Amazon ElastiCache Costs: Idle Nodes, Replicas and Reservations
ElastiCache has no stop operation for a node-based cluster. Savings come from fewer, smaller or reserved nodes, or from deleting a cache nobody uses.
The ElastiCache meters
AWS bills ElastiCache in three ways. Self-designed clusters on on-demand nodes are billed hourly from the time a node is launched until it is terminated. Serverless caches bill data stored in GB-hours plus requests in ElastiCache Processing Units. Reserved nodes take a discount off the hourly rate in exchange for a one- or three-year term.
“Launched until terminated” is the key phrase. There is no stopped state that pauses a node’s charge, so ElastiCache cannot be put on a nightly schedule the way an EC2 instance can. ZopNight does not claim to schedule it.
What can actually change
- Idle caches. A cluster with no connections, hits, misses or CPU activity is a delete candidate.
- Replicas outside production. Multi-AZ standby replicas double or triple node count for availability a test cache may not need.
- Node family. Older non-Graviton node types.
- Commitment. Steady production caches running on-demand.
ZopNight’s ElastiCache rules
- ElastiCache Cluster Idle requires all four signals, connections, cache hits, cache misses and CPU, to be covered for at least 30 days and below their floors on both average and maximum. If any one is missing it abstains.
- ElastiCache Multi-AZ Non-Critical prices removing the standby replicas from a non-production replication group, using the real replicas-per-shard count rather than a guess. It does not rely on the cache’s name to decide what is non-production.
- ElastiCache Non-Graviton Instance flags older node families.
- ElastiCache Reserved Node Opportunity flags steady on-demand nodes.
ElastiCache is treated as a stateful data service throughout. ZopNight’s safety gate will not auto-delete a stateful resource or auto-reboot one for a resize, and the remediation pause template excludes ElastiCache entirely, so every change on a cache goes through a person.
Related
Cluster and node details: ElastiCache fact sheet and ElastiCache node fact sheet. The commitment side is explained on Reserved Instances, and the idle method on idle detection.
Questions we get a lot.
Why doesn't ZopNight schedule ElastiCache?
On-demand ElastiCache nodes bill from launch until termination and there is no stopped state, so there is nothing for a schedule to pause. ZopNight works on it through idle, replica, node-family and reservation recommendations.
Could a recommendation delete my cache automatically?
No. ElastiCache is classed as stateful, so deletes and reboots are never automatic.