Webbench 源码分析 阅读笔记

一.简介

Webbench是有名的网站压力测试工具，它是由 Lionbridge公司开发。

Webbech能测试处在相同硬件上，不同服务的性能以及不同硬件上同一个服务的运行状况。webBech的标准测试可以向我们展示服务器的 两项 内容：每秒钟相应请求数和每秒钟传输数据量。webbench不但能具有便准静态页面的测试能力，还能对动态页面（ASP,PHP,JAVA,CGI）进 行测试的能力。还有就是他支持对含有SSL的安全网站例如电子商务网站进行静态或动态的性能测试。

Webbench最多可以模拟3万个并发连接去测试网站的负载能力，其源码仅仅500行左右。

二.源码解读

源码一共包含两个源文件：socket.c 和 webbench.c 。

1. socket.c

本文件仅包含一个Socket函数：

int Socket(const char *host, int clientPort)
{
/*
* 以host和clientPort构成一对TCP的套接字（host支持域名）
* 创建失败返回-1，成功返回一个sockt描述符
*
* 涉及的知识点：
* inet_addr：若字符串有效则将字符串转换为32位二进制
* 网络字节序的IPV4地址，否则为INADDR_NONE；
* gethostbyname：返回对应于给定主机名的包含主机名字
* 和地址信息的hostent结构指针；
* htons：将一个无符号短整型(s)的主机数值(h)转换为网络字节顺序(n)；
* socket：创建套接字
* connect：连接到相应的主机
*/
int sock;
unsigned long inaddr;
struct sockaddr_in ad;
struct hostent *hp;

memset(&ad, 0, sizeof(ad));
ad.sin_family = AF_INET;

inaddr = inet_addr(host);
if (inaddr != INADDR_NONE)
memcpy(&ad.sin_addr, &inaddr, sizeof(inaddr));
else
{
hp = gethostbyname(host);
if (hp == NULL)
return -1;
memcpy(&ad.sin_addr, hp->h_addr, hp->h_length);
}
ad.sin_port = htons(clientPort);

sock = socket(AF_INET, SOCK_STREAM, 0);
if (sock < 0)
return sock;
if (connect(sock, (struct sockaddr *)&ad, sizeof(ad)) < 0)
return -1;
return sock;
}

2.webbench.c

在webbench.c中，包含了下面几个函数：

static void alarm_handler(int signal);
static void usage(void);
static void build_request(const char *url);
static int bench(void);
static void benchcore(const char *host, const int port, const char *req);
int main(int argc, char *argv[]);

- Webbench工作流程：

进入main函数，解析命令行参数；

调用build_request()函数，构造http request信息；

调用bench()函数，根据client大小fork子进程；

子进程调用benchcore()函数进行测试，挂接信号处理函数alarm_handler，将结果写入管道。

主进程从管道读取消息，并输出到stdout。

下面逐个部分进行分析：

- 全局变量

/* values */
/* 超时标识，由命令行参数-t指定，当被设置为1时，所有子进程退出
* volatile：
*    - 让系统总是从内存读取数据，
*    - 告诉编译器不要做任何优化，
*    - 变量会在程序外被改变
*/
volatile int timerexpired=0;
int speed=0; //子进程成功得到服务器响应的总数
int failed=0; //子进程请求失败总数
int bytes=0; //读取到的字节总数
/* globals */
int http10=1; /* 0 - http/0.9, 1 - http/1.0, 2 - http/1.1 */ //HTTP协议版本
/* Allow: GET, HEAD, OPTIONS, TRACE */
#define METHOD_GET 0
#define METHOD_HEAD 1
#define METHOD_OPTIONS 2
#define METHOD_TRACE 3
#define PROGRAM_VERSION "1.5"
int method=METHOD_GET; //HTTP请求方法，默认为GET
int clients=1; //并发数，即子进程个数，默认为1，可以由命令行参数-c指定
int force=0; //是否等待从服务器获取数据，0表示等待
int force_reload=0; //是否使用cache，0为使用
int proxyport=80; //代理服务器端口，默认为80
char *proxyhost=NULL; //代理服务器IP，默认为NULL
/*
* 执行时间，默认为30秒，可由命令行参数-t指定，
* 当子进程执行时间超过这个秒数之后，
* 发送 SIGALRM 信号，将 timerexpired 设置为 1，
* 让所有子进程退出
*/
int benchtime=30;
/* internal */
int mypipe[2]; //创建管道(半双工)，用于父子进程间通信
char host[MAXHOSTNAMELEN]; //服务器IP
#define REQUEST_SIZE 2048
char request[REQUEST_SIZE]; //HTTP请求信息

