Page Index - 100-hours-a-week/5-yeosa-wiki GitHub Wiki
507 page(s) in this GitHub Wiki:
- Home
- 온세상의 기록, 온기(Ongi)
- AI 기반 사진정리 SNS, 온기입니다.
- 프로젝트 기간: 2025.03.31 ~ present
- 서비스 바로가기
- 기획 피그마 바로가기
- 목차
- 프로젝트 소개
- 주요 기능
- 팀원 소개
- 기술 스택
- 개발 일정
- 아키텍처 구조
- API 명세서
- 하위 위키 보기
- 3‐Tier 모니터링
- Please reload this page
- [AI] FastAPI 서버가 사용하는 임베딩 캐시와 GPU 요청 메시지 큐도 Backend 서버와 같은 Redis에 함께 올려도 될까?
- Please reload this page
- [AI] 개선 아이디어 및 확장 계획
- Please reload this page
- [AI] 단계별 설계를 통한 아키텍처 진화 요약
- Please reload this page
- [AI] 단일 스레드 환경인 파이썬에서, 왜 스레드 수 제한이 유효했을까?
- Please reload this page
- [AI] 로드밸런서 도입 검토
- Please reload this page
- [AI] 모니터링 대상 지표 목록과 수집 시각화 방법
- Please reload this page
- [AI] 모니터링 및 알림 체계
- Please reload this page
- [AI] 사용한 인프라 기술 및 컴퓨팅 자원
- Please reload this page
- [AI] 서비스 구성요소 간 연동 흐름
- Please reload this page
- [AI] 설계 단계별 회고
- Please reload this page
- [AI] 아키텍처 종합 평가
- Please reload this page
- [AI] 여러 인스턴스에서 Gemini API 요청이 동시에 나가면, NAT 게이트웨이에서 병목이 발생하지 않을까?
- Please reload this page
- [AI] 여러 인스턴스에서 동시에 GPU 스타일 변환 요청이 발생하면, GPU 서버에 과도한 부하가 집중되어 병목이 생기지 않을까?
- Please reload this page
- [AI] 오토스케일링 구성
- Please reload this page
- [AI] 완성된 시스템의 성능 품질 평가
- Please reload this page
- [AI] 이 설계가 팀 서비스의 향후 성장과 안정적 운영에 어떻게 기여하는가?
- Please reload this page
- [AI] 이미지 로딩 부하 테스트
- Please reload this page
- [AI] 전체 프로젝트 설계 회고
- Please reload this page
- [AI] 컴포넌트별 도입 이유와 효과
- Please reload this page
- [cloud ‐ trobuleshooting] Nginx Certbot (Docker 기반) 설정 트러블슈팅
- Please reload this page
- [cloud ‐ TroubleShooting] Dashboard 및 Metrics Server Pending 상태
- Please reload this page
- [cloud ‐ troubleshooting] Docker Named Volume 마이그레이션 트러블 슈팅 보고서
- Please reload this page
- [cloud ‐ troubleshooting] GitHub Actions Runner 환경 오동작 트러블슈팅 보고서
- Please reload this page
- [cloud ‐ TroubleShooting] kakao OAuth KOE303
- Please reload this page
- [cloud ‐ TroubleShooting] Kubernetes API 서버 연결 실패
- Please reload this page
- [cloud ‐ TroubleShooting] Kubernetes Master Node Kubelet 장애
- Please reload this page
- [cloud ‐ TroubleShooting] openvpn server를 통해 app server 접속 장애
- Please reload this page
- [cloud ‐ TroubleShooting] OpenVPN 멀티클라우드 양방향 통신 시도
- Please reload this page
- [cloud ‐ TroubleShooting] s3에 이상한 파일 생성
- Please reload this page
- [cloud ‐ troubleshooting] Sentry set‐commits 오류 트러블슈팅
- Please reload this page
- [cloud ‐ TroubleShooting] UTM → Termius 계속 다운되는 문제
- Please reload this page
- [cloud ‐ TroubleShooting] 카카오 맵 API 호출 X
- Please reload this page
- [cloud] GCP서버 자원 자동화 시스템 구축 (with Cloud Function Cloud Build)
- Please reload this page
- [cloud] 1차 릴리즈 BE Script
- Please reload this page
- [cloud] 1차 릴리즈 FE Script
- Please reload this page
- [cloud] 1차 릴리즈 구축 상세
- Please reload this page
