Apache Kafka is publish-subscribe messaging rethought as a distributed, partitioned, replicated commit log service. Kafka plays an important role in any streaming application. Let's see what happens without having Kafka in a streaming application. If the streaming application processing the streams is down for 1 minute for some reason, what will happen to the stream of data for that 1 minute? We will end up losing 1 minute's worth of data. Having Kafka as one more layer buffers incoming stream data and prevents any data loss. Also, if something goes wrong within the Spark Streaming application or target database, messages can be replayed from Kafka. Once the streaming application pulls a message from Kafka, acknowledgement is sent to Kafka only when data is replicated in the streaming application. This makes Kafka a reliable receiver.
There are two approaches to receive data from Kafka.