POC to parallelize long time processes using Kafka and Quarkus.
Imagine you have a big amount of items to process and each one takes a long time to finish.
This POC proposes one solution to diminish the overall process time.
The idea is simple, one application, the producer, publishes messages on a Kafka topic and another application, the processor, reads these messages and process them.
If the messages don't need to be processed in an ordered way, we can configure the application, using SmallRye Reactive Message, to create a pool of threads and process multiple messages simultaneously.
Kafka also allows that we consume messages from its topic from multiple instances of our application. To achieve this, we must configure the partition number of the topic. For instance, if our topic has a partition number of three and we run four instances of our application, three instances will be able to process one specific partition and the fourth will remain idle.
In this POC I use both ideas, that is I run the processing application in multiple instances and each instance has its own thread pool that processes multiple messages simultaneously.
Just run the application, access the Swagger-UI to inform the quantity of items to process, execute it and check the results on the console log of the producer and the processor instances.
The Swagger-UI is on http://localhost:8080/q/swagger-ui
Enter the processor folder and run:
$ ./mvnw package
Enter the producer folder and run:
$ ./mvnw package
On the root folder, run:
$ docker-compose up
If you want to run multiple instances of the processor, just pass the scale argument:
$ docker-compose up --scale processor=2
The NUM_PARTITIONS parameter of Kafka in the docker-compose file says how many partitions the topics will have.
Enter the processor folder and run:
$ mvn quarkus:dev
Enter the producer folder and run:
$ mvn quarkus:dev -Ddebug=5006
In this mode, Quarkus will automatically download one Kafka image and run it for you.
The num.partitions parameter defines how many partitions the topic will have.
The consumer group is defined by one id.
One consumer group will receive all messages sent to a topic.
One consumer group can have 'n' instances of applications running.
Each instance of the consumer group will process messages from some partitons of the topic.
Each instance of the consumer group will be able to read one or more partitions of the topic.
If you have more consumers than partitions, some consumers will remain idle.
If the messages don't need to be processed in order, you can use the following annotation:
@Incoming("extraction-requests")
@Blocking(ordered = false, value = "my-custom-pool")
public void read(Client client) {Inform the number of threads of your my-custom-pool in your application.properties:
smallrye.messaging.worker.my-custom-pool.max-concurrency=3You can inform the number of partitions the topic will have in your Kafka test container with this parameter, informing your topic name (in this case, extraction-requests) :
quarkus.kafka.devservices.topic-partitions.extraction-requests=3
https://quarkus.io/guides/kafka
https://strimzi.io/docs/operators/latest/using.html
https://smallrye.io/smallrye-reactive-messaging/smallrye-reactive-messaging/3.13/index.html