AWS - Elastic Cache

Posted on |

Elastic Cache

  • Managed in-memory cache service
  • key value stores
  • sub-mili sec latency to data
  • Redis / Memcached data store options
  • Multi-AZ capability
  • increase application throughput : 20M reads /sec ,4.8M write /sec
  • Scaling DB layer is much more expensive compared to scaling caching layer

Compare the two options

  • depending on project language & framework support
  • Redis feature is superset of Memcached

Memcached Vs Redis

  • Store json
    • in Memcached , use serialized string
    • in Redis , use hash

Memcached (mem cache d) store Option

  • Free and opensource
  • Object max size 1MB
  • Total max size 7 TiB
  • No persistence ; easy to adding node

Redis Store option

  • Free and open Source

  • Object max size 512M

  • Total max size 3.5 TiB

  • persistence; read replica

  • Support Notification from Redis Pub/Sub channel

  • Support more data structures including : bitmaps, hyperlogs,GeoSpacial command; geo indexes with radius queries ; and also those supported by Memcached

  • Support auto sorting of data

  • Support HA and Failover

    • Failover is automatic, will chose the read replica with lowest latency and will switch the dns automatically

  • API provided to query all read replica endpoints

  • Sharding : 16384 sharding slots (automatic client sharding; developer must use Redis Cluster Client )

  • Standard Redis use case

  • Leaderboard

  • Counters ; like & dislike

Redis key feature

  • Set : A group of objects with key. Each value has a unique key which will be unique.
  • Sorted Set: based on set but the keys are sorted
  • List: Push Pop ; like queue
  • Hashes: key

Updating cache

  • Database triggers or Database via Lambda can be used to trigger update cache
  • application can be used to trigger cache updates

Caching Strategies

  • Lazy Loading: load from cache if missed, load from DB and update cache
  • Write Through: write DB and then write cache
  • Adding TTL (memcache support seconds , Redis support both seconds or milliseconds)

Elastic Cache vs Cloud Front
Elastic Cache — Designed to cache in memory , caching query (dynamic cache)
CloudFront — designed at Edge location , SDN

A Node is the smallest building block of an ElastiCache deployment.
https://docs.aws.amazon.com/AmazonElastiCache/latest/mem-ug/WhatIs.html

Redis append-only files (AOF) ： Redis一种持久化的方式
Redis Multi-AZ with Auto Failover

Use Scenarios

  • Session Replication
  • IoT Device Data: make use of the 4.x million write / second capability
    IoT with Elastic Cache

IoT with Elastic Cache - another use case

IoT with Elastic Cache - use case 3

IoT with Elastic Cache - use case 4

  • GEO advertising (Redis support GeoSpacial )

Architecture

  • Master Node + read replica (also as fail over node)
  • Cluster mode : data will be sharded

how to use Cluster Mode

  • 16384 hash slots for keys ; slots are distributed across cluster into shards

  • MUST USE Redis Cluster Client

    • the client will store a map with shards , like
    1
    2
    shard1  = slot 1-3276
    shard2 = slot 3277- 6553

  • a cluster can have max=15 shards; max 5 replicas

How to migrate

  • Migrate from single server mode to cluster mode

Create new cluster mode cluster --> Snapshot old one --> restore snapshot on new mode cluster–> update client --> delete old one

  • Migrate from cluster mode shard N to shard M

Create new cluster mode cluster --> Snapshot old one --> restore snapshot on new mode cluster–> delete old one

Tuning

  • set reserved memory to 30%
  • Swap should be 0 or very low ; if not scale up
  • read replica deploy to different AZ with master
  • take snapshot from read replica
  • Odd Number of shards (support even number but not recommended)
  • Russian Doll Caching : The technique of nesting fragment caches to maximize cache hits is known as russian doll caching. By nesting fragment caches, it ensures that caches can be reused even when content changes.
  • Thundering Herd : sudden rise of large # of cache miss --> Spike in database load
    • App startup --> script to populate cache
    • adding/removing nodes --> Graduate scale nodes
    • key expiration (TTL ) --> Randomize the title
    • out of cache memories --> monitor cache evictions
  • Failover requires updating DNS CName ; so careful with application caching the CName

Caching Strategies with Database

Option 1: Caching the row
Option 2: Caching transformed Object like json string
Option 3: Caching serialized application object
Option 4: Caching as Redis specific data type (like sorted set)

Hands on

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
https://github.com/mikelabib/elasticache-memcached-php-demo

Step 1: Install php, apache, memcache client on the server.
e.g. yum install php httpd php-pecl-memcache

Step 2: Update /etc/php.ini file params:
session.save_handler = memcache session.save_path = "tcp://elasticache-memcache-node1-endpoint:11211,tcp://elasticache-memcache-node2-endpoint:11211, etc."

Step 3: Configure php.d/memcache.ini param values:
memcache.hash_strategy = consistent
memcache.allow_failover = 1
memcache.session_redundancy = 3

Step 4: Restart httpd
e.g. etc/init.d/httpd restart

Cost

  • CrossAZ data transfer is free

Reference

https://youtu.be/e9sN15a7utI

Deep Dive
https://youtu.be/zmDUDSYnAv4

Code Sample

Reward Makes Perfect
0%