🔥 TIL - Day 63

팀 프로젝트 2일차!

📌 오늘의 구현이슈

• 처음보는 에러.. 406 not acceptable
• 코드리뷰 상세정보 쿼리 최적화

---

📌 406 not acceptable

찾아보니 꽤나 흔한? 에러코드에 해당하는 것 같다.
발생하는 케이스가 굉장히 많은 것 같다.

나의 경우 Springboot 환경에서 발생했다.

일단 에러의 이유는 응답객체를 Json으로 변환하지 못하는 문제 때문이었다.
이게 왜 406인지는 잘 납득이 안 되지만 그렇다고 한다.. (객체에 접근을 못한다는 뜻..?)

응답객체 (응답dto)를 구성하고 Getter를 추가해줘야 Spring이 내부적으로 Json포맷으로 변환해줄 수 있는 것이다. 기존 dto에 게터를 추가함으로 문제가 해결됐다.

---

📌 코드리뷰 상세정보 쿼리 최적화

현재 코드리뷰 상세정보에는 코드리뷰에 대한 정보와 해당 코드리뷰를 남긴 User, 해당 코드리뷰에 남겨진 Comment(댓글)로 이뤄져 있다.

상세정보를 조회할 때 User에 대한 정보와 Comment(댓글)에 대한 정보를 함께 가져와야 한다.
처음 작성한 쿼리는 아래와 같다. (그냥 쪼인..)

    @Override
    public ReviewRequestDetailResponseDto getReviewRequestDetailWithComment(Long id) {

        ReviewRequest result = query.select(reviewRequest)
                .from(reviewRequest)
                .join(reviewRequest.reviewRequestComments, reviewRequestComment)
                .join(reviewRequest.requestUser, user)
                .where(reviewRequest.id.eq(id))
                .fetchFirst();

        List<CommentResponseDto> comments = result.getReviewRequestComments().stream().map(
                c -> new CommentResponseDto(c.getId(), c.getUser().getId(), c.getUser().getUsername(), c.getContent(), c.getCreatedAt())
        ).collect(Collectors.toList());


        return new ReviewRequestDetailResponseDto(
                result.getId(), result.getRequestUser().getUsername(), result.getTitle(), result.getCode(), result.getComment(),
                result.getStatus().toString(), result.getCreatedAt(), comments
        );
    }

쿼리결과를 보자.
한 개의 코드리뷰 요청서에 두 명의 유저가 각각 두 개씩 총 4개의 댓글이 달린 경우의 쿼리이다.

Hibernate: 
    select
        reviewrequ0_.id as id1_3_,
        reviewrequ0_.created_at as created_2_3_,
        reviewrequ0_.updated_at as updated_3_3_,
        reviewrequ0_.answer_user_id as answer_u9_3_,
        reviewrequ0_.code as code4_3_,
        reviewrequ0_.comment as comment5_3_,
        reviewrequ0_.language_name as language6_3_,
        reviewrequ0_.request_user_id as request10_3_,
        reviewrequ0_.review_answer_id as review_11_3_,
        reviewrequ0_.status as status7_3_,
        reviewrequ0_.title as title8_3_ 
    from
        review_request reviewrequ0_ 
    inner join
        review_request_comment reviewrequ1_ 
            on reviewrequ0_.id=reviewrequ1_.review_request_id 
    inner join
        user user2_ 
            on reviewrequ0_.request_user_id=user2_.id 
    where
        reviewrequ0_.id=? limit ?
Hibernate: 
    select
        reviewrequ0_.review_request_id as review_r5_4_0_,
        reviewrequ0_.id as id1_4_0_,
        reviewrequ0_.id as id1_4_1_,
        reviewrequ0_.created_at as created_2_4_1_,
        reviewrequ0_.updated_at as updated_3_4_1_,
        reviewrequ0_.content as content4_4_1_,
        reviewrequ0_.review_request_id as review_r5_4_1_,
        reviewrequ0_.user_id as user_id6_4_1_ 
    from
        review_request_comment reviewrequ0_ 
    where
        reviewrequ0_.review_request_id=?
Hibernate: 
    select
        user0_.id as id1_5_0_,
        user0_.created_at as created_2_5_0_,
        user0_.updated_at as updated_3_5_0_,
        user0_.password as password4_5_0_,
        user0_.ranking_point as ranking_5_5_0_,
        user0_.review_count as review_c6_5_0_,
        user0_.user_role as user_rol7_5_0_,
        user0_.username as username8_5_0_ 
    from
        user user0_ 
    where
        user0_.id=?
Hibernate: 
    select
        user0_.id as id1_5_0_,
        user0_.created_at as created_2_5_0_,
        user0_.updated_at as updated_3_5_0_,
        user0_.password as password4_5_0_,
        user0_.ranking_point as ranking_5_5_0_,
        user0_.review_count as review_c6_5_0_,
        user0_.user_role as user_rol7_5_0_,
        user0_.username as username8_5_0_ 
    from
        user user0_ 
    where
        user0_.id=?
Hibernate: 
    select
        user0_.id as id1_5_0_,
        user0_.created_at as created_2_5_0_,
        user0_.updated_at as updated_3_5_0_,
        user0_.password as password4_5_0_,
        user0_.ranking_point as ranking_5_5_0_,
        user0_.review_count as review_c6_5_0_,
        user0_.user_role as user_rol7_5_0_,
        user0_.username as username8_5_0_ 
    from
        user user0_ 
    where
        user0_.id=?

