Case study: Design twitter

Feb. 23, 2020

Introduction

It was my mock interview as a system design interviewer. I have to push myself hard to perform better as an interviewer. I had chance to interview an engineer with strong coding skills, since I am not sure how good he is in terms of system design, I also asked a coding question to double check. I like to write a short case study on this mock interview.

Case study

I will add more detail later.

	Design twitter, 1000 users,
	- users : tweet, follow people, and have followers
	users:
	userId(twitterId) : Name, location, email address
	tweets table:
	tweetId -> userId, content of tweet(text), time
	Likedtweets
	tweetId -> userId, time
	Retweets
	tweetId -> retrweeUserId, time
	FollowTable
	user1->user2
	user2->user3
	user3->user1
	POST CreateUser(email, name, password) OK (200)
	UpdateUser()
	Login(email, password) -> return auth token OK (200)
	POST WriteTweet() -> OK(200)
	DeleteTweet(tweetId)
	Retweet(tweetId)
	LikeTweet(tweetId)
	GET GetTweets(int GetNext) JSON {tweets}
	Post FollowUser
	TwitterService
	CAP theorom
	push message onto serach service queue with tweetID
	- analysis of words and hastags that are in tweet
	- go ahead and perform some sort of ML model on tweet
	- store searchable word -> tweetId
	- intwerviewing.io -> 50 tweets
	- system desing - > 30 tweets
	- rolling average of the last 60 minutes of trending stuff
	- we can start fresh at the start of every hour and count things that are trending
	Search Service, reads the tweet, parses out hashtags, words and word/hashtag -> tweetId
	trending -> count
	celebrity, topic, hashtag
	- availability over consistency
	show case your design:
	celebrity with milions of followers, tweet, notification to millions - bottleneck, network, cache, push/pull, search, ...
	- user A has 20M followers
	we're doing push? then we end up with 20M requests sent down
	- push down to users who phave been active in last 24 hours
	- periodically prepopulate feed for users
	Database scaling
	- shard based on userID consistent hashing
	- cluser of databases (read replicas) (master slave replication, where we're opting mostly for availaiblity) - async replication
	cache
	- search service (trending)
	- tweet itself
	- tweetID does exist in chace? yes, get from cache, no, hit database
	- cache users and their tweets
	database, storage - build index,
	- transaction
	deposit
	withdraw
	- consistency
	storage design - cloud storage
	- balance binary search tree - B-tree
	- transaction - log -
	log-structure merge tree -
	CA P
	-
	- large data stream, integer,
	1, 2, 3, 4, 5, 6, -> six, 3, 4,
	3 + 4/ 2
	1, 2,3,4,5 , odd number, median -
	assume that you store all integer into a few data structure, order can be lost, it is ok.
	Add(int, ) -> do not require ->
	findMedian() -> O(1) time complexity
	minheap, maxheap, keep size of two heap max diff < 1
	determine how to get median,
	LSM-tree -> video, Microsoft, merge, Apple employee ->
	Log -> append -> merge <- backward ->
	client-> load balanace-> reverse proxy -> twitter, socail profile servcice -> cache -> storage -> database
	long polling -> ...
	booking in United statemet
	diagram
	- hand
	https://gist.github.com/jianminchen/ff21b0a1cad67216376999512dd1b696
	booking
	design twitter ->
	URL -

view raw Design twitter hosted with ❤ by GitHub

Here is the link.

Actionable Items

The interviewee is Microsoft SDE II. I should learn from his performance. Good analysis and very good communication skills.