The word application is a bit of misnomer. Component would be a better term. Each library should be an OTP app, since that allows proper specification of runtime dependencies (with applications in your mix.exs). That in turn allows various tasks to work on the project. For example, exrm will include all dependent applications in the OTP release. Another example is dialyze which builds PLTs from runtime dependencies.

It’s worth noting that poison is a simple OTP app that doesn’t start any process. Such applications are called library applications.

Finally, it’s worth mentioning that any project you build will mix will generate an OTP app.

This is a question that interests me too. I have a vague notion in my mind. So far, this is what I’ve gathered and would like it cleared up.

Reach for OTP when some operation I might be performing could potentially take a long time. It might depend on outside resources out of my control, or be computationally expensive. Additionally, if I care about the result of that operation, OTP will allow me to retrieve the result in a safe way. If I call await too soon, I’m stuck waiting for the result.

If I need some form of state storage that doesn’t need to persist a long time, go for OTP. An agent or genserver is cheap to start up.

Ok yeah great point about the limits of async.