Trello has been using RabbitMQ for the last three years. Prior to RabbitMQ, we were relying on a Redis Pub-Sub implementation. Recently, due to RabbitMQ's questionable behavior when network partitions occur, we've made the switch over to Kafka. This blog post goes into depth on our RabbitMQ implementation, why we chose Kafka, and the Kafka-based architecture we ended up with. Current state of the