ElastiCache Metrics Reference

DB Optimizer collects the following CloudWatch metrics for every registered ElastiCache cluster (Redis and Valkey), at a 60-second period aligned to minute boundaries.

Collected metrics

Metric	Statistic	Unit	Description
EngineCPUUtilization	Average	Percent	CPU used by the Redis/Valkey engine thread
DatabaseMemoryUsagePercentage	Average	Percent	Memory used vs. maxmemory
CacheHits	Sum	Count	Successful key lookups (keyspace_hits)
CacheMisses	Sum	Count	Failed key lookups (keyspace_misses)
Evictions	Sum	Count	Keys removed due to maxmemory policy
CurrConnections	Maximum	Count	Current client connections
NetworkBytesIn	Sum	Bytes	Bytes received
NetworkBytesOut	Sum	Bytes	Bytes sent
ReplicationLag	Maximum	Seconds	Replica delay behind primary
BytesUsedForCache	Maximum	Bytes	Absolute memory used for cached data

That table is the whole set

Ten metrics, one statistic each. Nothing else is requested — no Reclaimed, SwapUsage, CurrItems, GetTypeCmds or SetTypeCmds, no serverless metrics such as ElastiCacheProcessingUnits, and no derived CacheHitRatio. CacheHits and CacheMisses are stored as collected, so any hit ratio is computed by whoever reads them.

Dimensions

CacheClusterId only.

Replication group metadata

For clusters in a replication group, each metric row carries:

• Replication group ID
• Node role: primary or replica
• Shard ID (always recorded; empty string for single-shard groups)

Data retention

ElastiCache metrics are retained for 15 days. There is no longer-term rollup for them.

Redis vs. Valkey

Redis and Valkey clusters use the same metric set — the engine value reported by the ElastiCache API is recorded on the instance record as either redis or valkey, but the metrics collected are identical.