看我怎么搞垮erlang

最近在做mapreduce系统的资源监控模块(项目全部基于erlang).

概要:监控各个map或者reduce的任务进程,每个进程都有各自的内存上限.

该监控系统提供进程监控注册,注册监控的进程在其内存超出其内存上限,该

监控系统则将告知mapreduce系统杀死进程.

        在测试时,需要模拟进程不断消耗内存,于是根据如何搞垮erlang的文章(

见前一篇转载文章)的原理,利用消息共泛,不断给自己发送2阶乘的二进制数据,

使其内存不断增大,达到自身内存上限,使监控系统能够监控到.

        具体代码片段如下:

consume_memory() ->
    consume_memory(100000).
consume_memory(TimeCount) ->
    Pid = spawn(fun() -> loop_receive() end),
    send_msg(Pid, TimeCount),
    {ok, Pid}.

send_msg(_Pid, 0) ->
    ok;
send_msg(Pid, N) ->
    Pid ! <<5:(N)>>,
    timer:sleep(300),
    send_msg(Pid, N-1).
    
loop_receive() ->
    Result = erlang:process_info(self(), [memory]),
    ?INFO_F("comsumed memory : ~w ~n", [Result]),
    receive
      _ ->
          loop_receive()
    end.

测试了一下不行,内存增加到5600Byte就不会继续增加了,发现loop_receive()

这边是递归调用,所以为了能够内存继续增加,于是做点小修改,使其不递归,

不调用同一个地址的loop_receive,改成去调用try语句.

代码如下(测试通过):

consume_memory() ->
    consume_memory(100000).
consume_memory(TimeCount) ->
    Pid = spawn(fun() -> loop_receive() end),
    send_msg(Pid, TimeCount),
    {ok, Pid}.

send_msg(_Pid, 0) ->
    ok;
send_msg(Pid, N) ->
    Pid ! <<5:(N)>>,
    timer:sleep(300),
    send_msg(Pid, N-1).
    
loop_receive_old() ->
    Result = erlang:process_info(self(), [memory]),
    ?INFO_F("comsumed memory : ~w ~n", [Result]),
    receive
      _ ->
          loop_receive_old()
    end.
    
loop_receive() ->
    Result = erlang:process_info(self(), [memory]),
    ?INFO_F("comsumed memory : ~w ~n", [Result]),
    try
	receive
	  _ ->
	      loop_receive()
	end
    catch
      _:_ ->
          loop_receive()
    end.

ps: error

send_msg(Pid, TimeCount),

改成

spawn(fun() -> send_msg(Pid, TimeCount) end),

查看哪些进程占用内存最高

spawn(fun() -> etop:start([{output, text}, {interval, 5}, {lines, 20}, {sort, memory}]) end)

新开辟一个进程,不影响erlang shell输入,以输出text方式启动etop,刷新间隔5s,输出行数20,按照内存降序排序.