aslamnotes

elasticache

2 min read

The Basics

  • Managed redis/memcached benefits like RDS
  • Help reduces load off of DBs
  • Makes application stateless, putting state in elasticache
  • Needs, heavy application code changes

DB Cache - Architecture

  • Application looks for cache hit

  • If not, queries DB

  • Writes to cache

  • Session Storage - Architecture

    • The seesion data gets stored on the cache
    • Multiple stateless application can access this state

Cache must have invalidation strategy. Elasticache Node is a network attached fixed size RAM block with a Compute

Redis

  • sharding helps reduce the load off a node, more like RAID0
  • Data can be sharded or replicated for HA
  • Supports failovers

Memcached

  • Data has to be sharded, and it our responisibilty to do so, accross nodes
  • Sharding will be done on the clinet sie

Replication

Cluster mode disabled

  • One primary node, up to 5 replicas
  • Async replication
  • Primary rw, other r
  • Multi AZ enabled by default Cluster mode enabled
  • Data is sharded partially
  • Each shard behaves like a node group, one primary, upto 5 replicas
  • Upto 500 nodes per cluster

500 shards - 1 replica (500 nodes in total) .. .. 83 shards - 5 replica (413 nodes in total)

Scaling

elasticache-scaling

  • Horizontal, upto 5 read replicas
  • Scaling vertically will create a new node group, then data replication and then dns update Auto Scaling
  • Metric: Elasticache primary engine cpu utiliation metric
  • Add Shards to maintain the metric

Connection

  • Standalone node - One endpoint
  • Cluster mode disabled
    • Primary Endpoing - rw
    • Reader endpoint - r (evently split acroos replicas)
    • NodeEndpoint - r
  • Cluster mode enabled
    • Configuration enpoint - one endpoint for r/w operation
    • NodeEndpoint - For r operation

Sysops Operations

Scale (cluster mode disabled) Horizontal - add upto 5 read replicas Vertical - Update the config, but internally aws creates new node group and replicates the data

Scale (Cluster mode enabled) Horizontal - Supports both online and offline - Resharding: Add or remove shards - Shard Rebalancing: Evenly distributes keysapces across the shards

Vertical - Supports online scaling - Adjust the node types

Metrics

  • Evictions (Memory is overfilled, add more memory or evict LRUs)
  • CPUUtilizations (Overall Host CPU usage)
  • SWAP Usage (Should not exceed 50MB)
  • CurrConncetions

Graph View

Backlinks