This example demonstrates the capability of the Kogito Event-Driven Predictions AddOn: to enable, when included as dependency of a simple service containing PMML files, to trigger evaluations of its queries and receive the corresponding results via specific CloudEvents.
The source and destination of these events are two configured Kafka topics.
The main goal behind the addon is to allow Kogito PMML services to be used as part of an event processing pipeline.
You will need:
Like the other Kogito AddOns, the only required step to enable it is to include it as dependency in the POM file:
<dependency> <groupId>org.kie</groupId> <artifactId>kie-addons-springboot-event-predictions</artifactId> </dependency>
The version is implicitly derived from the kogito-bom included in the dependencyManagement section.
The only configuration required is for the input and output topics.
Here is the important section of application.properties:
kogito.addon.cloudevents.kafka.kogito_incoming_stream=<input_topic_name> kogito.addon.cloudevents.kafka.kogito_outgoing_stream=<output_topic_name> kogito.addon.tracing.decision.kafka.bootstrapAddress=<kafka_bootstrap_address> spring.kafka.bootstrap-servers=<kafka_bootstrap_address> spring.kafka.consumer.group-id=<group_id>
Insert the value you need in <kafka_bootstrap_address>, <group_id>, <input_topic_name> and <output_topic_name>. Pre-configured values already works if you follow this guide without changes.
There's a useful docker-compose.yml in the root that starts a dedicated Kafka instance for quick tests.
Simply start it with this command from the root of the repo:
docker compose up -d
Once everything is started you can check the data contained in your small Kafka instance via Kafdrop at http://localhost:9000/.
mvn clean compile spring-boot:run
mvn clean package java -jar target/dmn-event-driven-springboot.jar
or on Windows
mvn clean package java -jar target\dmn-event-driven-springboot.jar
Here is an example of a input event that triggers the evaluation of the LinReg regression model included in this example. The data field contains the query input.
Just send this payload to the configured input topic:
{ "specversion": "1.0", "id": "a89b61a2-5644-487a-8a86-144855c5dce8", "source": "SomeEventSource", "type": "PredictionRequest", "subject": "TheSubject", "kogitopmmlmodelname": "LinReg", "data": { "fld1":3.0, "fld2":2.0, "fld3":"y" } }
And you should receive something similar to this in the output topic:
{ "specversion": "1.0", "id": "d54ace84-6788-46b6-a359-b308f8b21778", "source": "LinReg", "type": "PredictionResponse", "subject": "TheSubject", "kogitopmmlmodelname": "LinReg", "data": { "fld4": 52.5 } }
The data field contains the query output. Values of id fields will change, but the rest will be the same.
All the leaf subfolders of the test events resource folder contain a pair of input.json and output.json files.
There's one for every possible variation in the structure of the input/output events supported by the addon. Feel free to try them all.