Linux利器strace
strace常用来跟踪进程执行时的系统调用和所接收的信号。 在Linux世界,进程不能直接访问硬件设备,当进程需要访问硬件设备(比如读取磁盘文件,接收网络数据等等)时,必须由用户态模式切换至内核态模式,通 过系统调用访问硬件设备。strace可以跟踪到一个进程产生的系统调用,包括参数,返回值,执行消耗的时间。
strace使用参数
-p 跟踪指定的进程
-f 跟踪由fork子进程系统调用
-F 尝试跟踪vfork子进程系统调吸入,与-f同时出现时, vfork不被跟踪
-o filename 默认strace将结果输出到stdout。通过-o可以将输出写入到filename文件中
-ff 常与-o选项一起使用,不同进程(子进程)产生的系统调用输出到filename.PID文件
-r 打印每一个系统调用的相对时间
-t 在输出中的每一行前加上时间信息。-tt 时间确定到微秒级。还可以使用-ttt打印相对时间
-v 输出所有系统调用。默认情况下,一些频繁调用的系统调用不会输出
-s 指定每一行输出字符串的长度,默认是32。文件名一直全部输出
-c 统计每种系统调用所执行的时间,调用次数,出错次数。
-e expr 输出过滤器,通过表达式,可以过滤出掉你不想要输出
应用场景
#1.跟踪你的web服务器系统调用
系统调用优化,也是web性能优化的一个较为重要的方向,尤其是在I/O密集型web应用的情况。我们这里的测试环境是CentOS5.4+Nginx+FastCGI。
<?php
//file:hello.php
define
(
'DOCUMENT_ROOT'
,
dirname
(
__FILE__
)
)
;
include
(
"hello.inc"
)
;
include
(
"./hello.inc"
)
;
include
(
DOCUMENT_ROOT .
"/hello.inc"
)
;
?>
#strace -f -F -o strace_nginx strace /wwwchroot/nginx/sbin/nginx -c /wwwchroot/nginx/nginx.conf
... (有部分不重要的数据影响排版,在这里使用...代替)
//--接受来自客户端的http请求
4165 recv(16, "GET /hello.php HTTP/1.1\r\nHost: f"..., 32768, 0) = 391
4165 epoll_ctl(9, EPOLL_CTL_MOD, 16, {EPOLLIN|EPOLLOUT|EPOLLET, {u32=3081162952, u64=698098541354471624}}) = 0
//--进行DNS查找
4165 getsockname(16, {sa_family=AF_INET, sin_port=htons(80), sin_addr=inet_addr("222.73.211.214")}, [16]) = 0
//--新建一个socket,连接Fast-CGI,端口号为9000
4165 socket(PF_INET, SOCK_STREAM, IPPROTO_IP) = 17
4165 ioctl(17, FIONBIO, [1]) = 0
4165 epoll_ctl(9, EPOLL_CTL_ADD, 17, {EPOLLIN|EPOLLOUT|EPOLLET, {u32=3081163048, u64=697886249710965032}}) = 0
4165 connect(17, {sa_family=AF_INET, sin_port=htons(9000), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 )
4165 epoll_wait(9, {{EPOLLOUT, {u32=3081163048, u64=697886249710965032}}, {...}, 5\
12, 300000) = 2
4165 gettimeofday({1295420285, 130967}, NULL) = 0
4165 recv(16, 0xbfdd7d8b, 1, MSG_PEEK) = -1 EAGAIN (Resource temporarily unavailable)
4165 getsockopt(17, SOL_SOCKET, SO_ERROR, [0], [4]) = 0
//--将用户http请求交给Fast-CGI
4165 writev(17, [{"\1\1\0\1\0\10\0\0\0\1\0\0\0\0\0\0\1\4\0\1\3\30\0\0\21\7GATEWA"..., 832}], 1) = 832
4165 epoll_wait(9, {{EPOLLIN|EPOLLOUT, {u32=3081163048, u64=697886249710965032}}}, 512, 300000) = 1
4165 gettimeofday({1295420285, 131559}, NULL) = 0
//--接收Fast-CGI响应
4165 recv(17, "\1\6\0\1\0V\2\0X-Powered-By: PHP/5.2.10"..., 65536, 0) = 112
4165 readv(17, [{"\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0\0"..., 65424}], 1) = 0
4165 mmap2(NULL, 274432, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7514000
4165 close(17) = 0
4165 munmap(0xb7514000, 274432) = 0
//-- 响应客户端http请求,即http响应
4165 writev(16, [{"HTTP/1.1 200 OK\r\nServer: nginx/0"..., 228}, {"22\r\n", 4}, ..., 5) = 273
4165 write(5, "116.66.34.82 - - [19/Jan/2011:14"..., 191) = 191
4165 setsockopt(16, SOL_TCP, TCP_NODELAY, [1], 4) = 0
4165 recv(16, 0x9b024e8, 32768, 0) = -1 EAGAIN (Resource temporarily unavailable)
...
