[pmboyd]

I worked at a mass notification company that used a similar approach and it worked well with a handful of workers pulling in batches that could push out 300 to 500 of messages per second and could queue a quarter million messages in a few seconds. There wasn't really a need to optimized past that since the throughput on the rest of the system was the bottlenecked by outside components.

The problem with queuing in {insert new hot queueing system} is that persistent and failover support is often weak. They are by default are more efficient and simpler than RDBMS. If you're running Snapchat, Redis or RabbitMQ or whatever is entirely reasonable.