fix: 세션 설정 정리 + 분석 결과 cascade soft delete + 질문 풀 토큰 절감

ai/src/ai_server/config/settings.py

@@ -33,6 +33,9 @@ class Settings(BaseSettings):
     ai_callback_routing_feedback: str = "callback.feedback"
     ai_callback_routing_voice: str = "callback.voice"
     feedback_rag_top_k: int = 5
+    # 질문 풀 초기 크기. Core 의 applyPool 이 questions[0] 만 INSERT 하므로 1 로 고정해 토큰 낭비 차단.
+    # 후속 작업에서 풀 저장 도입 시 늘리기 (예: 5).
+    questions_initial_pool_size: int = 1
 
     # STT (음성 답변). "auto" 면 deepgram > openai_whisper > mock 순으로 키 보유 여부에 따라 자동 선택.
     stt_provider: Literal["auto", "mock", "openai_whisper", "deepgram"] = "auto"

ai/src/ai_server/messaging/consumers/questions_consumer.py

@@ -24,11 +24,15 @@ def __init__(
         publisher: CallbackPublisher,
         idempotency: LruIdempotencyStore,
         callback_routing_key: str,
+        initial_pool_size: int = 1,
     ) -> None:
         self._generator = generator
         self._publisher = publisher
         self._idempotency = idempotency
         self._callback_routing_key = callback_routing_key
+        # Core 의 QuestionsCallbackService.applyPool 은 questions[0] 만 INSERT 하고 나머지는 폐기.
+        # 토큰 낭비를 줄이기 위해 envelope.max_questions 대신 풀 크기를 강제. 후속 작업에서 풀 저장 도입 시 늘리기 쉬움.
+        self._initial_pool_size = max(1, initial_pool_size)
 
     async def handle(self, message: AbstractIncomingMessage) -> None:
         async with message.process(requeue=False):
@@ -53,20 +57,22 @@ async def handle(self, message: AbstractIncomingMessage) -> None:
                 return
 
             req = envelope.payload
+            effective_pool_size = self._initial_pool_size  # envelope.max_questions 무시
             log.info(
                 "questions.generate.start",
                 message_id=envelope.message_id,
                 session_id=req.session_id,
                 doc_count=len(req.documents),
                 max_questions=req.max_questions,
+                pool_size=effective_pool_size,
                 trace_id=envelope.trace_id,
             )
 
             context_text = _build_context(req.documents)
             pool = await self._generator.generate(
                 job_category=req.job_category,
                 interview_type=req.interview_type,
-                max_questions=req.max_questions,
+                max_questions=effective_pool_size,
                 context=context_text,
             )
 

ai/src/ai_server/messaging/runner.py

@@ -153,6 +153,7 @@ def __init__(self, settings: Settings) -> None:
             publisher=self._publisher,
             idempotency=self._idempotency,
             callback_routing_key=settings.ai_callback_routing_questions,
+            initial_pool_size=settings.questions_initial_pool_size,
         )
 
         # 꼬리질문 생성 (US-19)

ai/tests/test_questions_consumer.py

@@ -93,7 +93,8 @@ async def test_consumer_generates_questions_and_publishes_callback():
     call = generator.generate.await_args
     assert call.kwargs["job_category"] == "BACKEND"
     assert call.kwargs["interview_type"] == "TECHNICAL"
-    assert call.kwargs["max_questions"] == 5
+    # envelope.max_questions(=5) 무시하고 initial_pool_size(default 1) 로 강제. Core 가 첫 질문만 사용.
+    assert call.kwargs["max_questions"] == 1
     assert "Java" in call.kwargs["context"]
 
     publisher.publish.assert_awaited_once()

backend/src/main/java/com/stackup/stackup/document/application/AnalyzedDocumentCascadeListener.java

