I don’t think we disagree as much. I was very pro-PaaS when we started but having been burned a couple of times I’m more timid. E.g. right now parse is actually better as we can pick from multiple different providers and not just from Facebook… Still moving to this “vendor neutral” mode required more than just a recompile of the app.
I didn’t mean for this to be a Microservices > Monolithic post because I actually don’t think that.. I think all startups should launch as monolith first and then slowly migrate to microservices as they start to scale. Otherwise you will over-engineer the microservices. This is exactly the position we were in, we now understand the complexities of our busines and how to break it down into the right separate pieces.
I used to agree that PaaS provides a leg-up to quickly release something and it might be true for some services (it might be true in terms of price with the free GAE quota) but I recently did a bootcamp where I went thru the process of building a full startup and used Spring Boot + Linode for the production. This was relatively easy to do since Spring Boot is so great!
The one major pin point which app engine solves is HTTPS for which I used lets encrypt. The problem is that getting lets encrypt to work on Spring Boot is to say the least painful…
I think scaling is one of those things PaaS vendors oversell, you can get amazing scaling/uptime with one server and a good CDN. If you go beyond the limits of that then you probably have enough traffic to warrant working on infrastructure.