Get Started Free
Screen Shot 2021-07-23 at 4.13.50 PM

Sophie Blee-Goldman

Senior Software Engineer (Presenter)


Bill Bejeck

Integration Architect (Author)



Kafka Streams uses the concept of a task as a work unit. It's the smallest unit of work within a Kafka Streams application instance. The number of tasks is driven by the number of input partitions. For example, if you have a Kafka Streams application that only subscribes to one topic, and that topic has six partitions, your Kafka Streams application would have six tasks. But if you have multiple topics, the application takes the highest partition count among the topics.



Tasks are assigned to StreamThread(s) for execution. The default Kafka Streams application has one StreamThread. So if you have five tasks and one StreamThread, that StreamThread will work records for each task in turn. However, in Kafka Streams, you can have as many threads as there are tasks. So for five tasks, you could configure your application to have five threads. Each task gets its own thread, and any remaining threads are idle.


With respect to task assignment, application instances are similar to tasks. If you have an input topic with five partitions, you could spin up five Kafka Streams instances that all have the same application ID, and each application would be assigned and process one task. Spinning up new applications provides the same increase in throughput as increasing threads. Just like with threads, if you spin up more application instances than tasks, the extra instances will be idle, although available for failover. The advantage is that this behavior is dynamic; it doesn't involve shutting anything down. You can spin up instances on the fly, and you can take down instances.

Consumer Group Protocol

Because Kafka Streams uses a Kafka consumer internally, it inherits the dynamic scaling properties of the consumer group protocol. So when a member leaves the group, it reassigns resources to the other active members. When a new member joins the group, it pulls resources from the existing members and gives them to the new member. And as mentioned, this can be done at runtime, without shutting down any currently running applications.

So you should set up as many instances as you need until all of your tasks are accounted for. Then in times when traffic is reduced, you can take down instances and the resources will be automatically reassigned.


Use the promo code STREAMS101 to get $25 of free Confluent Cloud usage

Be the first to get updates and new content

We will only share developer content and updates, including notifications when new content is added. We will never send you sales emails. 🙂 By subscribing, you understand we will process your personal information in accordance with our Privacy Statement.