@@ -0,0 +1,52 @@
+package com.stackup.stackup.document.application;
+
+import com.stackup.stackup.document.domain.AnalyzedDocument;
+import com.stackup.stackup.document.domain.AnalyzedDocumentRepository;
+import com.stackup.stackup.github.application.event.RepositoryDeletedEvent;
+import com.stackup.stackup.resume.application.event.ResumeDeletedEvent;
+import java.util.List;
+import lombok.RequiredArgsConstructor;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+import org.springframework.context.event.EventListener;
+import org.springframework.stereotype.Component;
+import org.springframework.transaction.annotation.Transactional;
+
+// Resume / GithubRepository soft delete 발생 시 관련 AnalyzedDocument 도 cascade soft delete.
+// 도메인 cycle 회피 — resume/github 가 document 도메인을 직접 import 하지 않고 이벤트로 위임.
+@Component
+@RequiredArgsConstructor
+public class AnalyzedDocumentCascadeListener {
+
+    private static final Logger log = LoggerFactory.getLogger(AnalyzedDocumentCascadeListener.class);
+
+    private final AnalyzedDocumentRepository analyzedDocumentRepository;
+
+    @EventListener
+    @Transactional
+    public void on(ResumeDeletedEvent event) {
+        List<AnalyzedDocument> docs = analyzedDocumentRepository
+            .findActiveByResumeIdAndOwner(event.resumeId(), event.userId());
+        for (AnalyzedDocument doc : docs) {
+            doc.markDeleted();
+        }
+        if (!docs.isEmpty()) {
+            log.info("AnalyzedDocument cascade soft delete (resume). userId={}, resumeId={}, count={}",
+                event.userId(), event.resumeId(), docs.size());
+        }
+    }
+
+    @EventListener
+    @Transactional
+    public void on(RepositoryDeletedEvent event) {
+        List<AnalyzedDocument> docs = analyzedDocumentRepository
+            .findActiveByRepositoryIdAndOwner(event.repositoryId(), event.userId());
+        for (AnalyzedDocument doc : docs) {
+            doc.markDeleted();
+        }
+        if (!docs.isEmpty()) {
+            log.info("AnalyzedDocument cascade soft delete (repository). userId={}, repositoryId={}, count={}",
+                event.userId(), event.repositoryId(), docs.size());
+        }
+    }
+}

backend/src/main/java/com/stackup/stackup/document/domain/AnalyzedDocument.java

@@ -106,4 +106,8 @@ public void markFailed(String errorCode, String errorMessage) {
         this.errorCode = errorCode;
         this.errorMessage = errorMessage;
     }
+
+    public void markDeleted() {
+        this.deleted = true;
+    }
 }

backend/src/main/java/com/stackup/stackup/github/application/GithubRepositoryService.java

@@ -5,6 +5,7 @@
 import com.stackup.stackup.github.application.dto.CandidateRepositoryResult;
 import com.stackup.stackup.github.application.dto.GithubRepositoryResult;
 import com.stackup.stackup.github.application.dto.RegisterRepositoryCommand;
+import com.stackup.stackup.github.application.event.RepositoryDeletedEvent;
 import com.stackup.stackup.github.application.event.RepositoryRegisteredEvent;
 import com.stackup.stackup.github.domain.GithubRepository;
 import com.stackup.stackup.github.domain.GithubRepositoryRepository;
@@ -80,7 +81,9 @@ public GithubRepositoryResult get(Long userId, Long repositoryId) {
     @Transactional
     public void delete(Long userId, Long repositoryId) {
         GithubRepository repo = loadOwned(userId, repositoryId);
-        repositoryRepository.delete(repo);
+        repo.markDeleted();
+        // 분석 결과 cascade — document 도메인 listener 가 RepositoryDeletedEvent 수신해 AnalyzedDocument soft delete.
+        events.publishEvent(new RepositoryDeletedEvent(userId, repositoryId));
     }
 
     // US-08: GitHub 메타 재동기화. 등록된 레포의 default_branch / name / url 등을 최신화.

backend/src/main/java/com/stackup/stackup/github/application/event/RepositoryDeletedEvent.java

