A telecommunications company collects network usage data throughout each day at a rate of several thousand data points each second. The company runs an application to process the usage data in real…

Querying an Aurora database with a Lambda function for drops in network usage introduces inherent latency, as data must first be written and aggregated into Aurora, then periodically queried, which is not suitable for 'least latency' real-time detection of immediate drops.

Modifying the processing application to publish data to an Amazon Kinesis Data Stream enables real-time ingestion of high-volume data. An AWS Lambda function can then be configured to consume directly from the Kinesis Data Stream, processing records as they arrive with millisecond-level latency, allowing for immediate anomaly detection and triggering actions with the least possible delay.

Replacing Aurora with DynamoDB and using a Lambda function triggered by DynamoDB Streams still involves writing data to a database first. A direct stream-to-function approach (Kinesis to Lambda) is more immediate for detecting anomalies from the raw usage data stream.

Database Activity Streams in Aurora are primarily for auditing and compliance, reporting database changes rather than the raw, real-time network usage data itself before it hits the database. Detecting drops directly from the source stream is lower latency for the actual event.