- [cloud] 3Tier Max User Load Test
- Please reload this page
- [cloud] 5단계 : Kubeadm 오케스트레이션
- Please reload this page
- [cloud] 6단계 : Kubernetes(AWS ECS or EKS)기반 배포 자동화
- Please reload this page
- [cloud] APM 도입 ‐ SigNoz
- Please reload this page
- [cloud] CloudFront S3를 통한 커스텀 도메인 이미지 제공 구성 가이드
- Please reload this page
- [cloud] DB 재해 복구 전략
- Please reload this page
- [cloud] Fail2Ban 기반 침입 차단 및 Discord 실시간 알림 대응 보고서
- Please reload this page
- [cloud] GCP Cloud Armor 보안 정책 구성 보고서
- Please reload this page
- [cloud] GCP서버 자원 자동화 시스템 구축 및 운영시간 구분 (with Cloud Scheduler Cloud Function Cloud Build)
- Please reload this page
- [cloud] GCS Presigned URL vs Multipart
- Please reload this page
- [cloud] Kubeadm을 활용한 k8s cluster 구축
- Please reload this page
- [cloud] MySQL, Bind Mount → Docker Named Volume
- Please reload this page
- [cloud] Proxy Server?
- Please reload this page
- [cloud] Self‐Hosted Runner & Flyway Migration Documentation
- Please reload this page
- [cloud] Sentry 연동 및 Discord 알림 설정 보고서
- Please reload this page
- [cloud] Spring Boot 로그 S3 자동 저장 및 Discord 알림 시스템 구축 보고서
- Please reload this page
- [cloud] SSH 키 생성 및 GCP 연동 cursor 까지
- Please reload this page
- [cloud] VPN? IPsec VPN ?
- Please reload this page
- [cloud] 관련 문서
- Please reload this page
- [cloud] 버전 관리 정책 (Semantic Versioning)
- Please reload this page
- [Deprecated] Serial Task Queue 구현
- Please reload this page
- [FE] 테크 스펙
- Please reload this page
- [GPU 모델 병렬 처리] Single GPU Inference Using GPU to the Max Potential 요약
- Please reload this page
- [GPU 서버 이전] GPU‐Cuda의 이해
- Please reload this page
- [Gunicorn] gunicorn.config.py 구성 설계
- Please reload this page
- [Gunicorn] 멀티 프로세스 서버 로그 분석
- Please reload this page
- [Gunicorn] 멀티 프로세스 아키텍쳐 Docker 활용 관련
- Please reload this page
- [Gunicorn] 모델 preload 주의 사항 (프로세스 간 리소스 공유 이슈)
- Please reload this page
- [Redis] 도입 진행
- Please reload this page
- [도입 무산] CPU ‐ GPU 이미지와 태스크 결과 전송 구현
- Please reload this page
- [리팩토링] mypy를 이용한 타입힌트, 서브 클래스 시그니처 불일치, 타입 불일치 리팩토링
- Please reload this page
- [멀티 프로세스] CPU‐GPU 서버 모두 워커 프로세스 2 테스트
- Please reload this page
- [멀티 프로세스] GPU 서버 워커 프로세스 3→2 감축 테스트
- Please reload this page
- [멀티 프로세스] 단일 프로세스, 멀티 프로세스 성능 비교(리소스 사용량, 테스크별 지연시간)
- Please reload this page
- [멀티 프로세스][both server] 멀티 프로세스 (병목 개선 및 트러블 슈팅)
- Please reload this page
- [멀티 프로세스][cpu server] 단일 프로세스 → 멀티 프로세스 (병목 개선 및 트러블 슈팅)
- Please reload this page
- [이미지 스타일 변환] Stable Diffusion과 활용 방식
- Please reload this page
- [저품질 사진 필터링] Quality Assessment 방식
- Please reload this page
- [카테고리 분류] CLIP 이미지 임베딩과 텍스트 임베딩이 따로 군집을 이루는 이유
- Please reload this page
- [트러블 슈팅] artifact registry에 이미지 등록
- Please reload this page
- [트러블 슈팅] from diffusers import FluxKontextPipeline 불가
- Please reload this page
