System design

Question 1

db: disaster recovery

Answer 1

1. create a standby DB in different AZ, replicated from master DB
2. using the same DNS to connect to DB
3. if the master goes down, the replica is automatically up to replace

Question 2

db: replica and backup

Answer 2

1. aws RDS backup every 5m

2. can create 5 read replicas: within AZ, cross AZ or cross region

Question 3

aurora: how to connect to read/write to DB?

Answer 3

1. using reader endpoint, writer endpoint

2. aurora auto-routing to correct destination

Question 4

detect data conflicts

Answer 4

1. using versioning vector D([s1, v1], [s2, v2])
2. given 2 vectors if all v values in the one vector >= the other one –> no conflict
   conflict example:
   D1([s1,2], [s2,1]) -> populate D1 from s1 2 save into s2
   D2([s1, 1], [s2,2]) -> populate D2 from s1 1 save twice to s2

Question 5

rds: available network speed between instances

Answer 5

5 Gbps to 25 Gbps depends on the instances (same for both up and down)
home internet: 350 Mbps (both up and down)

Question 6

redis commands for list, set, sorted set

Answer 6

1. sorted set: zadd, zincrby, zrangebyscore, zrangebyrank, zrangebylex, zrembyrank
2. set: sadd, srem, scard
3. list: lpush, lpop, rpush, rpop, lindex, linsert, llen

Question 7

CNAME, A, AAAA record

Answer 7

A: map host to IP
AAAA: same as A but for IPv6
CNAME: host to another host

Question 8

sticky cookies

Answer 8

1. send cookies to client

2. server uses cookies to identify the server behind the LB

Question 9

aws load balancers

Answer 9

1. application load balancer: application layer e.g. HTTP

2. network load balancer: network layer e.g. TCP/IP

Question 10

terms using in the load balancer

Answer 10

1. target group i.e. group of instances

2. health check

Question 11

aws route 53

Answer 11

1. modes: latency, geographical, failover, simple
2. TTL
3. 53 is DNS port

Question 12

forward compatibility

Answer 12

is planning ahead

e. g.
1. reserved keywords in programming languages,
2. build something that is easy for adding new feature

argument:
1. mvp: build enough to release only

Question 13

Kafka consumer and offset

Answer 13

1. consumer is assigned to process messages in 1 or many partitions
2. 1 partition only be handled by 1 consumer
3. after polling messages, consumer will commit to save offset in Kafka so that if re-balancing/restart the delivered messages won’t be delivered again

Question 14

AAA pattern

Answer 14

Arrange, Act, Assert