed(stage: str, fn: Callable, *args, **kwargs):
         try:
             # submit_with_context가 제출 시점 컨텍스트(하위 토큰 바인딩)를 스레드로 넘긴다
             with bind_token(supplemental_token):
-                futures['ticker_info'] = submit_with_context(
-                    executor, timed, "ticker_info", self.collect_ticker_info, symbols
-                )
+                if include_ticker_info:
+                    futures['ticker_info'] = submit_with_context(
+                        executor, timed, "ticker_info", self.collect_ticker_info, symbols
+                    )
                 if need_prices:
                     futures['price_history'] = submit_with_context(
                         executor, timed, "price_history", self._fetch_price_histories,