流言终结者：Rust 真的能成为爆款吗？.pdf

1、We rewrote our platform in Rust.Was it worth it?Myth Busters:Is Rust A Slam Dunk?Ramya K|It BeginsEarly stage startup that offers a real-time data platform comprised of serverless infrastructure building blocks like caching,event bus and durable storage.Launched in 2022 with the first service Moment

2、o Cache:a Key-Value caching service.https:/ Level ArchitectureControl PlaneCreate/Read/Update/Delete actions on top level entities like cachesManage data plane infrastructureRouting Layer(Data Plane)Authentication and authorizationRoute requests to where the data livesStorage Layer(Data Plane)Run st

3、orage engines that manage data for top level entitiesArchitecture ComponentsMinimized time to launchEngineering team profile spanned the spectrum for tech stack experience.Kotlin=Safe and battle-tested choiceTech Varies by Storage EngineTechnology Stack v0It works.How good is it?Invest in observabil

4、ity and automated performance tests using rpc-perf.Identify representative customer workloads for cache.Single Node Baseline established for a single c6i.2xlarge routing node20K TPSKey-Value GET/SETGET:SET=1:14KB ItemsMetric=Client End-to-End LatencyP999=4.99 msP99=4.94 msv0 BenchmarkNew Product Req

5、uirement:Support caching data structures(dictionaries,lists,sets)in addition to key-value.Design changesNew storage engine to handle data structuresSupport multiple storage engines.Storage-engine agnostic abstraction needed for storage layer.Product Evolutionhttps:/ Storage service in Rust.Good cand

6、idate for an experimentBrand new componentNo external clientsTech Varies by Storage EngineTechnology Stack v1Is it still good enough?Performance regression for caches using the new storage engineCache workloadIncrease in P999 tail latencyMetricv0v1Client end-to-end p999 latency4.99 ms9.3 msClient en