working-with-raw-disk-drives-in-kubrenetes-ydbs-experience-nanokuberneteszhi-pin-kan-tan-jydbzha-lang-mo-ivan-blinkov-ydb.pdf

1、Working with Raw Disk Drives in Kubernetes YDBs ExperienceIvan BlinkovIvan BlinkovVP,Product and Open-Source在Kubernetes中使用原始磁盘驱动器YDB的经验2Ivan Blinkov Over a decade of experience in the database management systems(DBMS)development industry Talked with countless DBMS usersand stakeholders to understand

2、 how and why they ended up with a specific solution Worked on a handful of DBMS products,including two open-source ones:3123Nuances of working with raw disk drives in KubernetesWhats YDB and why it works with raw disk drives directly via block devices?Lessons learned along the wayAgendaWhy YDB works

3、 with raw disk drives directly?(without any filesystem)45Mission critical Highly availableData platform Survives AZ plus rack failure without human intervention Seamless upgrades Self-healing Smart SDKs Designed for services with 247 uptime requirements Serializable consistency Adapts to workloads S

4、ecurity features Row-oriented tables(OLTP)Column-oriented tables(OLAP)Topics(persistent queues)Federated queries MultitenancyYDB:Open-Source Distributed SQL Database6 Finance E-commerce Ride-hailing Advertisement Telecom Logistics AI services InfrastructureTypical YDBuse cases72014Started as an in-h

5、ouse infrastructure technology2020Provided as a managed cloud service2021Kubernetes compatibility prototype2022Published to open-source under Apache 2.0 license2023First production clusters running in KubernetesSummary of YDB history YDB high-level architectureDistributed Storage8TabletDistributedTr

6、ansactionsgRPC ProxyCluster InterconnectYDBSDKUserApplicationQueryProcessorTabletTabletTabletLegendStorage LayerClient SideCompute Layer9LatencyThroughput Hardware capacity Efficient algorithms Transaction contention No unnecessary overhead Network hops required for distributed transactions Caching