I needed to construct multipart messages recently, but also didn’t want to include Tesla in my dependencies, so I built a separate library for message construction. It should cover most use cases for multipart/form-data requests, but very open to suggestions to improve the API and functionality.

https://github.com/breakroom/multipart

So the intent behind creating streaming one is that I have to send multipart requests that contain files, and these files can be many megabytes in size. If I do N of these requests at the same time, I could easily end up exhausting my memory quota.

So, the intent was not to make it more performant, but make it not crash production server.

I actually have no idea how the streaming performance versus non-streaming compares. It’d be good test to do.

I don’t think Finch supports building actual multipart body. This is a task on it’s own.

What I am doing is using Tesla to help me out with that task.

First, build multipart body using Tesla.Multipart

  multipart =                                                                                                           
    Multipart.new()                                                                                                     
    |> Multipart.add_content_type_param("charset=utf-8")                                                                
    |> add_fields(form_data)                                                                                            
    |> add_files(file_parts)                                                                                            
                                                                                                                        
  mp_headers = Multipart.headers(multipart)                                                                             
  stream = Multipart.body(multipart)
  headers = ... # your other headers, optional

And then you can either POST it as a streaming body request, once this PR is merged https://github.com/keathley/finch/pull/107

  :post                                                                                                                 
  |> Finch.build(url, headers ++ mp_headers, {:stream, stream})                                                         
  |> Finch.request(YourFinch) 

Or, you can POSt it in non-streaming fashion (should work on current release/master):

  :post                                                                                                                 
  |> Finch.build(url, headers ++ mp_headers, stream |> Enum.join())                                                         
  |> Finch.request(YourFinch)