Yes we could have gone with either one. It would have worked for having both.

RabbitMQ is very reliable (At the cost of performance). It persists the data on disk. Even if the RabbitMQ server goes down, we can attach the disk onto a new instance ant it would restore to the previous state. We needed this for task queues. They should work no matter what.

Redis is much more performant (at the cost of reliability). Because, by default, everything happens in-memory. Without additional settings for persistence (Which will bring down performance), if the Redis instance goes down, then all data is lost. We needed the performance for quickest delivery of notifications. milli-seconds matter.

Plus, now we can scale tasks and notifications independent of each other.