最近看了点关于Dtrace的东西，它是个通用型的工具，但我主要集中于分析ruby程序的执行上面。关于操作系统的性能分析，比如cpu、内存、io、文件系统等，使用起来貌似挺复杂，木有细看。

这里简单的输出一条命令：

sudo dtrace -n 'ruby$target:::object-create {@objects[copyinstr(arg0)]=count();}' -c 'ruby -e "puts :hello"'

输出的结果是：

    dtrace: description 'ruby$target:::object-create ' matched 1 probe
    -e:1: unterminated string meets end of file
    dtrace: pid 15203 has exited

      #<Class:0x007fabf38e0700>                                         1
      ARGF.class                                                        1
      IOError                                                           1
      Mutex                                                             1
      NoMemoryError                                                     1
      SyntaxError                                                       1
      SystemStackError                                                  1
      ThreadGroup                                                       1
      Time                                                              1
      LoadError                                                         2
      Object                                                            2
      Gem::Specification                                                8
      Gem::Version                                                     10
      Hash                                                             11
      Gem::Requirement                                                 23
      Array                                                            96
      String                                                          260

ruby2.0已经有支持probe了，所以可以使用dtrace

2.0之前的如果要使用dtrace的话，要使用 ruby-dtrace这个gem

另外在学习的时候，写了dtrace的脚本，其实就是D语言了， rb_flowinfo.d，查看ruby方法的调用过程：

    #!/usr/sbin/dtrace -Zs
    #pragma D option quiet
    #pragma D option switchrate=10

    self int depth;

    dtrace:::BEGIN
    {
       printf("%s %6s %10s %16s:%-4s %-8s -- %s\n", "C", "PID", "DELTA(us)", "FILE", "LINE", "TYPE", "NAME");
    }

    ruby*:::method-entry, 
    ruby*:::method-return
    /self->last == 0/
    {
       self->last = timestamp;
    }

    ruby*:::method-entry
    /copyinstr(arg0) == "Object"/
    {
       this->delta = (timestamp - self->last) / 1000;
       this->name = strjoin(strjoin(copyinstr(arg0), "::"), copyinstr(arg1));
       printf("%d %6d %10d %16s:%-4d %-8s %*s-> %s\n", cpu, pid, this->delta,
             basename(copyinstr(arg2)), arg3, "method", self->depth * 2, "", this->name);

       self->depth++;
       self->last = timestamp;
    }

    ruby*:::method-return
    /copyinstr(arg0) == "Object"/
    {
       this->delta = (timestamp - self->last) / 1000;
       self->depth -= self->depth > 0 ? 1 : 0;
       this->name = strjoin(strjoin(copyinstr(arg0), "::"), copyinstr(arg1));
       printf("%d %6d %10d %16s:%-4d %-8s %*s<- %s\n", cpu, pid, this->delta,
             basename(copyinstr(arg2)), arg3, "method", self->depth * 2, "", this->name);
       self->last = timestamp;
    }

用于测试的ruby脚本， trace_method_call.rb

    dtrace_ruby $ cat trace_method_call.rb 
    #!/Users/wenleslie/.rvm/rubies/ruby-2.0.0-p0/bin/ruby
    def func_c
       puts "Function C"
       sleep 1
    end

    def func_b
       puts "Function B"
       sleep 6
       func_c
    end

    def func_a
       puts "Function A"
       func_b
    end

    func_a      

执行：

    sudo ./rb_flowinfo.d -c ./trace_method_call.rb

结果如下：

C    PID  DELTA(us)             FILE:LINE TYPE     -- NAME
Function A
Function B
4  15702      35986 trace_method_call.rb:13   method   -> Object::func_a
4  15702         55 trace_method_call.rb:7    method     -> Object::func_b
Function C0  15702    6001065 trace_method_call.rb:2    method       -> Object::func_c

6  15702    1000922 trace_method_call.rb:5    method       <- Object::func_c
6  15702         27 trace_method_call.rb:11   method     <- Object::func_b
6  15702         19 trace_method_call.rb:16   method   <- Object::func_a
^C

我的evernote上关于dtrace笔记 有稍微详细一点的内容，有兴趣的话，可以看看