Go Game

Topic: Go Game

Interviewer: FLAAG engineer

Interviewee: FLAAG engineer

Level: L4 (Experienced Individual Contributor)

Requirements

Multi-player game for Go

Each game is between 2 persons

Total 10k people at the same time

Web-based game frontend->backend

Matching pair - similar level

System design

7:05

User journey

Click start game -> request -> matching system

System sending request to another playerB

Game start

Inside game

Player A with black

In turn

Winning condition

System and game, grant score for winner for level up

7:20

API design:

putPeace(user_id, piece_type, ) -> return ACK(ok/error)

heartBeat(user_id)

resumeGame(user_id)

7:42

Diagram

Interviewer: What technology do you want to use?

Interviewee:

Lambda

Explain lambda

What database to use?

NonSQL: Cassandra, DynamoDB

What key and value for database?

Eventual consistency

What’s the cost? QPS

Max throughput:

10k QPS strong reads

3000 QPS per partition

Need 4 partitions

1 user N games

Database design:

7:57

API:

PutPiece(user_id, piece_type, coordinate_x, coordinate_y)

IsSecure(user_id) return error

Pseudo code

How to implement put piece function?

getLatestBoardFromDB(GameID)

Interviewer Comments

Soft skills

Requirement gathering

Make decisions - no need to be optimal, justified rationale

Describe the solution

System design

Vague requirement -

Improve: critical journey

Twitter - system scalability

Architecture diagram

Pseudo code, at right level - L4

Lambda: excellent, general solution, scaling, deployment

To Improve:

Action time - cannot trust client side, people cheat

Try not to fight the interviewer when the interviewer suggested something

Make decision and tradeoff

Heart beat

Javascript - websocket, push notification

Make decision to stick to it, make progress, design around

Key is to get to working design

Websocket: pipe, message type

Client ->server

Server -> client, define client

10k QPS

L4 lambda

L5-L6 under the hood

E.g. kubernetes, pod, application server, instance type, handle how many QPS

2-3, what kind of instance in AWS

1 instance, how many QPS

Order of 1000 times

Project life cycle. 3 methods.

How to handle this extension? How do you push?

How to rollback?

Feedback/comments from the audience

Not official rubrics

Feature flag - widely used in the industry

Justify, when to use, how to use

Audience:

题目 - candidate

Audience

How many server, hardware setting

L5 - how many core, handle simple logic, how many QPS

Audience

DynamoDB - consistency model

Eventual consistency. L5 - need candidate to tell me

Strong consistent

L5 - eventual consistent

Read and write,

Read your own writes

Write - mysql correctness

L5 - candidate - correct solution

Audience:

API call - server calling client