When using Kinesis Producer Library to aggregate records, send them to a Kinesis Data Stream, and run a Kinesis Data Analytics application, how is de-aggregation handled?

Question

The **Kinesis Producer Library** aggregates records before sending them to a **Kinesis Data Stream**.

When consuming records from that **Data Stream** with the **Kinesis Consumer Library**, record de-aggregation is automatic.

For consuming records from that **Data Stream** with a **Kinesis Data Analytics** application, how is de-aggregation handled? 
* Does the **Data Analytics** application de-aggregate records automatically?
* Does this case require running the **Kinesis Consumer Library** inside a **Data Analytics** application?
* Something else?

Answer

So this is a depends question.  The older Flink producer(org.apache.flink.streaming.connectors.kinesis.FlinkKinesisProducer) for kinesis did use de-aggregation by default, I believe, but there was a Flink configuration to turn it off.  The newer producer does not use de-aggregation.  I think that the version of KDA currently running today uses the older Flink Producer.
https://nightlies.apache.org/flink/flink-docs-master/docs/connectors/datastream/kinesis/

As for consumers receiving these de-aggregated records, we use Lambda's written in GoLang to read the de-aggregated records and write them to our data store.  We used this in our consumer to help pull out the records: https://github.com/awslabs/kinesis-aggregation

When using Kinesis Producer Library to aggregate records, send them to a Kinesis Data Stream, and run a Kinesis Data Analytics application, how is de-aggregation handled?

Relevant content