2018_ifkakao_cloud_andrew_final

Size: px

Start display at page:

Download "2018_ifkakao_cloud_andrew_final"

찬성 영
5 years ago
Views:

1 () 공용준 9. )( ( )

2 ) K P I ADII EI A FIN 7 B C F ) CLIK IK I K G M KI ) DN M 5KI )DIN C 1IK, M L )A ) MC A 5KI LLIK IK CA G KI MCI F 0 IKG MCI )A CA M ) 1 E I IK FIN 5 KM 2 7 FEL 6 F F 2I F K PCMB IK B LMK MIK G K FIN L ( I LM E L EIK.NFF KINM L MPIKE PCMB FC N M 6, PCMBINM 6, I LM E IN K Supervise d, Korean edition Korean Edition. LM )A M CLIK ( M 7IE I IGGCMM EIK

4 ( ) ) ( Programmable Resource Management 2016 What is Resource? Programmable Resource Life Cycle Management 2017 What is Life Cycle? Programmable Service Management 2018 What is Service?

5 CMMI Model Capability Maturity Model Integration developed by CMU ITIL Service Delivery Strategy Service Support Strategy Service Security Strategy

6 ) ) ( ( Maturity Targets Cross functional Cells & Teams Orchestrated Deployment Fully automatic acceptance tests Trend Reports kakao cloud

7 if (kakao): IaaS What is the purpose of doing IaaS? CMM0 CMM1 legacy output: ITF self service Dev resource output: krane (openstack cloud)

8 ) ) ))( ) Inhouse Auth. keystone Authentication KRANE Service Topology IMS DNS AEZORO BareMetal TBD VM LB DB

if (kakao): KRANE(The Tech.) Data Center Level Scalable Network Neutron Floating IP 192.1.1.202 Compute node1 neutron-l3-agent Switch Namespace linux bridge IP:10.10.100.

9 if (kakao): KRANE(The Tech.) Data Center Level Scalable Network Neutron Floating IP Compute node1 neutron-l3-agent Switch Namespace linux bridge IP: /32 IPTable DNAT connected Veth pair empty empty Gateway Host dest /32 to Route dest to eth Compute Node Router Routing Table Default GW eth1 Host Route dest /32 to dest to connected eth0 eth deactivate VIP global name space LB vm neutron-dhcp-agent neutronlinuxbridge-agent dhclient vm agent LB vm

10 Cloud CMMI2-MaaS What is the sole purpose of doing MaaS? CMM0 CMM1 CMM2 legacy output: cloudtf self service Dev resource output: krane (openstack cloud) limited Prod resources output: kemi (MaaS)

키워드빈도기반알림 전사 Elastic Search 서비스 KEMI Log 로그데이터 ETL*

11 ) () ) IMS 서비스정보 KEMI Stats 메트릭데이터 ETL* 시계열 DB 전사 HBASE 메트릭정보수집 KRANE 일반시스템 KEMI Alert 로깅정보수집 임계치기반알림 키워드빈도기반알림 전사 Elastic Search 서비스 KEMI Log 로그데이터 ETL* 실시간 DB 장기 DB 전사하둡 Secure DB ETL: Extract Transform Loading

12 .(.( )

13 .(.( ) Lambda Architecture RealTime KEMI Log Tailer (CLI) App Log KEMI Log Aggregator KEMI DIKE Container Log every 5m every 5~15m Batch

14 if (kakao): MaaS (The Key.) Monitoring Abstraction curl -XPOST -d '{ "start": , "end": , "queries": [{ "metric": "-", "aggregator": "max", "downsample": "15m-max", "filters": [{ "filter": "data.load.avg_5m", "groupby": true, "tagk": "metric", "type": "literal_or" }, { "filter": "dkos-marathon", "groupby": true, "tagk": "host", "type": "literal_or" }] }]' Connection with IMS(CMDB) by Events SAURON API Retrieval Standardization RBAC API limiting

15 What is the sole purpose of doing CaaS(Container As A Service)? CMM0 CMM1 CMM2 CMM3 legacy self service Dev resource limited Prod resources Automated CloudUsage output: cloudtf output: krane (openstack cloud) output: kemi (MaaS) output: DKOS (CaaS)

16 ) () ) 사내 LDAP keystone 사용자정보 KRANE 서비스정보 dns 정보 IMS 메트릭정보수집 설정정보 KEMI Stats 설정정보 메트릭데이터 ETL* 시계열DB 전사 HBASE AEZORO KEMI Alert 임계치기반알림 키워드빈도기반알림 물리머신 가상머신 데이터베이스 컨테이너 로드밸런서 public cloud 일반서비스 로깅정보수집 KEMI Log 로그데이터 ETL* 실시간 DB 장기 DB Secure DB 전사 Elastic Search 전사하둡 인증, 컴퓨팅 API 메트릭로깅 API Container Orchestration Container Scheduling Component Platform 메타케이지 DKOS D2Hub 사내 Github

17 .(.( ) Network Service Architecture: LB(L4)-LB(L7)-Container DKOS Worker DKOS LoadBalancer DKOS Worker 물리장비 LoadBalancer DKOS Worker DKOS LoadBalancer SSL Offloading L3DSR(Direct Server Return)

