Spark Structured Streaming Data Deduplication using State Store

This is a project showing how the spark in-built HDFS backed state store can be used to deduplicate data in a stream. The related blog post can be found at https://barrelsofdata.com/data-deduplication-spark-state-store

Build instructions

From the root of the project execute the below commands

• To clear all compiled classes, build and log directories

./gradlew clean

• To run tests

./gradlew test

• To build jar

./gradlew build

Run

Ensure your local hadoop cluster is running (hadoop cluster tutorial) and start two kafka brokers (kafka tutorial).

• Create kafka topic

kafka-topics.sh --create --bootstrap-server localhost:9092 --replication-factor 2 --partitions 2 --topic streaming-data

• Start streaming job

spark-submit --master yarn --deploy-mode cluster build/libs/spark-state-store-data-deduplication-1.0.0.jar <KAFKA_BROKER> <KAFKA_TOPIC> <FULL_OUTPUT_PATH> <DEDUPLICATED_OUTPUT_PATH> <WINDOW_SIZE_SECONDS>
Example: spark-submit --master yarn --deploy-mode client build/libs/spark-state-store-data-deduplication-1.0.0.jar localhost:9092 streaming-data fullOutput deduplicatedOutput 5

• You can feed simulated data to the kafka topic
• Open new terminal and run the shell script located at src/test/resources/dataProducer.sh
• Produces the two instances of the following json structure every 1 second: {"ts":1594307307,"usr":"user1","tmp":98}

cd src/test/resources
./dataProducer.sh localhost:9092 streaming-data

View Results

Open a spark-shell and use the following code, do change the paths to where the outputs are stored.

spark.read.parquet("fullOutput").orderBy("user","eventTime").show(truncate = false)

spark.read.parquet("deduplicatedOutput").orderBy("user","eventTime").show(truncate = false)