solaris dtrace系列二：dtrace应用基础案例-找出应用系统性能瓶颈&内部数据

这次我们修改模拟服务器守护进程的mydtrace.c源码，给原有的函数增加一些参数，算法上做一些字符串拼接和加法操作并打印出参数和返回值，然后利用sleep模拟系统瓶颈，a_1函数sleep(2)停顿2秒，b_1函数sleep(3)停顿3秒，最后我们编写跟踪探测脚本mydtrace.sh收集系统运行数据，分析瓶颈所在。

1. 修改mydtrace.c源码，增加参数，增加字符串拼接和加法操作并打印出参数和返回值。

#include <stdio.h>
#include <string.h>

void a(char *p1, int p2);
int  a_1(char *p1, int p2);
int  a_1_1(char *p1, int p2);
void b(char *p1, int p2);
int  b_1(char *p1, int p2);
void c(char *p1, int p2);
 
int main ()
{
   while (1)
   {
   a("dtrace-test-in-a",100);
   b("dtrace-test-in-b",200);
   c("dtrace-test-in-c",300);
   }
}
 

void a(char *p1, int p2)
{  char str[100];
   int rst;
   printf("a(%s,%d)\n",p1,p2);
   strcpy(str,p1);
   strcat(str,"&a_1");
   rst=a_1(str,p2+10);  
   printf("a_1=%d\n",rst);
}

int a_1(char *p1, int p2)
{
   int rst;
   printf("a_1(%s,%d)\n",p1,p2);
   rst=a_1_1(strcat(p1,"&a_1_1"),p2+1);
   printf("a_1_1=%d\n",rst); 
   sleep(2);
   return rst; 
}

int a_1_1(char *p1, int p2)
{
   int rst;
   rst=p2+1;
   printf("a_1_1(%s,%d)\n",p1,p2);   
   return rst; 
}

void b(char *p1, int p2)
{  char str[100];
   int rst;
   strcpy(str,p1);
   strcat(str,"&b_1");
   printf("b(%s,%d)\n",p1,p2);
   rst=b_1(str,p2+10);
   printf("b_1=%d\n",rst);  
}

int b_1(char *p1, int p2)
{
   int rst;
   rst=p2+1;
   printf("b_1(%s,%d)\n",p1,p2); 
   sleep(3);  
   return rst;
}

void c(char *p1, int p2)
{
   printf("c(%s,%d)\n",p1,p2); 
}

2. 编译&运行此应用

root@IPS_TEST # gcc mydtrace.c -o mydtraced
root@IPS_TEST # ./mydtraced

a(dtrace-test-in-a,100)
a_1(dtrace-test-in-a&a_1,110)
a_1_1(dtrace-test-in-a&a_1&a_1_1,111)
a_1_1=112
a_1=112
b(dtrace-test-in-b,200)
b_1(dtrace-test-in-b&b_1,210)
b_1=211
c(dtrace-test-in-c,300)
a(dtrace-test-in-a,100)
a_1(dtrace-test-in-a&a_1,110)

3. 编写跟踪程序的探测脚本mydtrace，见mydrace.sh

#!/usr/sbin/dtrace -s
#pragma D option quiet

uint64_t last_timestamp;
pid$target:mydtraced::entry
{
 last_timestamp = walltimestamp;
}

pid$target:mydtraced::return
{
 t = (walltimestamp - last_timestamp) / 1000000;
 printf("<--%s Spent %d ms\n", probefunc,t);
}

4. 运行此脚本，查看输出结果;初始统计有些不准确，需要忽略，后面的a_1,b_1准确的反映了耗时，当然由于它们分别被a,b调用，在a,b的调用统计上也有所反映。分析出瓶颈的关键所在之后，可以更有真对性处理。

root@IPS_TEST # ./mydtrace.sh -p `pgrep mydtraced`
<–b_1 Spent 1537253575408 ms
<–b Spent 1537253575408 ms
<–c Spent 0 ms
<–a_1_1 Spent 0 ms
<–a_1 Spent 1999 ms
<–a Spent 1999 ms
<–b_1 Spent 2999 ms
<–b Spent 2999 ms
<–c Spent 0 ms
<–a_1_1 Spent 0 ms
<–a_1 Spent 1999 ms
<–a Spent 1999 ms
<–b_1 Spent 3009 ms
<–b Spent 3009 ms
<–c Spent 0 ms
<–a_1_1 Spent 0 ms
^C

5. 分析系统性能瓶颈有时需要一些重要参数数据来协助诊断，比如文件名，分配的内存大小等，修改mydtrace.sh脚本

#!/usr/sbin/dtrace -s
#pragma D option quiet

pid$target:mydtraced:a_1:entry
{
 printf("parameters pass into a_1:%s,%d\n",copyinstr(arg0),arg1);
}

6. 运行修改后的脚本，查看输出结果

root@IPS_TEST # ./mydtrace.sh -p `pgrep mydtraced`
parameters pass into a_1:dtrace-test-in-a&a_1,110
parameters pass into a_1:dtrace-test-in-a&a_1,110
^C