Kafka on the other hand is a queuing system that was first developed at LinkedIn for its own internal purposes. It is written in Scala and is designed from the ground up for horizontal scalability and the best performance possible.

Focusing on performance is a key differentiator for Apache Kafka but it means that in order to achieve performance we need to sacrifice something. Messages in Kafka don't hold unique IDs but are addressed by their offset in the log. Apache Kafka consumers are not tracked by the system; it is the responsibility of the application design to do so. Message ordering is implemented at the partition level and it is the responsibility of the consumer to identify if a message has been delivered already.

Exactly once semantics were introduced in version 0.11 and are part of the latest 1.0 release so that now messages can be both strictly ordered within a partition and always arrive exactly once at each consumer: