Subversion Repositories SE.SVN

Rev

Blame | Last modification | View Log | RSS feed

-module(keepalive).

%% your web app can push data to clients using a technique called comet long
%% polling.  browsers make a request and your server waits to send a
%% response until data is available.  see wikipedia for a better explanation:
%% http://en.wikipedia.org/wiki/Comet_(programming)#Ajax_with_long_polling
%%
%% since the majority of your http handlers will be idle at any given moment,
%% you might consider making them hibernate while they wait for more data from
%% another process.  however, since the execution stack is discarded when a
%% process hibernates, the handler would usually terminate after your response
%% code runs.  this means http keep alives wouldn't work; the handler process
%% would terminate after each response and close its socket rather than
%% returning to the big @mochiweb_http@ loop and processing another request.
%%
%% however, if mochiweb exposes a continuation that encapsulates the return to
%% the top of the big loop in @mochiweb_http@, we can call that after the
%% response.  if you do that then control flow returns to the proper place,
%% and keep alives work like they would if you hadn't hibernated.

-export([loop/1, start/1]).

%% internal export (so hibernate can reach it)
-export([resume/3]).

-define(LOOP, {?MODULE, loop}).

start(Options = [{port, _Port}]) ->
    mochiweb_http:start([{name, ?MODULE}, {loop, ?LOOP}
                         | Options]).

loop(Req) ->
    Path = mochiweb_request:get(path, Req),
    case string:tokens(Path, "/") of
      ["longpoll" | RestOfPath] ->
          %% the "reentry" is a continuation -- what @mochiweb_http@
          %% needs to do to start its loop back at the top
          Reentry = mochiweb_http:reentry(?LOOP),
          %% here we could send a message to some other process and hope
          %% to get an interesting message back after a while.  for
          %% simplicity let's just send ourselves a message after a few
          %% seconds
          erlang:send_after(2000, self(), "honk honk"),
          %% since we expect to wait for a long time before getting a
          %% reply, let's hibernate.  memory usage will be minimized, so
          %% we won't be wasting memory just sitting in a @receive@
          proc_lib:hibernate(?MODULE, resume,
                             [Req, RestOfPath, Reentry]),
          %% we'll never reach this point, and this function @loop/1@
          %% won't ever return control to @mochiweb_http@.  luckily
          %% @resume/3@ will take care of that.
          io:format("not gonna happen~n", []);
      _ ->
          ok(Req, io_lib:format("some other page: ~p", [Path]))
    end,
    io:format("restarting loop normally in ~p~n", [Path]),
    ok.

%% this is the function that's called when a message arrives.
resume(Req, RestOfPath, Reentry) ->
    receive
      Msg ->
          Text =
              io_lib:format("wake up message: ~p~nrest of path: ~p",
                            [Msg, RestOfPath]),
          ok(Req, Text)
    end,
    %% if we didn't call @Reentry@ here then the function would finish and the
    %% process would exit.  calling @Reentry@ takes care of returning control
    %% to @mochiweb_http@
    io:format("reentering loop via continuation in "
              "~p~n",
              [mochiweb_request:get(path, Req)]),
    Reentry(Req).

ok(Req, Response) ->
    mochiweb_request:ok({_ContentType = "text/plain",
                         _Headers = [], Response},
                        Req).