The problem is that network APIs still suck. The boundary between components is the place where strong typing is most vital, but microservices are still largely based around stringly-typed JSON.
> If everything's an API call, calling to it is trivial.
Calling a library API is a lot easier than calling a web API
> The interfaces can be their own container.
No they can't. That doesn't make any sense. What do you even mean?
> I know that a container does its job and does it well.
Maybe. I'm never clear on what exactly the job of the "container" is supposed to be - it seems to involve process isolation (I'd think ordinary processes should do that?), and maybe doing funky network namespacing and PID namespacing and a few other things like that, only then other people are saying you shouldn't do that?
> A container makes it evident if it is the result of bad programming. You'll see good data go in, and bad come out.
Only if you can tell good data from bad data. Which is much easier to do within the language, where you have a type system to help you.
> This plays right in to containerization. Compute/storage is cheap.
Compute/storage is cheap, so you probably won't ever have to make your system distributed. The biggest problem with microservices is that they slow development by making it much harder to run the whole system. Anything that makes your development feedback loop slow is a big problem.
> Yeah, LISP and FP still hasn't taken off.
FP is taking off. It's a lot bigger than it was 5 or 10 years ago.
> Hence containers. It's a ball of privilege separation, data segregation, clean interfaces, and easily maintainable.
On the contrary, containers are being pushed as the one true solution to everything, and seem to aggregate a bunch of unrelated concerns.
You don't have to use a RESTful API and JSON. If you feel that configuration is lacking you can use a binary format and a non-RESTful API.
Sure, and that improves things a lot. I do use Thrift (via Scrooge) sometimes. I get the sense that isn't what the microservice advocates are talking about though.
But even then, why introduce a network boundary when you don't need to? All it does is make your system slower, add overhead converting back and forth from DTOs, and give more opportunity for calls to fail.
...anything that doesn't involve a network connection? You're arguing that spinning up entire OSs and networking them together is "simpler" than calling methods/functions inside the same process. Once you start including the machinery to deal with all of these failure scenarios, your code is absolutely no longer simple. If you don't do that, you're code is dangerously simplistic.