Correct
Well, the tcp stack is implemented in erlang so you need it. I don’t there are any HTTP clients implemented in elixir yet though but all go through erlang ones.
The key is to do the measurement on the actual TCP request, not the HTTP client abstraction. As I mentioned before Hackney (erlang HTTP client) has metric doing this measurement. This is not to say that there are functionality further below that skews the metric but I think that is the best we can get for now. The tcp uses async model too.
Erlang has got an HTTP client in the standard “distribution”. It also has got HTTP packet parsers in the tcp handling which is part of the standard library.
If you mean the elixir standard library I can’t see it being any different than a separate OTP application distributed with elixir rather than being built-in to the standard library (which doesn’t make sense to me)
As a final alternative, I know that the OTP team is working on improving tracing. Perhaps it is now possible to actually do metrics through erlang tracing without too much overhead. This means you can measure anything, even 3rd party software.