通过这些,我们只能够大概地了解,Nginx这里启用了epoll。同时,还可以了解到Nginx和Fast-CGI底层是如何运作的。奇 怪,hello.php文件中有三个inclue,即加载了三次文件,这里没有看到相应的i/o逻辑操作,是为什么呢?这是因为,Nginx并没解析处理PHP脚本,而是交给Fast-CGI去做这部事情了。
#strace -f -F -o php-cgi-strace /wwwchroot/php/bin/php-cgi --fpm-config /wwwchroot/php/etc/php-fpm.conf
//--接收来自Nginx发出的请求
4510 <... accept resumed> {sa_family=AF_INET, sin_port=htons(35983), sin_addr=inet_addr("127.0.0.1")}, [16]) = 3
4510 clock_gettime(CLOCK_MONOTONIC, {22638545, 869965681}) = 0
4510 poll([{fd=3, events=POLLIN}], 1, 5000) = 1 ([{fd=3, revents=POLLIN}])
4510 read(3, "\1\1\0\1\0\10\0\0", 8) = 8
4510 read(3, "\0\1\0\0\0\0\0\0", 8) = 8
4510 read(3, "\1\4\0\1\0035\3\0", 8) = 8
4510 read(3, "\21\7GATEWAY_INTERFACECGI/1.1\17\5SERV"..., 824) = 824
4510 read(3, "\1\4\0\1\0\0\0\0", 8) = 8
4510 time(NULL) = 1295425149
//--加载请求资源文件hello.php
4510 lstat64("/var", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
4510 lstat64("/var/www", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
4510 lstat64("/var/www/ep", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
4510 lstat64("/var/www/ep/hello.php", {st_mode=S_IFREG|0644, st_size=119, ...}) = 0
4510 clock_gettime(CLOCK_MONOTONIC, {22638545, 870893872}) = 0
4510 setitimer(ITIMER_PROF, {it_interval={0, 0}, itvalue={60, 0}}, NULL) = 0
4510 rt_sigaction(SIGPROF, {0x835c120, [PROF], SA_RESTART}, {SIG_DFL, [], 0}, 8) = 0
4510 rt_sigprocmask(SIG_UNBLOCK, [PROF], NULL, 8) = 0
4510 time(NULL) = 1295425149
4510 open("/var/www/ep/hello.php", O_RDONLY) = 4
4510 fstat64(4, {st_mode=S_IFREG|0644, st_size=119, ...}) = 0
4510 time(NULL) = 1295425149
4510 chdir("/var/www/ep") = 0
4510 fstat64(4, {st_mode=S_IFREG|0644, st_size=119, ...}) = 0
4510 mmap2(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0xb7fe7000
4510 read(4, "\n", 8192) = 29
4510 read(4, "", 8192) = 0
4510 read(4, "", 8192) = 0
4510 close(4) = 0
//-- 加载hello.inc, 对应php代码include './hello.inc'
4510 getcwd("/var/www/ep"..., 4096) = 12
4510 time(NULL) = 1295425149
4510 open("/var/www/ep/hello.inc", O_RDONLY) = 4
4510 fstat64(4, {st_mode=S_IFREG|0644, st_size=29, ...}) = 0
4510 read(4, "\n", 8192) = 29
4510 read(4, "", 8192) = 0
4510 read(4, "", 8192) = 0
4510 close(4) = 0