- [트러블 슈팅] GPU 서버 멀티 프로세스 도입 (gunicorn cuda does not allow reinitialize)
- Please reload this page
- [트러블 슈팅] Redis too many connection
- Please reload this page
- [트러블 슈팅] 레디스 일부 요청 임베딩 캐싱 실패 원인 파악
- Please reload this page
- [트러블 슈팅] 멀티프로세스 도입 테스트 Value is not an object: null
- Please reload this page
- [트러블 슈팅][하이라이트 사진 추천] Linear Probing 학습 과정
- Please reload this page
- [트러블슈팅] torch 2.5.1 버전 설치 문제
- Please reload this page
- [트러블슈팅] 도커로 서버 실행 시 문제점 파악을 위해 로깅 추가
- Please reload this page
- [하이라이트 사진 추천] CLIP knows image aesthetics 논문 요약
- Please reload this page
- Aesthetic Score 고도화 방안
- Please reload this page
- AI CPU SERVER 인스턴스 검토
- Please reload this page
- AI GPU SERVER 인스턴스 검토
- Please reload this page
- AI 컨벤션 상세 페이지
- Please reload this page
- aiohttp signedURL 테스트 및 미반영 결론 도출
- Please reload this page
- API 명세서
- Please reload this page
- API 설계 ‐ 1단계 과제
- Please reload this page
- Backend Wiki
- Please reload this page
- BE 테크스펙(Album)
- Please reload this page
- BE 테크스펙(User)
- Please reload this page
- bitsandbytes 작동 원리
- Please reload this page
- CDN?
- Please reload this page
- CI CD 최적화
- Please reload this page
- CLIP 모델 활용 플로우
- Please reload this page
- CLIP 성능 테스트
- Please reload this page
- CLIP 최적화 계획과 기대 성능 지표
- Please reload this page
- Cloud 1차 릴리즈 ‐ 단일 인스턴스 기반 MVP
- Please reload this page
- Cloud 2차 릴리즈 ‐ Docker 기반 3‐Tier
- Please reload this page
- Cloud wiki
- Please reload this page
- Cloud 과제 1단계: Big Bang 방식 수작업 배포
- Please reload this page
- Cloud 과제 2단계: CI(지속적 통합) 파이프라인 구축
- Please reload this page
- Cloud 과제 3단계: CD(지속적 배포) 파이프라인 구축
- Please reload this page
- Cloud 과제 4단계: Docker 컨테이너화 배포
- Please reload this page
- Cloud 스프린트 10단계 로드맵
- Please reload this page
- CORS 검토
- Please reload this page
- CPU GPU 사용량 로깅(in GCP)
- Please reload this page
- CPU ‐ GPU 연동 부하테스트
- Please reload this page
- CPU 교체 후 테스트
- Please reload this page
- CPU‐Heavy 코루틴 테스트
- Please reload this page
- def → async def 변환
- Please reload this page
- Embedding 세마포어 설정 후 테스트
- Please reload this page
- ERD 설계
- Please reload this page
- face recognition 병목 요소 식별
- Please reload this page
- face recognition 최적화 계획과 기대 성능 지표
- Please reload this page
- face_recognition 모델 활용 플로우
- Please reload this page
- face_recognition 성능 테스트 결과
- Please reload this page
- face‐recognition 모델 선정 이유
- Please reload this page
- Fast API Server Architecture
- Please reload this page
- Fast API 사용 이유
- Please reload this page
- FastAPI Uvicorn 동작 원리
- Please reload this page
- FastAPI server wiki
- Please reload this page
- FastAPI 동작 원리를 활용한 전략
- Please reload this page
- FastAPI 로깅
- Please reload this page
- FastAPI 요청 처리 흐름
- Please reload this page
- FE‐ 기술 스택 선정
- Please reload this page
- FE‐User flow chart
- Please reload this page
