This project shows the power of serverless, event-driven architectures to execute code in response to messages or to handle streams of data records.

It demonstrates two OpenWhisk actions (written in JavaScript) that read and write messages with Message Hub (based on Apache Kafka). The use case demonstrates how actions can work with data services and execute logic in response to message events.

One action receives message streams of one or more data records, and these are in turn piped to another action in an OpenWhisk sequence (a way to link actions declaratively in a chain). The second action aggregates the message and posts a transformed message to another topic.

• OpenWhisk
• Message Hub (Kafka)

You should have a basic understanding of the OpenWhisk programming model. If not, try the action, trigger, and rule demo first. Also, you'll need a Bluemix account and the latest OpenWhisk command line tool (wsk) installed and on your PATH.

1. Provision Message Hub
2. Create OpenWhisk actions, triggers, and rules
3. Test new message events
4. Delete actions, triggers, and rules
5. Recreate deployment manually

Log into Bluemix, provision a Message Hub instance, and name it kafka-broker. On the "Manage" tab of your Message Hub console create two topics: in-topic and out-topic.

Copy template.local.env to a new file named local.env and update the KAFKA_INSTANCE_NAME, SRC_TOPIC, and DEST_TOPIC values for your instance if they differ. Update the API_KEY, USER, and PASSWORD values from the "Credentials" tab.

deploy.sh is a convenience script reads the environment variables from local.env and creates the OpenWhisk actions, triggers, and rules on your behalf. Later you will run these commands yourself.

./deploy.sh --install

Note: If you see any error messages, refer to the Troubleshooting section below.

Note: deploy.sh will be replaced with wskdeploy in the future. wskdeploy uses a manifest to deploy declared triggers, actions, and rules to OpenWhisk.

Open one terminal window to poll the logs:

wsk activation poll

There are two helper scripts that simulate a message producer and a message consumer.

# Produce a message, will trigger the sequence ./kafka_publish.sh # Consume a message after processing is complete ./kafka_consume.sh

Use deploy.sh again to tear down the OpenWhisk actions, triggers, and rules. You will recreate them step-by-step in the next section.

./deploy.sh --uninstall

This section provides a deeper look into what the deploy.sh script executes so that you understand how to work with OpenWhisk triggers, actions, rules, and packages in more detail.

Make the Kafka instance in Bluemix available as an event source.

wsk package refresh wsk package create kafka wsk package bind kafka kafka-out-binding \ --param api_key ${API_KEY} \ --param kafka_rest_url ${KAFKA_REST_URL} \ --param topic ${DEST_TOPIC} wsk package get --summary kafka-out-binding

Create the kafka-trigger trigger that listens for new messages.

wsk trigger create kafka-trigger \ --feed /_/Bluemix_${KAFKA_INSTANCE_NAME}_Credentials-1/messageHubFeed \ --param isJSONData true \ --param topic ${SRC_TOPIC}

Upload the mhget-action action as a single file Node.js action. This downloads messages when they arrive via the trigger.

wsk action create mhget-action actions/mhget/mhget.js

Upload the mhpost-action action as a zipped action, in order to include dependencies that are not in the default Node.js environment on OpenWhisk. This aggregates information from the action above, and sends the summary JSON back to Kafka.

DIR=`pwd` cd actions/mhpost npm install --loglevel=error zip -r mhpost.zip * cd ${DIR} wsk action create kafka/mhpost-action actions/mhpost/mhpost.zip --kind nodejs:6

Declare a linkage between the mhget-action and mhpost-action in a sequence named kafka-sequence.

wsk action create kafka-sequence --sequence mhget-action,kafka-out-binding/mhpost-action

Declare a rule named kafka-inbound-rule that links the trigger kafka-trigger to the sequence named kafka-sequence.

wsk rule create kafka-inbound-rule kafka-trigger kafka-sequence

# Produce a message, will trigger the sequence ./kafka_publish.sh # Consume a message after processing is complete ./kafka_consume.sh

Check for errors first in the OpenWhisk activation log. Tail the log on the command line with wsk activation poll or drill into details visually with the monitoring console on Bluemix.

If the error is not immediately obvious, make sure you have the latest version of the wsk CLI installed. If it's older than a few weeks, download an update.

wsk property get --cliversion

Apache 2.0

This project was inspired by and reuses significant amount of code from this article.