Webmachine streamed bodies

Webmachine allows the resource developer to handle request and response bodies as either whole units (binary or iolist) to be handed around at once, or else to choose to “stream” the body.

The body-handling functions are:

The last of these, wrq:set_resp_body/2, is also called implicitly with the return value of any content-producing function such as to_html/2.

The first of these (req_body) is the simplest. It will provide the whole incoming request body as a binary. (Unless the body is too large, as set by wrq:set_max_recv_body/2or defaulting to 50M) For the majority of resources, this is the easiest way to handle the incoming request body.

If a resource wants to handle the incoming request body a hunk at a time, it may call wrq:stream_req_body/2instead. Instead of a binary, this produces a StreamBodystructure.

(It is an error to call both wrq:req_body/1and wrq:stream_req_body/2in the execution of a single resource.)

A StreamBodyis a pair of the form {Data,Next}where Datais a binary and Nextis either the atom donesignifying the end of the body or else a 0-arity function that, when called, will produce the “next” StreamBodystructure.

The integer parameter to wrq:stream_req_body/2indicates the maximum size in bytes of any Hunkfrom the resulting StreamBody.

When a resource provides a body to be sent in the response, it should use wrq:set_resp_body/2. The parameter to this function may be either an iolist, representing the entire body, or else a pair of the form {stream, StreamBody}.

An example may make the usage of this API clearer. A complete and working resource module using this API in both directions:

-export([init/1, allowed_methods/2, process_post/2]).


init([]) -> {ok, undefined}.

allowed_methods(ReqData, State) -> {['POST'], ReqData, State}.

process_post(ReqData, State) ->
    Body = get_streamed_body(wrq:stream_req_body(ReqData, 3), []),
    {true, wrq:set_resp_body({stream, send_streamed_body(Body,4)},ReqData), State}.

send_streamed_body(Body, Max) ->
    case Body of
        <<A:HunkLen,Rest/binary>> ->
            io:format("SENT ~p~n",[<<A:HunkLen>>]),
            {<<A:HunkLen>>, fun() -> send_streamed_body(Rest,Max) end};
        _ ->
            io:format("SENT ~p~n",[Body]),
            {Body, done}

get_streamed_body({Hunk,done},Acc) ->
    io:format("RECEIVED ~p~n",[Hunk]),
get_streamed_body({Hunk,Next},Acc) ->
    io:format("RECEIVED ~p~n",[Hunk]),

If you use this resource in place of the file /tmp/mywebdemo/src/mywebdemo_resource.erlin the quickstart setup, you should then be able to issue curl -d '1234567890' the command line and the io:formatcalls will show you what is going on.

Obviously, a realistic resource wouldn’t use this API just to collect the whole binary into memory or break one up that is already present – you’d use req_bodyand put a simple iolist into set_resp_bodyinstead. Also, the choices of 3 and 4 bytes as hunk size are far from optimal for most reasonable uses. This resource is intended only as a demonstration of the API, not as a real-world use of streaming request/response bodies.