Demo Flink project to show how to react on interaction events in real-time and trigger CRM campaigns for customer engagement based on campaign configurations.
The project also utilizes the Broadcast State Pattern in order to update the rules (campaigns) at runtime without restarting the project, using a dedicated, low-frequency, Kafka topic.
Each customer engagement campaign runs for a specific product (game) and can have a start and end date. Additionally, a campaign can be enabled and disabled and has a unique ID.
The key element of a campaign is the filter. The filter is a JavaScript function that is executed for each event and must return a boolean value. With this filter, we can define the conditions that must be met in order to trigger the campaign based on an event.
Campaigns can be configured by simply writing the JSON representation to the crm-campaigns
Kafka topic.
Interaction events are written as JSON to the events-valid
Kafka topic. For each event, the Flink Application will
evaluate all active campaigns and check if the filter matches. If that is the case, a trigger is written to the
crm-triggers
Kafka topic, indicating that the user is eligible to receive the offer connected to the campaign.
Please refer to the local execution example to get a better understanding of how the system works.
Requirements:
- Docker
- (optional) kcat (see: https://github.com/edenhill/kcat)
Default configuration is defined in src/main/resources/config.json
. You can override it by creating a
custom config file and pass the path as a parameter named config
when running the Flink job.
Just run the following command in the project dir to start ZooKeeper and Kafka as defined in docker-compose.yml
:
docker-compose up -d
We are using kcat
to interact with Kafka. We start the producer and exit (CTRL+D) again to just create the
necessary topics:
kcat -b localhost:9093 -t events-valid -P
kcat -b localhost:9093 -t crm-campaigns -P
kcat -b localhost:9093 -t crm-triggers -P
Then use kcat
to list topics:
kcat -b localhost:9093 -L | grep topic
The output should include the three topics above:
Metadata for all topics (from broker 1: localhost:9093/1):
4 topics:
topic "crm-triggers" with 1 partitions:
topic "events-valid" with 1 partitions:
topic "crm-campaigns" with 1 partitions:
topic "__consumer_offsets" with 50 partitions:
Run com.innogames.analytics.rtcrm.App
and use kcat
to create a campaign, send and event and check the trigger:
Open http://localhost:8081/ to see the Flink Web UI.
kcat -b localhost:9093 -t crm-campaigns -P
{"campaign_id":4711,"enabled":true,"game":"foe","event_name":"fight","start_date":"2023-09-01T10:00:00.00Z","end_date":"2023-12-01T10:00:00.00Z","filter":"function(event) { return event.getData().get('result') === 'loss' }"}
kcat -b localhost:9093 -t events-valid -P
{"schema_version":"2023110110010101","event_id":"7a8d0614-4e8e-4464-b439-36e31441a040","system_type":"game","system_name":"foe-backend","game":"foe","market":"de","player_id":12345,"event_type":"progress","event_name":"fight","event_scope":"user","created_at":"2023-11-20T11:00:00.00Z","received_at":"2023-11-20T11:00:02.00Z","hostname":"foe-backend-1.ig.com","context":{},"data":{"result":"loss"}}
kcat -b localhost:9093 -t crm-triggers -C -o beginning -u
When Kafka and Flink are running, you can also use the dedicated Demo UI project to interact with the system via a web frontend.