/* 命令行的选项配置表，参见man文档：man getopt_long */
static const struct option long_options[]=
{
{"force",no_argument,&force,1},
{"reload",no_argument,&force_reload,1},
{"time",required_argument,NULL,'t'},
{"help",no_argument,NULL,'?'},
{"http09",no_argument,NULL,'9'},
{"http10",no_argument,NULL,'1'},
{"http11",no_argument,NULL,'2'},
{"get",no_argument,&method,METHOD_GET},
{"head",no_argument,&method,METHOD_HEAD},
{"options",no_argument,&method,METHOD_OPTIONS},
{"trace",no_argument,&method,METHOD_TRACE},
{"version",no_argument,NULL,'V'},
{"proxy",required_argument,NULL,'p'},
{"clients",required_argument,NULL,'c'},
{NULL,0,NULL,0}
};

- main()函数

int main(int argc, char *argv[])
{
int opt=0;
int options_index=0;
char *tmp=NULL;

if(argc==1)
{
usage();
return 2;
}

/* 参数检查，并设置相应选项，参见 man getopt_long */
while((opt=getopt_long(argc,argv,"912Vfrt:p:c:?h",long_options,&options_index))!=EOF )
{
switch(opt)
{
case  0 : break;
case 'f': force=1;break;
case 'r': force_reload=1;break;
case '9': http10=0;break;
case '1': http10=1;break;
case '2': http10=2;break;
case 'V': printf(PROGRAM_VERSION"\n");exit(0);
case 't': benchtime=atoi(optarg);break;
case 'p':
/* proxy server parsing server:port */
tmp=strrchr(optarg,':');
proxyhost=optarg;
if(tmp==NULL)
{
break;
}
if(tmp==optarg)
{
fprintf(stderr,"Error in option --proxy %s: Missing hostname.\n",optarg);
return 2;
}
if(tmp==optarg+strlen(optarg)-1)
{
fprintf(stderr,"Error in option --proxy %s Port number is missing.\n",optarg);
return 2;
}
*tmp='\0';
proxyport=atoi(tmp+1);break;
case ':':
case 'h':
case '?': usage();return 2;break;
case 'c': clients=atoi(optarg);break;
}
}

/* 扫描参数选项时，optind标识下一个选项的索引；
* 扫描结束后，标识第一个非选项参数索引;
* 如果optind=argc，说明非选项参数即服务器URL缺失。
* 此变量是系统定义的。
*/
if(optind==argc) {
fprintf(stderr,"webbench: Missing URL!\n");
usage();
return 2;
}

if(clients==0) clients=1;
if(benchtime==0) benchtime=60;
/* Copyright */
fprintf(stderr,"Webbench - Simple Web Benchmark "PROGRAM_VERSION"\n"
"Copyright (c) Radim Kolar 1997-2004, GPL Open Source Software.\n"
);
build_request(argv[optind]); //最后为URL，所以optind应该是URL的位置
/* print bench info */
printf("\nBenchmarking: ");
switch(method)
{
case METHOD_GET:
default:
printf("GET");break;
case METHOD_OPTIONS:
printf("OPTIONS");break;
case METHOD_HEAD:
printf("HEAD");break;
case METHOD_TRACE:
printf("TRACE");break;
}
printf(" %s",argv[optind]);
switch(http10)
{
case 0: printf(" (using HTTP/0.9)");break;
case 2: printf(" (using HTTP/1.1)");break;
}
printf("\n");
if(clients==1) printf("1 client");
else
printf("%d clients",clients);

