Kafka is a distributed streaming platform written in Java for the Apache ecosystem.

Backbeat uses Kafka as a journaling system to track jobs to be performed and as a persistent queue for failed operations to retry. Jobs come to Kafka from Node.js streams (from producers), and Kafka queues them for Backbeat to follow.

The lifespan of jobs queued in Kafka’s journal is configurable (the current default lifespan of a job in Backbeat’s Kafka queue is seven days), and after the job’s age reaches the configured retention time, Kafka purges it from the queue. This solves two problems: job instructions are stored in a stable, non-volatile system (not, for instance, in a fugitive database, such as Redis) that is also not enduring (once completed, the job information is not preserved).