4510 time(NULL) = 1295425149
//-- 加载hello.inc, 对应php代码include DOCUMENT_ROOT . '/hello.inc'
4510 lstat64("/var", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
4510 lstat64("/var/www", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
4510 lstat64("/var/www/ep", {st_mode=S_IFDIR|0755, st_size=4096, ...}) = 0
4510 lstat64("/var/www/ep/hello.inc", {st_mode=S_IFREG|0644, st_size=29, ...}) = 0
4510 open("/var/www/ep/hello.inc", O_RDONLY) = 4
4510 fstat64(4, {st_mode=S_IFREG|0644, st_size=29, ...}) = 0
4510 read(4, "\n", 8192) = 29
4510 read(4, "", 8192) = 0
4510 read(4, "", 8192) = 0
4510 close(4) = 0
//-- 将响结果输出给Nginx,并且关闭连接
4510 write(3, "\1\6\0\1\0V\2\0X-Powered-By: PHP/5.2.10"..., 96) = 96
4510 setitimer(ITIMER_PROF, {it_interval={0, 0}, itvalue={0, 0}}, NULL) = 0
4510 write(3, "\1\3\0\1\0\10\0\0\0\0\0\0\0ere", 16) = 16
4510 shutdown(3, 1 /* send */) = 0
4510 recv(3, "\1\5\0\1\0\0\0\0", 8, 0) = 8
4510 recv(3, "", 8, 0) = 0
4510 close(3) = 0
通过跟踪php-cgi,我们可以知道,相较与其它二种方法include ‘./hello.inc’的性能是最高的。这里看到strace输出都被截断了,如果你需要看到更多的输出,可以通过-s选项,让strace输出更多内容。
当你发现某个http请求造成CPU占用效骤然升高,你可以通过strace跟踪查找问题的根源。同时,你也可以通过strace -c统计监控你的优化是否生效
#2. MySQL执行语句列表
当发生个http请求的时候,很多时候希望得到这个http请求发生了多少次数据库SELECT操作,是否在同一个mysql connection连接里面完成。这里以访问本页为例子,通过strace来跟踪这些MySQL SELECT查询语句。
//-9514是mysqld的进程号,为了看到整条SQL语句,我们通过-s 1024希望输出更多内容
#strace -f -F -ff -o strace-mysqld -s 1024 -p 9514
#find . -name "strace-mysqld*" -type f -print |xargs grep -n "SELECT.*FROM wp_"
./strace-mysqld.19203:64:
read(19, "\3SELECT option_name, option_value FROM wp_options WHERE autoload = 'yes'", 72) = 72
./strace-mysqld.19203:165:
read(19, "\3SELECT * FROM wp_users WHERE user_login = 'admin'", 50) = 50
./strace-mysqld.19203:184:
read(19, "\3SELECT meta_key, meta_value FROM wp_usermeta WHERE user_id = 1", 63) = 63
./strace-mysqld.19203:295:
read(19, "\3SELECT option_value FROM wp_options WHERE option_name = 'rewrite_rules' LIMIT 1", 80) = 80
./strace-mysqld.19203:311:
read(19, "\3 SELECT wp_posts.* FROM wp_posts WHERE 1=1 AND wp_posts.ID = 501
AND wp_posts.post_type = 'post' ORDER BY wp_posts.post_date DESC ", 136) = 136
... (这里省去了一些)
其他
strace远不止这么强大,你可以善用之,我想你会离不开它的。同时,你还可以联合gdb和ltrace,你的工作会更加高效。