- FLUX.1 Kontext (dev) 모델
- Please reload this page
- Frontend Wiki
- Please reload this page
- Git Convention
- Please reload this page
- GPU 사용 최적화를 위해 개선한 코드
- Please reload this page
- GPU 서버에서 이미지 로딩 및 디코딩 최적화 시도
- Please reload this page
- gpu 서버에서 인물 분류 속도 최적화 시도
- Please reload this page
- LangChain 기반 멀티스텝 AI 구현 검토(4단계 과제)
- Please reload this page
- MCP 활용 설계
- Please reload this page
- Multi Stream 도입 테스트
- Please reload this page
- Multi Stream 재도입 테스트
- Please reload this page
- MVP에서의 요청 과부하 대응 방안
- Please reload this page
- Product‐Backlog
- Please reload this page
- Python GIL 해제 연산의 병렬 처리 구조
- Please reload this page
- Python 성능 최적화(feat. HYPERCONNECT Tech Blog)
- Please reload this page
- Pytorch 추론 가속 기법 정리
- Please reload this page
- Pytorch 추론 가속 기법 테스트 계획
- Please reload this page
- RAG 적용 설계(5단계 과제)
- Please reload this page
- Redis 도입 근거 및 예상 테스트 시나리오
- Please reload this page
- Release‐Planning
- Please reload this page
- Roadmap
- Please reload this page
- semaphore 설정
- Please reload this page
- Service Planning
- Please reload this page
- Sprint 2 Review and Retrospect
- Please reload this page
- Sprint 3 Planning
- Please reload this page
- Sprint 4 planning
- Please reload this page
- sprint 4 review and retrospect
- Please reload this page
- sprint 5 planning
- Please reload this page
- sprint 6 planning
- Please reload this page
- sprint 7 planning
- Please reload this page
- sprint 8 planning
- Please reload this page
- Sprint Backlog
- Please reload this page
- Sprint Planning
- Please reload this page
- sprint1 retrospect
- Please reload this page
- Sprint2 Planning
- Please reload this page
- sprint3 review and retrospect
- Please reload this page
- stable diffusion 모델 활용 플로우
- Please reload this page
- stable diffusion 병목 지점 식별
- Please reload this page
- stable diffusion 성능 테스트 결과
- Please reload this page
- stable diffusion 최적화 계획과 기대 성능 지표
- Please reload this page
- torch.set_num_threads 제한 후 테스트
- Please reload this page
- TroubleShooting && Test
- Please reload this page
- V1 로직
- Please reload this page
- V1 로직(CLIP Prompt Embedding)
- Please reload this page
- V1 로직(CLIP 활용)
- Please reload this page
- V1 로직(Linear Probing)
- Please reload this page
- V2 로직(OpenCV Laplacian Variance)
- Please reload this page
- V2 로직(OpenCV 활용)
- Please reload this page
- V2 로직(계층 구조 도입)
- Please reload this page
- Vision
- Please reload this page
- yolo arcface
- Please reload this page
- 개발 세팅
- Please reload this page
- 고려 사항
- Please reload this page
- 구조 요약
- Please reload this page
- 기술 검토 및 선정
- Please reload this page
- 기술 도입 전 부하테스트 및 VT , WebClient 도입 비교
- Please reload this page
- 도커 기반 실행
- Please reload this page
- 도커 없이 멀티프로세스 cgroup 사용
- Please reload this page
