Even though my feature team won’t communicate directly with Kafka, we’ll use a wrapper organized by another feature team, it is important to understand what is underneath.
Hence first I just drafted a small app which first produces some records, then goes to sleep and then consumes them. The production was easy. I could see them through the CLI consumer, but couldn’t consume them through my app. I tried a bunch of stuff, I checked tutorials, I upgraded my Kafka version. Nothing. In fact not nothing. Sometimes I managed to consume the messages, but then for a lot of restarts of the app, nothing again.
Then I ran into another article where the groupid was not a constant but it was a UUID.
That was finally something different, so I tried it. And it worked like a charm.
Only one instance of a consumer group will receive a message. Put it differently the messages are load balanced between the instances of a consumer group. The more I restarted the app the more instances I got in the same group I think. Even though the other processes were stopped, I think they still left some marks behind them, thus my new instance barely received any message. Of course when all my consumers started to belong to a unique group, this problem vanished.
So at that point I split the one dummy app, into two dummies. One produced and one consumed and on my two screens I could see both running at the same time. Great!
Next step is to do the same on OpenShift.