kafka-streams-on-heroku

Kafka Streams example on Heroku with a multi-project gradle build

Dependencies

1. Postgres
2. Kafka (+ Zookeeper) 0.10+ (this uses 0.11 brokers against 1.0 client)
3. Java 8
4. Gradle 4.3 (use sdkman)

Local Development

Building

$ ./gradlew clean build

Testing

$ ./gradlew clean test

Building FatJar Artifacts

$ ./gradlew clean stage

Running Locally

Topologies are organized as subprojects. You can run any or all of them

(start postgres - optional, zookeeper - required, kafka - required)
$ ./gradlew streams-text-processor:run
$ ./gradlew streams-aggregator:run
$ ./gradlew streams-anomaly-checker:run

Deployment: Heroku

Dependencies

1. Postgres
2. Kafka
3. Heroku CLI
4. Heroku Kafka CLI Plugin

Config Vars

1. SENDGRID_API_KEY (optional via SendGrid addon)
2. TESTING_EMAIL (optional for sinking to a test email using SendGrid addon)

Setup

Install the Heroku CLI: https://devcenter.heroku.com/articles/heroku-cli

Install the Heroku Kafka CLI Plugin:

heroku plugins:install heroku-kafka

Clone the application:

$ git clone git@github.com:kissaten/kafka-streams-on-heroku.git

Create the application:

$ cd kafka-streams-on-heroku
$ heroku apps:create <application name>
$ heroku buildpacks:add heroku/ruby
$ heroku buildpacks:add heroku/gradle

Deploy the application:

$ git push heroku master

Run the setup script:

$ ./setup <app name> <plan>

Smoke Testing

$ heroku kafka:topics:write [prefix]textlines "hello world" -a <app>
$ heroku pg:psql -c 'select * from windowed_counts' HEROKU_POSTGRESQL_URL -a <app>

Example Use Cases

Now let's use Kafka Streams with some example use cases. The data-generators directory contains some simple Ruby scripts to generate streams of data. Instructions on how to use them are below.

Word Count

First we'll do word count over a large stream of text. This will produce into Kafka lines from Alice's Adventures in Wonderland.

$ heroku run ruby data-generators/text-generator/stream-lines-to-kafka.rb data-generators/text-generator/alice-in-wonderland.txt --app sushi

Alternatively, if you have Ruby and Bundler installed locally, you can run the data generator locally

$ bundle install --path=vendor/gems
$ cd data-generators/text-generator
$ HEROKU_KAFKA_URL=$(heroku config:get HEROKU_KAFKA_URL) \
HEROKU_KAFKA_CLIENT_CERT=$(heroku config:get HEROKU_KAFKA_CLIENT_CERT) \
HEROKU_KAFKA_CLIENT_CERT_KEY=$(heroku config:get HEROKU_KAFKA_CLIENT_CERT_KEY) \
HEROKU_KAFKA_TRUSTED_CERT=$(heroku config:get HEROKU_KAFKA_TRUSTED_CERT) \
HEROKU_KAFKA_PREFIX=$(heroku config:get HEROKU_KAFKA_PREFIX) \
bundle exec ruby stream-lines-to-kafka.rb alice-in-wonderland.txt

Now we can see the word count for specific time windows:

$ heroku pg:psql -c 'select * from windowed_counts order by time_window desc' HEROKU_POSTGRESQL_URL

Anomaly Detection

Let's look at a more interesting use case -- not only because it is more realistic but also because it better showcases continuously updating caluculations based on a stream of data. You'll need two separate terminal windows for this.

In the first one, tail the Heroku application logs

$ heroku logs --tail --app sushi

In the second one, we'll generate some data. This will produce into Kafka fake log data at a rate of 10 messages per second with a 20% chance of anomaly.

$ heroku run ruby data-generators/log-generator/stream-logs-to-kafka.rb 10 .2 --app sushi

Alternatively, if you have Ruby and Bundler installed locally, you can run the data generator locally

$ bundle install --path=vendor/gems
$ cd data-generators/log-generator
$ HEROKU_KAFKA_URL=$(heroku config:get HEROKU_KAFKA_URL) \
HEROKU_KAFKA_CLIENT_CERT=$(heroku config:get HEROKU_KAFKA_CLIENT_CERT) \
HEROKU_KAFKA_CLIENT_CERT_KEY=$(heroku config:get HEROKU_KAFKA_CLIENT_CERT_KEY) \
HEROKU_KAFKA_TRUSTED_CERT=$(heroku config:get HEROKU_KAFKA_TRUSTED_CERT) \
HEROKU_KAFKA_PREFIX=$(heroku config:get HEROKU_KAFKA_PREFIX) \
bundle exec ruby stream-logs-to-kafka.rb 10 .2

Looking at the Heroku applications logs, you will see STDOUT output showing an anomaly has been detected.