printf(", running %d sec", benchtime);
if(force) printf(", early socket close");
if(proxyhost!=NULL) printf(", via proxy server %s:%d",proxyhost,proxyport);
if(force_reload) printf(", forcing reload");
printf(".\n");
return bench();
}

- build_request()函数

/****************
创建URL请求连接,HTTP头
@url:url地址
创建好的请求放在全局变量request中
****************/
void build_request(const char *url)
{
char tmp[10];
int i;

bzero(host,MAXHOSTNAMELEN);
bzero(request,REQUEST_SIZE);

//协议适配
if(force_reload && proxyhost!=NULL && http10<1) http10=1;
if(method==METHOD_HEAD && http10<1) http10=1;
if(method==METHOD_OPTIONS && http10<2) http10=2;
if(method==METHOD_TRACE && http10<2) http10=2;

switch(method)
{
default:
case METHOD_GET: strcpy(request,"GET");break;
case METHOD_HEAD: strcpy(request,"HEAD");break;
case METHOD_OPTIONS: strcpy(request,"OPTIONS");break;
case METHOD_TRACE: strcpy(request,"TRACE");break;
}

strcat(request," "); //追加空格

if(NULL==strstr(url,"://")) //strstr(str1,str2)用于判断str2是否是str1的子串
{
fprintf(stderr, "\n%s: is not a valid URL.\n",url);
exit(2);
}
if(strlen(url)>1500)
{
fprintf(stderr,"URL is too long.\n");
exit(2);
}
if(proxyhost==NULL)
if (0!=strncasecmp("http://",url,7)) //比较前7个字符串
{ fprintf(stderr,"\nOnly HTTP protocol is directly supported, set --proxy for others.\n");
exit(2);
}
/* protocol/host delimiter */
i=strstr(url,"://")-url+3;
/* printf("%d\n",i); */

if(strchr(url+i,'/')==NULL) { //url+i指向http://后第一个位置
fprintf(stderr,"\nInvalid URL syntax - hostname don't ends with '/'.\n");
exit(2);
}
if(proxyhost==NULL)
{
/* get port from hostname */
if(index(url+i,':')!=NULL &&
index(url+i,':')<index(url+i,'/')) //判断url中是否指定了端口号
{
strncpy(host,url+i,strchr(url+i,':')-url-i); //取出主机地址
bzero(tmp,10);
strncpy(tmp,index(url+i,':')+1,strchr(url+i,'/')-index(url+i,':')-1);
/* printf("tmp=%s\n",tmp); */
proxyport=atoi(tmp); //端口号转换为int
if(proxyport==0) proxyport=80;
} else
{
strncpy(host,url+i,strcspn(url+i,"/"));
}
// printf("Host=%s\n",host);
strcat(request+strlen(request),url+i+strcspn(url+i,"/"));
} else
{
// printf("ProxyHost=%s\nProxyPort=%d\n",proxyhost,proxyport);
strcat(request,url);
}
if(http10==1)
strcat(request," HTTP/1.0");
else if (http10==2)
strcat(request," HTTP/1.1");
strcat(request,"\r\n");
if(http10>0)
strcat(request,"User-Agent: WebBench "PROGRAM_VERSION"\r\n");
if(proxyhost==NULL && http10>0)
{
strcat(request,"Host: ");
strcat(request,host);
strcat(request,"\r\n");
}
if(force_reload && proxyhost!=NULL)
{
strcat(request,"Pragma: no-cache\r\n");
}
if(http10>1)
strcat(request,"Connection: close\r\n");
/* add empty line at end */
if(http10>0) strcat(request,"\r\n");
// printf("Req=%s\n",request);
}