- 로깅을 통해 태스크별 소요 시간 파악
- Please reload this page
- 멀티 프로세스 & 레디스 터미널 명령어 정리
- Please reload this page
- 메시지 큐
- Please reload this page
- 메시지 큐 도입 시나리오
- Please reload this page
- 모니터링 대상 지표 목록과 수집 시각화 방법
- Please reload this page
- 모델 API 설계
- Please reload this page
- 모델 검토 및 추론 성능 최적화
- Please reload this page
- 배치 처리 도입 테스트
- Please reload this page
- 병렬로딩 및 버킷 인증&연결 사전 처리 후 부하 테스트 진행
- Please reload this page
- 병목 요소 식별
- Please reload this page
- 병목지점 식별 테스트(Embedding 생략)
- Please reload this page
- 부하 용량 변경 및 인스턴스 유형 변경 후 부하 테스트 진행
- Please reload this page
- 비동기 구조 개선
- Please reload this page
- 사용한 인프라 기술 및 서비스 명세
- Please reload this page
- 서비스 아키텍처 모듈화
- Please reload this page
- 성능 테스트 결과 요약
- Please reload this page
- 성능 테스트 설계
- Please reload this page
- 세마포어 대신 Task Queue 도입 테스트
- Please reload this page
- 세마포어 위치 테스트
- Please reload this page
- 스레드 분리(run_in_executor)
- Please reload this page
- 스타일 변환 gpu 사용률 체크(코랩 환경)
- Please reload this page
- 스타일 변환 리소스 사용률 테스트
- Please reload this page
- 스타일 변환 후보 모댈 개요 및 비교, 선정 이유
- Please reload this page
- 실행 스크립트 정리
- Please reload this page
- 왜 Multi Stream이 효과가 없는가
- Please reload this page
- 요청 병렬 처리 후 테스트 진행
- Please reload this page
- 우리 서비스에 메시지 큐 도입이 적절한 이유
- Please reload this page
- 이미지 업로드 성능 비교 – Multipart vs Presigned URL GCS
- Please reload this page
- 이미지 임베딩 CLIP 모델 선정 이유
- Please reload this page
- 이미지 임베딩 성능 테스트 결과
- Please reload this page
- 이미지 전처리 과정 개선
- Please reload this page
- 이미지 처리 모델 개요(CLIP, RAM)
- Please reload this page
- 이미지 처리 모델 성능 테스트(CLIP, RAM)
- Please reload this page
- 이미지 처리 모델 최적화 전략
- Please reload this page
- 이미지 처리 컨테이너(임베딩, 태깅, 중복 필터링)
- Please reload this page
- 인덱싱 도입 결과 및 DB 마이그레이션 적용기
- Please reload this page
- 인물 분류 리소스 사용량 테스트
- Please reload this page
- 인물 분류 성능 향상 (DBSCAN vs HDBSCAN 비교, 클러스터링 후처리 적용)
- Please reload this page
- 인물 분류 후보 모델 개요 및 비교
- Please reload this page
- 인물분류 api 명세 수정
- Please reload this page
- 제안한 인프라 모니터링 설계가 팀 서비스의 향후 성장과 안정적 운영에 어떻게 기여하는지에 대한 설명
- Please reload this page
- 초기 부하 테스트 진행
- Please reload this page
- 최대 실행 스레드 수 제한(max_workers)
- Please reload this page
- 최종 통합된 서비스 아키텍처 다이어그램 및 설명
- Please reload this page
- 추가적으로 개선할 사항
- Please reload this page
- 추론 시간 부하테스트
- Please reload this page
- 추후 도입할 사항들
- Please reload this page
- 카프카 도입 설계
- Please reload this page
- 컨벤션
- Please reload this page
- 코루틴에서 CPU‐Heavy 태스크를 별도 스레드로 분리 개선 후 테스트
- Please reload this page
- 쿠버네티스 환경
- Please reload this page
- 태스크별 리소스 제한 방안
- Please reload this page
- 테크스펙(Feed)
- Please reload this page
- 하이라이트 사진 추천 후보 모델 검토
- Please reload this page
- 하이라이트 선정 CLIP 모델 선정 이유
- Please reload this page
- 하이라이트 스코어에 CLIP을 활용할 수 있다?
- Please reload this page
- 확장성을 고려한 배포 아키텍처 다이어그램
- Please reload this page