ElastiCache — Managed In-Memory Cache

Creating a cache starts a real Redis, Valkey or Memcached container; without Docker the same calls are metadata only.

Status: ⚠️ Partial

Quick start

export AWS_ENDPOINT_URL=http://localhost:4566

aws elasticache create-cache-cluster \
  --cache-cluster-id sessions \
  --engine redis --cache-node-type cache.t3.micro --num-cache-nodes 1

aws elasticache describe-cache-clusters --cache-cluster-id sessions \
  --show-cache-node-info \
  --query 'CacheClusters[0].[CacheClusterStatus,CacheNodes[0].Endpoint]'
# once "available": ["127.0.0.1", 63790] on the host
redis-cli -p 63790 ping

A Redis or Valkey cluster carries its address on the node, not on ConfigurationEndpoint — that one is Memcached’s auto-discovery endpoint and is null for the others, as on AWS.

Any credentials work; with none configured, run eval "$(overcast env)" first — see Using AWS SDKs and CLI.

What works

Area	Behaviour
Real engines	`CreateCacheCluster`, `CreateReplicationGroup` and `CreateServerlessCache` each start a container, ports allocated from `ELASTICACHE_PORT_BASE` (default 63790)
Honest readiness	The status moves to `available` only after the engine answers its own protocol — `PING` for Redis and Valkey, `version` for Memcached. A published port that merely accepts a connection is not enough
Failure is terminal	An engine that never answers settles in the status AWS documents for that shape — `incompatible-network` for a cache cluster, `create-failed` for a replication group or serverless cache — with the reason recorded
VPC placement	A `CacheSubnetGroupName` puts the cache on that subnet group’s VPC network and nothing else; a name Overcast has no record of is refused with `CacheSubnetGroupNotFoundFault`, as on AWS. Without one the cache is in the default VPC, which is the shared data plane
Per-caller endpoints	`{id}.{region}.cfg.{base}` resolves to the engine container from a Lambda or ECS task, on the engine’s own port; the host gets the published port, and `127.0.0.1` when `{base}` has no wildcard DNS. `DescribeCacheClusters` mints it for whoever asks, so a function reading it at runtime gets a name it can dial — Data-plane endpoints
CloudFormation	`AWS::ElastiCache::CacheCluster` and `AWS::ElastiCache::ReplicationGroup`; `Fn::GetAtt` gives `RedisEndpoint.Address`/`.Port` for Redis and Valkey and `ConfigurationEndpoint.Address`/`.Port` for Memcached, the pair each engine has on AWS
Without Docker	Every operation still works as metadata, and statuses settle immediately

Supported engines: redis (redis:6, redis:7), valkey (valkey/valkey:7, valkey/valkey:8), memcached (memcached:1.5, memcached:1.6).

Differences from AWS

Area	On AWS	Overcast
Cluster size	Replicas, cluster mode and failover	A replication group starts a single primary container and reports one node group, always
Replication-group endpoints	A reader endpoint load-balances the replicas	`PrimaryEndpoint` and `ReaderEndpoint` are the same address: one container serves both roles
Node-level detail	Each node has its own address, status and zone	Every `CacheNodes` entry reports the one container’s endpoint and the cluster’s own status
Parameter and security groups	Applied to the engine	`CacheParameterGroupName` is recorded and echoed, never pushed into the engine; `SecurityGroupIds` are dropped entirely
Snapshots	Snapshot, backup and restore	Not implemented
Scaling	Node-count changes and `IncreaseReplicaCount` add nodes	`ModifyCacheCluster` node-count changes and `IncreaseReplicaCount` do not add containers
`CacheSubnetGroupName` on `ReplicationGroup`	Not returned	Not returned either
`AWS::ElastiCache::CacheCluster` naming	The property is `ClusterName`, not `CacheClusterId`	Omit it and the name is generated from the stack, the logical id and a suffix, lowercased and capped at 50 characters

Gotchas

Important

A cache in a VPC is reachable only from that VPC, as on AWS — ElastiCache has no PubliclyAccessible escape hatch. Create both the cache and its caller with the same subnet group, or leave both out of one. A CfnCacheCluster with no cacheSubnetGroupName is in the default VPC, and a Fargate task in the stack’s own VPC cannot resolve it — the data-plane-name-refused advisory says which two resources disagree. See Lambda, ECS and VPCs.

The cache id is constrained too, by Docker rather than by AWS.

Warning

Container names derive from the cache id you choose, and Docker requires them to be unique per daemon. Two Overcasts sharing a daemon can each hold a cache called sessions only if they are given different names — the second to start one fails it with a reason saying so, rather than quietly sharing the first’s.

Operations

22 of 24 listed operations are implemented. Per-operation status, notes and AWS API links: ElastiCache operations.

RDS — the same Docker-backed lifecycle for databases
All service pages
Service names and state overrides
Lambda, ECS and VPCs
AWS API reference