# ElastiCache Multi-AZ Non-Critical

> ElastiCache replication groups with an explicit dev/test environment tag and Multi-AZ enabled carry standby replicas non-critical caches rarely need. ZopNight prices removing them as cost times replicas divided by replicas-plus-one, never the old flat 0.50. Abstains fleet-wide today because the replication-group parent row prices at $0, a known producer gap.

Source: https://zop.dev/integrations/aws/recommendations/elasticache-multi-az-non-critical
Updated: 2026-08-19

---

## Why an explicit dev or test env tag is required

- env tag (env/environment/stage/tier): authoritative and required (RC-053 HARDEN). A prod env tag is a hard veto (never strip HA from production), and firing requires an explicit dev/test env tag; concrete-or-abstain on the env signal.
- resource.Name: not consulted. The dev/test name-substring fallback (isDevTestName) is removed (it misfired on a prod cache literally named "...test..." and is unverifiable); absent/non-env tag => abstain.
- resource.Metadata: multi_az_enabled (bool; producer exists): applicability gate.
- resource.Metadata: replicasPerShard (float; discoverer elasticache.go): the real removed-node fraction. Utilisation series such as those in [the ElastiCache CloudWatch metric set](https://docs.aws.amazon.com/AmazonElastiCache/latest/red-ug/CacheMetrics.html) play no part in this gate; the decision rests entirely on tags and node metadata.

## Node count as the cost lever

removing the standby replica nodes (going to a single primary per shard) is the cost lever. Every node in a replication group is the same type and [each one accrues its own node-hour charge](https://aws.amazon.com/elasticache/pricing/), so node cost is proportional to node count: savings = cost × replicasPerShard / (replicasPerShard + 1) (the shard count cancels), capped at cost. The hardcoded 0.50 halving fraction is removed; abstains when replicasPerShard is missing or ≤ 0. PRODUCER-GAP: RC-053 is structurally dead in production until a producer surfaces a real parent-level cluster MonthlyCost (or standby-cost rollup) on the replication-group "elasticache" UID. The RG parent row carries multi_az_enabled/replicasPerShard but no InstanceType → aggregator resolves MonthlyCost 0 (no rateFallbackKeys["elasticache"]), so the cost gate always abstains. The abstain logic is correct (emits nothing wrong); it just cannot fire until the parent cost lands. No longer a "fixed estimate".

## Removing the standby replicas

1. Verify this is a non-production cache cluster
2. Disable Multi-AZ first (prerequisite only, required before the last replica can be removed; this step alone removes no nodes and has no cost effect)
3. Go to ElastiCache Console → Modify → decrease replica count to 0 (or use DecreaseReplicaCount / lower NumCacheClusters)
4. This is the step that actually removes and stops billing for the standby replica nodes