erlang R17新socket选项{active,N}

erlang R17带来了新的socket选项{active,N} ，与{active,once}一起为应用层提供流量控制。为什么要多了这个选项，{active,once}不是可以有效抑制大量socket消息吗？

我们知道，{active,once}在每次接收到包都要重新设置active选项，才能继续接收erlang的消息通知。实际上，每次设定{active,once}都意味着调用一次epoll_ctl， 如果请求过于频繁，就会有大量的epoll_ctl调用。erlang目前只有一个线程会收割epoll_wait事件，epoll_wait要轮询已经就绪的ctl队列，如果大量的ctl事件将会阻塞了epoll_wait的操作，造成网络处理能力的下降。

那么，我们能不能设定接收N个的socket消息后再执行一次epoll_ctl，这样可以有效减少epoll_ctl的调用，{active,N}就是这样出现的。

下面来看一下{active,N}的说明

Add the {active,N} socket option for TCP, UDP, and SCTP,where N is an integer in the range -32768..32767, to allow acaller to specify the number of data messages to be deliveredto the controlling process. Once the socket's deliveredmessage count either reaches 0 or is explicitly set to 0 withinet:setopts/2 or by including {active,0} as an option whenthe socket is created, the socket transitions to passive({active, false}) mode and the socket's controlling processreceives a message to inform it of the transition. TCPsockets receive {tcp_passive,Socket}, UDP sockets receive{udp_passive,Socket} and SCTP sockets receive{sctp_passive,Socket}.

The socket's delivered message counter defaults to 0, but itcan be set using {active,N} via any gen_tcp, gen_udp, orgen_sctp function that takes socket options as arguments, orvia inet:setopts/2. New N values are added to the socket'scurrent counter value, and negative numbers can be used toreduce the counter value. Specifying a number that wouldcause the socket's counter value to go above 32767 causes aneinval error. If a negative number is specified such that thecounter value would become negative, the socket's countervalue is set to 0 and the socket transitions to passive mode.If the counter value is already 0 and inet:setopts(Socket,[{active,0}]) is specified, the counter value remains at 0but the appropriate passive mode transition message isgenerated for the socket.

设定了{active,N}选项后，进程在接收了N个包后，会收到{tcp_passive, Socket}消息，意味着这个Socket进入被动模式，需要重新设置active选项。

接下来，在实际的例子中测试这个参数：

-module(server).

-export([start/0]).
-export([continue/1]).
-define( PORT, 8888).

start() ->
  {ok, LSock} = gen_tcp:listen(?PORT, [binary, {packet, 0},{active, false}]),
  io:format("socket listen: ~p on ~p ~n",[LSock, ?PORT]),
  accept(LSock).

accept(LSock) ->
  {ok, ASock} = gen_tcp:accept(LSock), 
  Pid = spawn(fun() -> do_loop(ASock) end),
  gen_tcp:controlling_process(ASock, Pid),
  inet:setopts(ASock, [{active, 3}]),
  accept(LSock).

do_loop(ASock) ->
  receive
    {tcp, Socket, Data} ->
       io:format("socket ~p recv: ~p ~n",[Socket, Data]);
    {tcp_closed, Socket} ->
       io:format("socket ~p close ~n",[Socket]);
    {tcp_passive,Socket} ->
       io:format("socket ~p is passive, please call continue/1 ~p ~n",[Socket, self()]);
    release_passive ->
       inet:setopts(ASock, [{active, 3}]);
    Err ->
       io:format("socket may error: ~p ~n",[Err])
  end,
  do_loop(ASock).

continue(Pid) ->
  Pid ! release_passive,
  ok.

编译启动这个模块后，我们创建一个客户端来请求这个服务端：

1> f(S), {ok,S} = gen_tcp:connect({127,0,0,1},8888,[{packet,0}]).
{ok,#Port<0.526>}
2> gen_tcp:send(S,<<"hello">>).
ok
3> gen_tcp:send(S,<<"hello">>).
ok
4> gen_tcp:send(S,<<"hello">>).
ok
5> gen_tcp:send(S,<<"hello">>).
ok

服务端控制台打印了这样的信息：

D:\tmp>erl -s server
socket listen: #Port<0.422> on 8888
Eshell V6.0  (abort with ^G)
1> socket #Port<0.479> recv: <<"hello">>
1> socket #Port<0.479> recv: <<"hello">>
1> socket #Port<0.479> recv: <<"hello">>
1> socket #Port<0.479> is passive, please call continue/1 <0.33.0>
1> server:continue(pid(0,33,0)).
socket #Port<0.479> recv: <<"hello">>
ok
2>

在上面的例子中，我们设定了{active,3}的选项，在接收到客户端3次数据后，socket进入了passive状态，在重新设置{active,N}后继续接收tcp消息。

那么，如何在实际项目中运用{active,N}选项？

inet:setopts(Socket, [{active, 300}]),
erlang:send_after(30 * 1000, self(), release_passive);

大概思路是，在30秒内最多接收300个包，超过就不接收，等待这30秒完成后继续接收，如此反复。

利用这点还可以加多一个计数器，如果超过10次进入passive状态，说明这个Socket存在问题，有攻击的行为。

参考：

http://blog.csdn.net/mycwq/article/details/24814843
http://www.erlang.org/download/otp_src_17.0.readme
http://blog.yufeng.info/archives/2970