yynice / clickstream

"Real-time" clickstream analysis using Kafka and Spark

Clickstream

Sample project that processes clickstream data using Kafka and Apache Spark.

Install scala, kafka and apache-spark using homebrew.

export JAVA_HOME="$(/usr/libexec/java_home)"
export PATH=$JAVA_HOME:$PATH

export SCALA_HOME="/usr/local/Cellar/scala/2.12.4"  # find it using `brew info`
export PATH=$SCALA_HOME/bin:$PATH

Make sure that /usr/local/bin is also added to your $PATH.

Use pyenv or similar to manage your python versions and virtual environments. After creating a virtual environment, install dependencies with: pip install -r requirements.txt.

To use production data, copy the CSV file into data/production.csv.

Quickstart

See the make commands in for running the services locally.

iterm walkthrough

1. Start Zookeeper: make zookeeper
2. Start Kafka: make kafka
3. New tab, create the clickstream topic with make create_topic (unless it already exists).
4. Start the simple Spark stream that monitors the clickstream topic and prints the messages to the command line: make spark_read
5. New tab, stream some sample data to Kafka: make sample_data
6. The sample data should appear in the simple stream in the previous tab.
7. Make sure that your production data (a really big CSV) is found under data/production.csv
8. Start importing production data with make production_data
9. Start the categories stream with make spark_categories
10. The categories should appear counted, with the sliding interval of 10 seconds.
11. The output of the previous stream should also appear writen to the file system in the output directory.

TODO

• Remove log4j logs! Could not find a way to use the with the spark-submit command.
• Continue with setting up Docker. Likely by creating a new Docker image, with Spark and Python3 on it.