Flow summaries are determined by creating the "union" over multiple PCollections that contain the flow summaries for certain aggregations. The union operation requires that flow summary information is encoded / decoded. At the moment a JacksonJsonCoder is used that is not efficient because it is based on reflection and uses JSON as the representation for transmission.

Considering the expected amount of generated flow summaries an efficient custom coder for flow summary information should be used. The JSON representation of flow summaries is created at the output stages of the pipeline, i.e. when they are written to Elastic and Kafka.