Prototype to test high availability between two apps using kafka. Also to see how to scale up when there is high concurrency of requests.
To accomplish this there are two application, producer and consumer, producer will create messages by batches and send it to kafka and on the other end the consumer app will start dequeuing messages. The project nft will execute stress tests and have scenarios to scale up the consumers and see what is the effect on the metrics.
Both apps are going to expose basic metrics to monitor the resources.
###install all platform by executing
./install.sh
this will install kafka, zookeeper, 3 consumers, 1 producer, Grafana and Prometheus
Take the IP address shown at the end of the installation process and use it to access all the apps
eg.
grafana: http://192.168.1.100:3000 Prometheus: http://192.168.1.100:9090
Each consumer is using a pool thread to consume in parallel 200 messages at the time and because I am adding a random delay between 500ms and 10000ms you might see some spikes but nothing out of the ordinary.
In grafana there are custom metrics that you can check like
Producer:
- batch_batches_total: is a counter of batches that has been processed.
Consumer:
- consumer_number_messages_total: Total number of messages that has been consumed.
###Kill all the pods execute
./killAll.sh
To start the process of messages you need to execute a endpoint in the producer app and declaring how many messages will the batch contains.
POST http://192.168.1.100:8080/batches Request Payload
{
"numberOfMessages": "1000"
}
After that it will start the first batch of 1000 messages and after that finished it will generate another one. so it is endless process of executing batches.
To stop the process og generating new batches, but does not stop the current running.