架構師訓練營 week5 總結

Distributed cache architecture

Cache:

• CPU cache

• OS cache

• DB cache

• CDN cache

• Reverse proxy cache

• Frontend cache

• Application cache

• Distributed object cache



What items affect the "cache hit ratio"

• How many cache keys

• How many memory

• TTL of cache object



Why do we use cache?

• faster

• We usually save final result in the cache, we don’t need to compute again

• Reduce DB, disk, network loading



We use cache on

• Frequently read data

• Complex computing result

• Read more write less

• Low consistency requirements



We don’t use cache on

• Date which is updated frequently (read/write < 2)

• No hot data

• High consistency requirements



缓存雪崩 - Cache Avalanche

A lot of request go to database directly and cause database down



How to prevent?

• Cluster 

• redis cluster

• Distributed expired time

• setRedis(Key, value, time + Math.random() * 10000)；

• Service degradation



缓存穿透 - Cache Penetration

Same as Cache Avalanche, only different is the scale.

Cache Penetration => fewer

Cache Avalanche => a lot



How to prevent it?

• Setup null/default value

• Short TTL

• Bloom Filter



缓存击穿 Cache Breakdown

A lot of requests go to one cache key => kill that cache service



How to prevent it?

• Update lock logic

• Asynchronous update

• Cache warm-up



Redis

• support complex data structure

• support asynchronous

• Maser/slave high availability

• Cluster and share nothing mode 



Message queue and asynchronous structure

• Point to point 



• Publish and subscribe



What is the benefit of the message queue?

1. Asynchronous, increase performance

2. Good scalability

3. Failure isolation

4. Decouple



Event-Driven structure



Load balancer

• HTTP

• DNS

• Reverse proxy

• IP

• MAC address



Load balancer algorithm

• Round Robin

• Weighted Round Robin

• Random

• Least Connection

• Weighted Least Connection

• Source IP Hash