- bench()函数

/**********************
创建管道和子进程，对http请求进行测试
**********************/
/* vraci system rc error kod */
static int bench(void)
{
int i,j,k;
pid_t pid=0;
FILE *f;

/* check avaibility of target server */
/* 测试远程主机是否能够连通 */
i=Socket(proxyhost==NULL?host:proxyhost,proxyport);
if(i<0) {
fprintf(stderr,"\nConnect to server failed. Aborting benchmark.\n");
return 1;
}
close(i);
/* create pipe */
if(pipe(mypipe))
{
perror("pipe failed.");
return 3;
}

/* not needed, since we have alarm() in childrens */
/* wait 4 next system clock tick */
/*
cas=time(NULL);
while(time(NULL)==cas)
sched_yield();
*/

/* fork childs */
for(i=0;i<clients;i++)
{
pid=fork();
if(pid <= (pid_t) 0) //pid=0->子进程,pid<0->error
{
/* child process or error*/
sleep(1); /* make childs faster */
break;
/* 子进程跳出循环orfork出错父进程跳出循环 */
}
}

if( pid< (pid_t) 0) //fork出错
{
fprintf(stderr,"problems forking worker no. %d\n",i);
perror("fork failed.");
return 3;
}

if(pid== (pid_t) 0) //子进程
{
/* I am a child */
/* 子进程执行请求：尽可能多的发送请求，直到超时返回为止 */
if(proxyhost==NULL)
benchcore(host,proxyport,request);
else
benchcore(proxyhost,proxyport,request);

/* write results to pipe */
f=fdopen(mypipe[1],"w");
if(f==NULL)
{
perror("open pipe for writing failed.");
return 3;
}
/* fprintf(stderr,"Child - %d %d\n",speed,failed); */
fprintf(f,"%d %d %d\n",speed,failed,bytes);
fclose(f);
/* 子进程完成任务，返回退出 */
return 0;
} else
{
/* 父进程读取管道，打印结果 */
f=fdopen(mypipe[0],"r");
if(f==NULL)
{
perror("open pipe for reading failed.");
return 3;
}
/* _IONBF(无缓冲）：直接从流中读入数据或直接向流中写入数据，而没有缓冲区 */
setvbuf(f,NULL,_IONBF,0);
/* 虽然子进程不能污染父进程的这几个变量，但用前重置一下，在这里是个好习惯 */
speed=0;
failed=0;
bytes=0;
/* 从管道中读取每个子进程的任务执行情况，并计数 */
while(1)
{
pid=fscanf(f,"%d %d %d",&i,&j,&k);
if(pid<2)
{
fprintf(stderr,"Some of our childrens died.\n");
break;
}
speed+=i;
failed+=j;
bytes+=k;
/* fprintf(stderr,"*Knock* %d %d read=%d\n",speed,failed,pid); */
if(--clients==0) break;
}
fclose(f);

printf("\nSpeed=%d pages/min, %d bytes/sec.\nRequests: %d susceed, %d failed.\n",
(int)((speed+failed)/(benchtime/60.0f)),
(int)(bytes/(float)benchtime),
speed,
failed);
}
return i;
}

- benchcore()函数

/******************************
真正测试http的地方
@host:地址
@port：端口
@req：http格式方法
********************************/
void benchcore(const char *host,const int port,const char *req)
{
int rlen;
char buf[1500];
int s,i;
struct sigaction sa;

