Understanding the request reference `request_ref` in Mint

When calling Mint.HTTP.request/5, a request identifier request_ref is returned.
Then when streaming the responses with Mint.HTTP.stream/2, I need to wait for messages and make sure that the request reference inside the message received matches with the request reference that was returned by request/5. More can be read here: https://hexdocs.pm/mint/Mint.HTTP.html#stream/2-responses

I do not actually understand how the messages could include another request reference; I was thinking, maybe it can happen if I execute multiple times in parallel request/5 and stream responses; however I tried spawning multiple processes executing requests, but never I get a message with another request_ref.

How can I reproduce it?

Well, you can start many requests from the same process, all of them will have a unique ref. You can use it to collect individual replies into individual result sets.

1 Like

I’m trying but can’t reproduce it :sweat_smile: I try something similar:

{:ok, conn, request_ref1} = HTTP.request(conn, method, path, headers, nil)
{:ok, conn, request_ref2} = HTTP.request(conn, method, path, headers, nil)
{:ok, conn, request_ref3} = HTTP.request(conn, method, path, headers, nil)

receive do
  message ->
    {:ok, conn, mint_messages} = HTTP.stream(conn, message)

mint_messages will always contain the first request_ref1, there are no other incoming messages including the second or third request ref; the second and third call to request/5 seem to be ignored.

Maybe I misunderstood you; at least I thought the code above is what you meant by ‘multiple request from the same process’.

Probably subsequent requests are dropped by the socket process in HTTP1, have you tried with something you can ensure to be HTTP2?

Or with different conns?

I’ve got a small client I wrote for the ex_ari library, you can check out the source here, https://github.com/CityBaseInc/ex_ari/blob/master/lib/ex_ari/http_client.ex

and it’s usage here, https://github.com/CityBaseInc/ex_ari/blob/master/lib/ex_ari/http/asterisk.ex

and here

1 Like

The model here, is we start one process per REST resource, and that process manages concurrent requests for that endpoint.

You could easily have one process manage all of your endpoints. But I didn’t want one hot resource to slow everything else down. For example, the Channel resource probably sees 80% more traffic than any other one.

1 Like

Sorry for the - probably very trivial - question, but what is the purpose of the process? I thought it was to re-use a connection (stored in a process as state) but I see that for every single request, you create a new connection.

Sorry for the delay, was on vacation.

It’s not a trivial question, and actually something I meant to address. I just forgot to revisit the connection stuff once it was working :wink:

Thanks for the reminder, it is supposed to maintain one connection.