# Amazon ElastiCache Global Datastore

> ElastiCache Global Datastore runs a complete, continuously billed replica cluster in every secondary region, so 2 secondaries triple the node-hour spend of the primary cache. ZopNight discovers global datastores via a dedicated provider on the 6-hour cycle, attributes cost per region, and flags secondaries no application reads from.

Source: https://zop.dev/integrations/aws/elasticache-global
Updated: 2026-08-19

---

An ElastiCache Global Datastore replicates a Redis cluster across AWS regions for disaster recovery and local reads. Every secondary region runs a full, continuously billed replica cluster, so unused secondaries double or triple cache spend.

## Cache replication at cluster prices

Global Datastore has no meter of its own; its cost is that every member is a full ElastiCache cluster billing node-hours at its instance sizes, all day, in its own region, plus cross-region data transfer for the replication stream. Caches amplify the multiplication in a way databases do not: cache clusters are typically sized for throughput headroom, so each secondary inherits a generous node configuration whose entire purpose in the secondary region may be to sit warm awaiting a failover that has never been rehearsed.

## Regional replicas, individually accounted

A dedicated provider discovers global datastores on the 6-hour cycle, and cross-region cost attribution splits the spend per member cluster. That per-region view is the finding: a secondary whose local applications hold no connections and issue no reads is pure standby cost, and the recommendation asks whether the failover requirement justifies a same-sized cluster, a smaller one, or none. Re-creating a cache from a warm primary is often acceptable for workloads whose cache is genuinely a cache rather than a datastore.

## Standby caches that never serve

The common shapes: a secondary added during a multi-region initiative that shipped only its database tier, leaving the cache replica readerless; geo-local read replicas for a user base that never materialized in that geography; and DR postures where the application tier fails over to a region whose cache exists but whose connection strings were never wired, meaning the replica would not even be used in the disaster it exists for. Each bills node-hours continuously.

## Where to inspect a global datastore

The ElastiCache console's Global datastores view lists each datastore with its primary and secondary clusters, regions, and node configurations. Per-secondary CloudWatch connection and hit metrics answer the only question that matters (does anything read from this region), and the failover runbook, if one exists, answers whether anything ever could.