/* setup alarm signal handler */
/* 当程序执行到指定的秒数之后，发送 SIGALRM 信号 */
sa.sa_handler=alarm_handler;
sa.sa_flags=0;
if(sigaction(SIGALRM,&sa,NULL))
exit(3);
alarm(benchtime);

rlen=strlen(req);
/* 无限执行请求，直到接收到 SIGALRM 信号将 timerexpired 设置为 1 时 */
nexttry:while(1)
{
if(timerexpired)
{
if(failed>0)
{
/* fprintf(stderr,"Correcting failed by signal\n"); */
failed--;
}
return;
}
/* 连接远程服务器 */
s=Socket(host,port);
if(s<0) { failed++;continue;}
/* 发送请求 */
if(rlen!=write(s,req,rlen)) {failed++;close(s);continue;}
/* 如果是 http/0.9 则关闭 socket 的写操作 */
if(http10==0)
if(shutdown(s,1)) { failed++;close(s);continue;}
/* 如果等待响应数据返回，则读取响应数据，计算传输的字节数 */
if(force==0)
{
/* read all available data from socket */
while(1)
{
if(timerexpired) break;
i=read(s,buf,1500);
/* fprintf(stderr,"%d\n",i); */
if(i<0)
{
failed++;
close(s);
goto nexttry;
}
else
if(i==0) break;
else
bytes+=i;
}
}
/* 关闭连接 */
if(close(s)) {failed++;continue;}
/* 成功完成一次请求，并计数，继续下一次相同的请求，直到超时为止 */
speed++;
}
}

三.测试

$ webbench www.jiangezh.com
Webbench - Simple Web Benchmark 1.5
Copyright (c) Radim Kolar 1997-2004, GPL Open Source Software.

www.jiangezh.com: is not a valid URL.

$ webbench ftp://www.jiangezh.com Webbench - Simple Web Benchmark 1.5
Copyright (c) Radim Kolar 1997-2004, GPL Open Source Software.

Only HTTP protocol is directly supported, set --proxy for others.

$ webbench http://www.jiangezh.com Webbench - Simple Web Benchmark 1.5
Copyright (c) Radim Kolar 1997-2004, GPL Open Source Software.

Invalid URL syntax - hostname don't ends with '/'.

$ webbench http://www.jiangezh.com:8000/ Webbench - Simple Web Benchmark 1.5
Copyright (c) Radim Kolar 1997-2004, GPL Open Source Software.

Benchmarking: GET http://www.jiangezh.com:8000/ 1 client, running 30 sec.

Connect to server failed. Aborting benchmark.

$ webbench http://www.jiangezh.com/ Webbench - Simple Web Benchmark 1.5
Copyright (c) Radim Kolar 1997-2004, GPL Open Source Software.

Benchmarking: GET http://www.jiangezh.com/ 1 client, running 30 sec.

Speed=126 pages/min, 83208 bytes/sec.
Requests: 63 susceed, 0 failed.

$ webbench -t 10 -c 20
webbench: Missing URL!
webbench [option]... URL
-f|--force               Don't wait for reply from server.
-r|--reload              Send reload request - Pragma: no-cache.
-t|--time <sec>          Run benchmark for <sec> seconds. Default 30.
-p|--proxy <server:port> Use proxy server for request.
-c|--clients <n>         Run <n> HTTP clients at once. Default one.
-9|--http09              Use HTTP/0.9 style requests.
-1|--http10              Use HTTP/1.0 protocol.
-2|--http11              Use HTTP/1.1 protocol.
--get                    Use GET request method.
--head                   Use HEAD request method.
--options                Use OPTIONS request method.
--trace                  Use TRACE request method.
-?|-h|--help             This information.
-V|--version             Display program version.

$ webbench -t 10 -c 500 http://www.jiangezh.com/article/SYNFloodSrc.html Webbench - Simple Web Benchmark 1.5
Copyright (c) Radim Kolar 1997-2004, GPL Open Source Software.

Benchmarking: GET http://www.jiangezh.com/article/SYNFloodSrc.html 500 clients, running 10 sec.

Speed=126 pages/min, 121629 bytes/sec.
Requests: 21 susceed, 0 failed.

呜呼T^T

发现我的博客网站太菜了扛不起多个连接==

不过没关系！因为根本就没什么访问量！