Mike Trienis bio photo

Mike Trienis

All about data product and services that scale; from design to implementation

Email Twitter LinkedIn Github Stackoverflow

The most efficient way to load data into AWS Redshift is first upload your data to S3 and then execute the copy command on Redshift. From the documentation

A COPY command is the most efficient way to load a table. When you load data from Amazon S3, the COPY command is able to read from multiple data files simultaneously. Whether you load from data files on Amazon S3 or from an Amazon DynamoDB table, Amazon Redshift distributes the workload to the cluster nodes and performs the load process in parallel.

Therefore for each batch of data in your Spark Streaming application,

  1. Write batch to S3 folder identified by the batch interval
  2. Copy the batch to Redshift by executing a copy command
graph LR A[AWS Kinesis] --> B[Spark Streaming Application] B --> C[AWS S3] B --> D[AWS Redshift] C -.-> D

Pushing data from Spark Streaming to S3 is fairly straight forward as Spark exposes the saveAsTextFile output operation that supports s3n hadoop connection point. The endpoint will likely have a structure that resembles:


The output operation can be applied to both RDDs and DStreams. However, in order to uniquely identify the batch interval we will need to expose the spark streaming time parameter. Simply override the foreachRDD function:

def dispatch(RDD[T], Time): Unit {
    // 1. Write the batch to an _S3_ folder that is identified by the batch interval 
    // 2. Copy the batch to Redshift by executing a `copy` command

// perform etl process
dstream.foreachRDD(dispatch _)