RDS Metrics Reference

DB Optimizer collects the following CloudWatch metrics for every registered RDS instance, at a 60-second period aligned to minute boundaries.

Common metrics (all engines)

Metric	Statistic	Unit	Description
CPUUtilization	Average	Percent	CPU usage of the DB instance
FreeableMemory	Minimum	Bytes	Available RAM including reclaimable page cache
DatabaseConnections	Average	Count	Active client connections
ReadIOPS	Average	Count/sec	Read I/O operations per second
WriteIOPS	Average	Count/sec	Write I/O operations per second
ReadLatency	Average	Seconds	Average time per read I/O
WriteLatency	Average	Seconds	Average time per write I/O
DiskQueueDepth	Average	Count	Outstanding I/O requests waiting
FreeStorageSpace	Minimum	Bytes	Available storage on the instance
ReplicaLag	Maximum	Seconds	Replication delay (replicas only)
SwapUsage	Maximum	Bytes	Swap space in use

That table is the whole set

Eleven metrics, requested identically for every engine. There are no PostgreSQL-specific, MySQL-specific or Aurora-specific requests — no MaximumUsedTransactionIDs, BinLogDiskUsage, AuroraReplicaLag, FreeLocalStorage or Volume* metric is collected. The engine recorded on the instance selects one of exactly two branches (mysql,postgres for RDS, redis,valkey for ElastiCache), not a per-engine metric list.

Dimensions

DBInstanceIdentifier only.

Scrape interval

The default scrape interval is 300 seconds (5 minutes). At this interval, each poll requests 5 minutes of 60-second datapoints — the same resolution as a 60-second interval but at one fifth the CloudWatch API cost.

DB_OPTIMIZER_SCRAPE_INTERVAL is enforced at 300 to 3600 seconds. A value outside that range fails core's startup with an error rather than being clamped, and there is no override.

Data retention

RDS metrics are retained for 15 days, the same as raw node and pod metrics. There is no longer-term rollup for them.

Null handling

If CloudWatch returns no datapoint for a metric in a period, no row is stored and the API reports that metric as null. A measured value of 0 is stored and reported as 0. The console renders null as an em dash (—) and 0 as a zero — two visually distinct states so "not measured" is never mistaken for "measured and idle".