Google IO 21 Extended double S- Serverless k8s with cloud run

Google IO 21 Extended double S- Serverless k8s with cloud run

발표 보기 : https://www.youtube.com/watch?v=LNVs_iKQLzg

Transcript

1. Architecture Serverless K8s with Cloud Run
2. 이성우 MegazoneCloud Google Cloud Platform Team
3. GCP외 다양한 Cloud Platform을 섭렵하려 합니다 기술 블로그를 운영하고 있습니다 “Nangman14” Google it! Cloud native 기술에 관심이 많고, 지식을 공유하고 싶습니다
5. Cloud Run
6. 1. K8s + Serverless?
9. Resource request Scheduling Persistent volume claim Taint Toleration
14. 2. Cloud Run for Anthos
15. Cloud run fully managed Cloud run for anthos
16. Cloud run fully managed Cloud run for anthos
17. Knative 17 Serverless kubernetes 오픈소스 프로젝트 • Event-driven 기반 아키텍쳐 • 네트워킹, 스케일링 등 추상화 • 느슨한 결합(약결합) 컴포넌트 • 다양한 Kubernetes 환경 호환성
18. Cloud run for anthos Bringing Serverless to K8s • GKE 환경과 완전히 통합(GKE on-prem, GKE on GCP) • Knative 기술의 장점을 그대로 • Automatic routing , scaling • NVIDIA GPU accelerator
19. Pay-as-you-go pricing • 사용한 만큼만 비용 지불 • 요청이 없다면 사용량도 없음 ◦ 비용 절감 효과 ◦ 리소스 측면에서도 효율적 • GKE와 통합된 Pricing
20. Service • Cloud run for anthos의 메인 리소스 • Automatic scaling feature • Configuration 관리
22. Revision • 다른 버전의 컨테이너 이미지 집합 • Service가 항상 최신 버전으로 라우팅 • Traffic splitting
24. 3. DEMO
25. Smile,Shirt,Product,Natural environment,Coat,Tie... THANKS MARK RUFFALO FOR THINKING GREEN! HE CARES ABOUT OUR ENVIRONMENT! Resizing & Wartermarking Original Labeling Text detecting
26. Image processing pipeline
27. • -> GKE Cluster image -processing GCP resources
28. -> Cloud run for anthos services filter labeler resizer textdector watermarker GCP resources
29. 1 2 3 2 Event-driven architecture Smile,Shirt,Product,Natural environment,Coat,Tie...
30. Running Pods CPU Utilization Scale to 0 feature
31. CPU Utilization Running Pods Scale to 0 feature
32. 30% Traffic splitting 70%
33. 4. Conclusion
34. Loosely coupled modules Translator Text detector
35. Concurrency control
36. Cold start Request latency Peak latency (cold start)
37. Without Cloud run 1. Ingress 2. Service 3. Deployment 4. etc...
38. With Cloud run 1. Service 2. Trigger
39. 요약 Cloud run for anthos Serverless container • 사용한 만큼만 지불 • 효율적인 리소스 사용 • Stateless workload • Scale to 0 (cold start) Less operational burden • 네트워킹 레이어를 추상화 • Automatic scaling, routing • Traffic splitting Bring Knative to GCP • Service, Revision • GCS, Pub/sub 등 GCP서비스와 통합 • GKE 기반 add-on • Loosely coupled architecture
40. 이성우 - MegazoneCloud Thank you! https://nangman14.tistory.com/ kallusis369@gmail.com https://www.ny2cl.org/post/c elebrity-activism https://github.com/meteata mel/knative-tutorial/ Resources