18 .(.( ) Autoscale KEMI Alert Setup DKOS Autoscale Group Setup

19 .(.( ) Autoscale Result

dns 정보 IMS 메트릭정보수집 설정정보 KEMI Stats 설정정보 메트릭데이터 ETL* 시계열 DB 전사 HBASE

로드밸런서 컨테이너메트릭 컨테이너로깅 public cloud 일반서비스 로깅정보수집 KEMI Log 로그데이터 ETL*

20 if (kakao): DKOS(The key.) DKOS is connected to everything 사내 LDAP keystone 사용자정보 KRANE 서비스정보 dns 정보 IMS 메트릭정보수집 설정정보 KEMI Stats 설정정보 메트릭데이터 ETL* 시계열 DB 전사 HBASE AEZORO KEMI Alert 임계치기반알림키워드빈도기반알림 물리머신 가상머신 컨테이너 인증, 컴퓨팅 API 데이터베이스 로드밸런서 컨테이너메트릭 컨테이너로깅 public cloud 일반서비스 로깅정보수집 KEMI Log 로그데이터 ETL* 실시간 DB 장기 DB Secure DB 전사 Elastic Search 전사하둡 컨테이너오케스트레이션 컨테이너스케줄링 DKOS 머신러닝플랫폼 알람, 오토스케일조건설정 D2Hub 메타케이지 사내 Github

21 if(kakao) Product & Service in Kakao Cloud Category Target Product Description Connected internal service Authentication Compute Project&Role based Authentication KEYSTONE 프로젝트 / 롤기반사용자인증및권한관리내부 LDAP 와연동 Virtual machine KRANE 가상머신제공가상볼륨, 가상머신기반 DB/LB 제공 Autoscale 기능준비중 Container DKOS 컨테이너 orchestrator(marathon) 제공. 컨테이너용 Loadbalancer 제공컨테이너기반 Machine learning framework 제공 (Spark, GPU) IMS, *AEZORO, *Meta-kage 연동 *Tenth 연동 KRANE, KEMI, AEZORO 와연동 Management Monitoring KEMI Stats 전사메트릭데이터 (cpu, memory, disk) & custom (application, component) 데이터수집및분석 Logging KEMI Log 전사로그데이터 (syslog, app, container) 로그수집 / 저장 / 분석 Alert KEMI Alert KEMI 데이터기반알람 ( 톡및 url 전송 ) *IMS 와연동 sauron 연계 IMS 와연동공용하둡, Hbase 연동 Developer tools Package Repository Container Registry / Builder repo.kakao.com OS 패키지, Python/Ruby 패키지제공 D2Hub 컨테이너용이미지제작및등록 Keystone 연동 DNS as a service AEZORO 가상머신 / 컨테이너기반 DNS 자동등록 / 삭제내부 DNS 연동 *AEZORO( 사내 DNS as a Service API) *IMS(Infra Management System) *Meta-kage(swift[ 오픈스택스토리지 API] 용 kage) *Ceph-Tenth: Ceph API 호환 Tenth

22 ) ( ( IT Services Service Repo. Service topology Notification setup Declarative Scheduling Analytics Initial Setup Event / Alert Change Data Cloud API Platform Centralized Measuring API Abstract Resource API Automated Deploying API Automated Orchestration API DataCenter Contol/Data plane

23 What is cloud native? (by CNCF) Containerized Micro-service oriented Dynamically orchestrated

24 Little bit more specific Cloud Native Applications Characteristics Micro-services Health Reporting Telemetry Data Resiliency Declarative First Ask Why we need all these? Dynamically Orchestrated Platform

25 if (kakao): Characteristics to Requirement C. N. Applications Characteristics Microservices Health Reporting Telemetry Data Resiliency Declarative Platform Requirement Isolation Resource Allocation /Scheduling Service Discovery Monitoring/Logging Metric Aggregation Debugging and Tracing

26 - - - What is the sole purpose of doing Cloud Native Platform? CMM0 CMM1 CMM2 CMM3 CMM4 legacy self service Dev resource limited Prod resources Automated CloudUsage Integrated Service Platform output: cloudtf output: krane (openstack cloud) output: kemi (MaaS) output: DKOS (CaaS) output: 9rum (C.N.P)

27 ) (

28 (. ) ) Cloud Native Applications Characteristics Platform Requirement Isolation Microservices Health Reporting Telemetry Data Resiliency Resource Allocation and Scheduling Service Discovery Monitoring/Logging Metric Aggregation Declarative Debugging and Tracing

29 if (kakao): 9rum (The key) Integrated SSO. KRANE DKOS KEMI SSO IMS (CMDB) ERP Authorization QUOTA RBAC Keystone DB Project authentication Inhouse auth (Hello-MIS)

30 if ( kakao ) : some numbers Virtual to Physical ratio (%) Container slave ratio over Cloud (%) Indexing Data (TB/Day) 80 0 Expense $MM/Yr (compared to public cloud, excluding traffic/storage) License Expense

31 . / / 2 (. ( / / /. ). ( / / )

digital_transformation_software_engineering

digital_transformation_software_engineering Digital Transformation and Software engineering Andrew Yongjoon Kong Who am I Andrew. Yongjoon kong Cloud Technical Advisory for Government Broad Cast Agency Adjunct Prof. Ajou Univ Korea Data Base Agency