지____저분
굳이 한개 한개 추적해보지 않아도 N+1이 터진 것을 알 수 있다.

쿼리의 결과를 DTO로 변환할 때 Lazy로 설정된 프록시 객체를 실제 객체로 가져오기 위해 쿼리를 날린다. N+1 !..!

N+1의 특효약은 Fetch Join 이다.
Querydsl을 이용해서 패치조인을 적용해봤다.

    @Override
    public ReviewRequestDetailResponseDto getReviewRequestDetailWithComment(Long id) {

        ReviewRequest result = query.select(reviewRequest).distinct()
                .from(reviewRequest)
                .join(reviewRequest.reviewRequestComments, reviewRequestComment).fetchJoin()
                .join(reviewRequest.requestUser, user).fetchJoin()
                .where(reviewRequest.id.eq(id))
                .fetchFirst();

        List<CommentResponseDto> comments = result.getReviewRequestComments().stream().map(
                c -> new CommentResponseDto(c.getId(), c.getUser().getId(), c.getUser().getUsername(), c.getContent(), c.getCreatedAt())
        ).collect(Collectors.toList());


        return new ReviewRequestDetailResponseDto(
                result.getId(), result.getRequestUser().getUsername(), result.getTitle(), result.getCode(), result.getComment(),
                result.getStatus().toString(), result.getCreatedAt(), comments
        );
    }

컬렉션(댓글)을 기준으로 패치조인을 하는 경우 발생할 수 있는 문제를 해결하기 위해 distinct()를 추가했다. (컬렉션 쪽에 맞춰서 조회결과가 늘어나는 문제)

fetchJoin() 을 적용한 결과이다.

Hibernate: 
    select
        distinct reviewrequ0_.id as id1_3_0_,
        reviewrequ1_.id as id1_4_1_,
        user2_.id as id1_5_2_,
        reviewrequ0_.created_at as created_2_3_0_,
        reviewrequ0_.updated_at as updated_3_3_0_,
        reviewrequ0_.answer_user_id as answer_u9_3_0_,
        reviewrequ0_.code as code4_3_0_,
        reviewrequ0_.comment as comment5_3_0_,
        reviewrequ0_.language_name as language6_3_0_,
        reviewrequ0_.request_user_id as request10_3_0_,
        reviewrequ0_.review_answer_id as review_11_3_0_,
        reviewrequ0_.status as status7_3_0_,
        reviewrequ0_.title as title8_3_0_,
        reviewrequ1_.created_at as created_2_4_1_,
        reviewrequ1_.updated_at as updated_3_4_1_,
        reviewrequ1_.content as content4_4_1_,
        reviewrequ1_.review_request_id as review_r5_4_1_,
        reviewrequ1_.user_id as user_id6_4_1_,
        reviewrequ1_.review_request_id as review_r5_4_0__,
        reviewrequ1_.id as id1_4_0__,
        user2_.created_at as created_2_5_2_,
        user2_.updated_at as updated_3_5_2_,
        user2_.password as password4_5_2_,
        user2_.ranking_point as ranking_5_5_2_,
        user2_.review_count as review_c6_5_2_,
        user2_.user_role as user_rol7_5_2_,
        user2_.username as username8_5_2_ 
    from
        review_request reviewrequ0_ 
    inner join
        review_request_comment reviewrequ1_ 
            on reviewrequ0_.id=reviewrequ1_.review_request_id 
    inner join
        user user2_ 
            on reviewrequ0_.request_user_id=user2_.id 
    where
        reviewrequ0_.id=?
Hibernate: 
    select
        user0_.id as id1_5_0_,
        user0_.created_at as created_2_5_0_,
        user0_.updated_at as updated_3_5_0_,
        user0_.password as password4_5_0_,
        user0_.ranking_point as ranking_5_5_0_,
        user0_.review_count as review_c6_5_0_,
        user0_.user_role as user_rol7_5_0_,
        user0_.username as username8_5_0_ 
    from
        user user0_ 
    where
        user0_.id in (
            ?, ?
        )

select 절이 폭발적으로 늘었지만 결과적으로 쿼리의 수가 대폭 줄어들었다.

---

이번 프로젝트는 필수적으로 부하테스트를 포함해야 하기 때문에 이런 최적화 포인트를 남겨놓고 프로젝트 막바지에 성능비교를 해봐야겠다.