系统设计
第一步理解系统设计需求，需要明确系统所用于何种商业目的，要求的功能&技术，来成功定义面试官内心的“标答” 【功能性需求】
第二步明确所设计系统需求的资源, 例如分析QPS 【非功能性需求】
latency sensitive的场景，要避免使用msg queue
第三步画出关系清晰的架构图，搞清楚service怎么划分
第四步设计数据结构与存储、核心子服务、接口等，接上database
技术选型：SQL vs No-SQL，Sync VS Async，RPC VS MQ等技术选择
第五步明确所设计系统的可扩展性、容错性、延迟要求等
扩展: 加缓存，数据库读写分离，数据库 sharding 等等
瓶颈: 一般都在数据库
第六步解决缺陷并处理可能遇到的问题
Template
Functional requirement
NonFunctional requirement
Capacity planning
Diagram + API 【接口和业务entity】
DataSchema + Scale
Monitoring/Metrics
常见面试问题
Design a URL shortener (e.g. Bitly)
Design a video watching website (e.g. YouTube)
Design a chatting service (e.g. Telegram, Slack, Discord)
Design a file sharing service (e.g. Google Drive, Dropbox)
Design a ride sharing service (e.g. Uber, Lyft)
Design a photo sharing service (e.g. Flickr, Pinterest)
Design an e-commerce website (e.g. Amazon, eBay)
Design a jobs portal (e.g. LinkedIn, Indeed)
Design a web crawler (e.g. Google)
Design AI自动写作系统设计
Design Auction system
Design search autocomplete system
Design large scale rate limiter
Design Netflix
Design Amazon inventory system
Design Venmo
Design large scale notification platform
Design AirBNB platform
Design Ticket master
Design high velocity bank account platform
Design top10 scorersina large scale mobilegame
Design large scale real time chat platform
Design electric bike rental platform
Design grocery store order processing system
Design website building platform
Design giftcard system
Design large scale devices location tracker
auto complete: trie database
瓶颈，scale, tradeoff
High level design
client -> load balancer -> web service/API -> memory cache -> DB
web/mobile -> HAproxy/ELB -> API Gateway -> Nginx server/Kube deployment/REST -> Redis -> MySQL
client -> reverse proxy -> web service -> message queue -> application server -> cache -> DB
问题
How to do large scale batch processing
How to do model serving readiness
How to do model rollout
两个service要互相发消息，怎么解决
高并发系统
缓存
降级
限流
计数器、漏桶和令牌桶
负载均衡算法
轮询、加权轮询、随机算法、一致性Hash
消息队列
解耦，异步处理，削峰/限流
一致性
热点数据处理
基础
push and pull
consistent hashing
event sourcing
paxos and raft
cache
redis
数据库规范化(Normalization)
Normalisation Form(NF)，其中包括第一范式、第二范式、第三范式、第四范式以及第五范式(1NF、2NF、3NF、4NF、5NF)
SQL provides ACID transaction
网络基础
Restful API
HTTP/HTTPS
RPC
SQL and NoSQL
需要支持transaction和join的，需要用SQL 需要high TPS和灵活schema的，用NoSQL
SQL(ACID)
consistency
structured data (fixed schema)
transactions
joins
NOSQL
high performance
unstructured data (flexible schema)
availability
easy scalability
scale
load balancer
Horizontal scaling
caching
database sharding(partition)
horizontal scaling, sharded by userId, caching, etc
Reference
Web Application and Software Architecture 101
Last updated