Kinesis Connector for Spark Structured Streaming

Implementation of Kinesis Source Provider in Spark Structured Streaming. SPARK-18165 describes the need for such implementation. More details on the implementation can be read in this blog

This is a fork of https://github.com/qubole/kinesis-sql with the build configuration and source code updated for building against Spark 3.2.1 in order to fix a number of bugs involving the consumer not receiving new messages after a period of no new messages being added to the Kinesis data stream.

WARNING: Relatively Unmaintained Software

This project is relatively unmaintained. If you're already using Amazon Kinesis Data Streams, instead of using Spark for processing streaming data, a far better solution is to use Kinesis Firehose Delivery Streams with AWS Lambda Functions for processing the micro-batches. This is a far more economical solution, and is also much easier to develop. Additionally, if you select ARM architecture for your Lambda Functions, you will save even more in terms of ongoing operating costs. Because of these facts, use of this project is strongly discouraged in favor of a more AWS-centric approach.

Downloading and Using the Connector

The connector is available from the Maven Central repository. It can be used using the --packages option or the spark.jars.packages configuration property. Use the following connector artifact

Spark 3.2: com.roncemer.spark/spark-sql-kinesis_2.13/1.2.3_spark-3.2

Developer Setup

Clone spark-sql-kinesis from the source repository on GitHub.

You need maven, sbt, and openjdk 17 (not 21!) to successfully build this project.

Spark version 3.2.x

git clone git@github.com:roncemer/spark-sql-kinesis.git
git checkout master
cd spark-sql-kinesis
mvn install -DskipTests

This will create target/spark-sql-kinesis_2.13-1.2.3_spark-3.2.jar file which contains the connector code and its dependency jars.

Deploying to Maven Central

Apply for a Sonatype JIRA account, and create a ticket requesting access to OSSRH. You must provide them with a domain which you can prove that you own by adding a TXT message to the DNS zone for that domain. Wait for them to approve it and create your account.

Create a GPG key (RSA, 4096 bits) and upload to both GitHub and the Ubuntu key servers. Wait for it to propagate. Set up a $HOME/.m2/settings.xml with a servers section with a server entry for the OSSHR server with your Sonatype username and password, as well as a profiles section with a profile entry for OSSRH which tells it how to sign JAR files using GPG. Google is your friend for getting these things set up.

Once you have all of the above prerequisites in place, you can build and publish the package to staging using this command:

mvn -DskipTests clean source:jar verify gpg:sign install:install deploy:deploy

After deployment to staging, you can go to https://s01.oss.sonatype.org/#stagingRepositories and log in with your username and password, then select the repository which was created by the deployment, and click Close. Click the Activity tab and click Refresh at the top left of the top table. If you did everything correctly, all tests will pass and the last line will say Repository closed.

Once the repository has been closed successfully, you can click the Release button at the top of the top table, and your release will be pushed out to the Maven repositories and become available for public consumption.

How to use it

Setup Kinesis

Refer Amazon Docs for more options

Create Kinesis Stream

$ aws kinesis create-stream --stream-name test --shard-count 2

Add Records in the stream

$ aws kinesis put-record --stream-name test --partition-key 1 --data 'Kinesis'
$ aws kinesis put-record --stream-name test --partition-key 1 --data 'Connector'
$ aws kinesis put-record --stream-name test --partition-key 1 --data 'for'
$ aws kinesis put-record --stream-name test --partition-key 1 --data 'Apache'
$ aws kinesis put-record --stream-name test --partition-key 1 --data 'Spark'

Example Streaming Job

Refering $SPARK_HOME to the Spark installation directory.

Open Spark-Shell

$SPARK_HOME/bin/spark-shell --jars target/spark-sql-kinesis_2.13-1.2.3_spark-3.2.jar

Subscribe to Kinesis Source

// Subscribe the "test" stream
scala> :paste
val kinesis = spark
    .readStream
    .format("kinesis")
    .option("streamName", "spark-streaming-example")
    .option("endpointUrl", "https://kinesis.us-east-1.amazonaws.com")
    .option("awsAccessKeyId", [ACCESS_KEY])
    .option("awsSecretKey", [SECRET_KEY])
    .option("startingposition", "TRIM_HORIZON")
    .load

Check Schema

scala> kinesis.printSchema
root
|-- data: binary (nullable = true)
|-- streamName: string (nullable = true)
|-- partitionKey: string (nullable = true)
|-- sequenceNumber: string (nullable = true)
|-- approximateArrivalTimestamp: timestamp (nullable = true)

Word Count

// Cast data into string and group by data column
scala> :paste

     kinesis
    .selectExpr("CAST(data AS STRING)").as[(String)]
    .groupBy("data").count()
    .writeStream
    .format("console")
    .outputMode("complete") 
    .start()
    .awaitTermination()

Output in Console

+------------+-----+
|        data|count|
+------------+-----+
|         for|    1|
|      Apache|    1|
|       Spark|    1|
|     Kinesis|    1|
|   Connector|    1|
+------------+-----+ 

Using the Kinesis Sink

// Cast data into string and group by data column
    scala> :paste
    kinesis
    .selectExpr("CAST(rand() AS STRING) as partitionKey","CAST(data AS STRING)").as[(String,String)]
    .groupBy("data").count()
    .writeStream
    .format("kinesis")
    .outputMode("update") 
    .option("streamName", "spark-sink-example")
    .option("endpointUrl", "https://kinesis.us-east-1.amazonaws.com")
    .option("awsAccessKeyId", [ACCESS_KEY])
    .option("awsSecretKey", [SECRET_KEY])
    .start()
    .awaitTermination()

Kinesis Source Configuration

Kinesis Sink Configuration

Roadmap

• We need to migrate to DataSource V2 APIs for MicroBatchExecution.
• Maintain Per Micro-Batch Shard Commit state in Dynamo DB

Acknowledgement

This connector would not have been possible without reference implemetation of Kafka connector for Structured streaming, Kinesis Connector for Legacy Streaming and Kinesis Client Library. Structure of some part of the code is influenced by the excellent work done by various Apache Spark Contributors.