@@ -0,0 +1,5 @@
+package com.stackup.stackup.github.application.event;
+
+// GithubRepository soft delete commit 후 발화. document 도메인이 수신해 관련 AnalyzedDocument cascade soft delete.
+public record RepositoryDeletedEvent(Long userId, Long repositoryId) {
+}

backend/src/main/java/com/stackup/stackup/github/domain/GithubRepository.java

@@ -113,4 +113,8 @@ public void markAnalyzed() {
     public void markFailed() {
         this.status = RepositoryStatus.FAILED;
     }
+
+    public void markDeleted() {
+        this.deleted = true;
+    }
 }

backend/src/main/java/com/stackup/stackup/resume/application/ResumeService.java

@@ -5,6 +5,7 @@
 import com.stackup.stackup.common.storage.ObjectStorageClient;
 import com.stackup.stackup.resume.application.dto.ResumeResult;
 import com.stackup.stackup.resume.application.dto.ResumeUploadCommand;
+import com.stackup.stackup.resume.application.event.ResumeDeletedEvent;
 import com.stackup.stackup.resume.application.event.ResumeUploadedEvent;
 import com.stackup.stackup.resume.domain.Resume;
 import com.stackup.stackup.resume.domain.ResumeFileType;
@@ -65,8 +66,10 @@ public ResumeResult get(Long userId, Long resumeId) {
     @Transactional
     public void delete(Long userId, Long resumeId) {
         Resume resume = loadOwned(userId, resumeId);
-        // 객체(S3 본문) 삭제는 후속 정책(보존 vs 즉시 cleanup)에 따라. 우선 row soft delete.
-        resumeRepository.delete(resume);
+        resume.markDeleted();
+        // 분석 결과 cascade — document 도메인 listener 가 ResumeDeletedEvent 받아 AnalyzedDocument soft delete.
+        // 직접 의존 회피 (ArchUnit 도메인 cycle 방지).
+        events.publishEvent(new ResumeDeletedEvent(userId, resumeId));
     }
 
     private Resume loadOwned(Long userId, Long resumeId) {

backend/src/main/java/com/stackup/stackup/resume/application/event/ResumeDeletedEvent.java

@@ -0,0 +1,5 @@
+package com.stackup.stackup.resume.application.event;
+
+// Resume soft delete commit 후 발화. document 도메인이 수신해 관련 AnalyzedDocument cascade soft delete.
+public record ResumeDeletedEvent(Long userId, Long resumeId) {
+}

backend/src/main/java/com/stackup/stackup/resume/domain/Resume.java

@@ -80,4 +80,8 @@ public void markAnalyzed() {
     public void markFailed() {
         this.status = ResumeStatus.FAILED;
     }
+
+    public void markDeleted() {
+        this.deleted = true;
+    }
 }

backend/src/main/java/com/stackup/stackup/session/presentation/dto/SessionCreateRequest.java

@@ -13,18 +13,21 @@
 public record SessionCreateRequest(
     @Size(max = 200) String title,
     @Size(max = 4000) String memo,
-    @NotNull SessionMode mode,
+    // mode: 현재 시연 범위에선 ONLINE 만 유효. 미입력 시 default 적용 (Sprint 4 에서 정식 분기).
+    SessionMode mode,
     @NotNull InterviewType interviewType,
     @NotNull JobCategory jobCategory,
-    @Min(1) @Max(30) Integer maxQuestions,
-    @Min(5) @Max(180) Integer maxDurationMinutes,
+    // 첫 질문 + 꼬리질문 최소 1쌍 보장을 위해 하한 2. applyFollowup 자동 종료 흐름 결합.
+    @Min(2) @Max(30) Integer maxQuestions,
+    // maxDurationMinutes: 시간 기반 자동 종료 미구현. DB 저장만 유지, 입력 검증 풀어 미입력 허용.
+    Integer maxDurationMinutes,
     List<Long> contextDocumentIds
 ) {
     public SessionCreateCommand toCommand() {
         return new SessionCreateCommand(
             title,
             memo,
-            mode,
+            mode == null ? SessionMode.ONLINE : mode,
             interviewType,
             jobCategory,
             maxQuestions,