[mcharawi]

Hey, cofounder of Laudspeaker here. At a high level, there are two main problems to solve for when it comes to scaling this type of system. One is the ability of your infrastructure to handle resource intensive customer requests (sending 10 million emails) that are instantiated at arbitrary unpredictable times—this sort of work is usually best handled using job queues and job processing threads that are horizontally scalable. The second is the channel specific failure modes that are common at higher message sending volumes, for example IP blacklisting by spam filters for email, or sending the same notifications repeatedly when dealing with mobile push. We’ve set up our architecture to tackle the first problem and are currently performing stress testing to find and fix any bottlenecks there; with respect to the second type of problem, we are actively experimenting with ways to mitigate these failure modes as our product matures.