http://files.cnblogs.com/daizhj/DTrace.pdf
//
syscall
dtrace -ln 'syscall::write*:' //显示可使用的probe
dtrace -ln 'syscall::*read*:entry' //显示可使用的probe
dtrace -n 'syscall::write:entry {@dist[execname] = quantize(arg0)}' //之后CTRL+C
dtrace -n 'syscall::socket:entry {@dist[execname] = quantize(arg0)}' //之后CTRL+C
dtrace -n 'syscall:::entry { @sc[execname, probefunc] = count(); }' #dtrace -n 'syscall:::entry'
dtrace -n 'syscall::open:entry { printf("%s %s", execname, copyinstr(arg0)); }'
dtrace -n 'syscall::fork*: { trace(pid); }'
dtrace -n 'syscall::exec*: { trace(execname); }'
Showing Read Byte Distributions by Process
dtrace -n 'syscall::read:return { @[execname] = quantize(arg0); }'
一秒打印一次进程数
dtrace -n 'profile-997 { @[execname] = count(); } tick-1s { printa(@); trunc(@); }'
Most function calls will return from the same thread that they enter,6 so a thread- local variable can be used to associate these events. Here a time stamp is saved on the write(2) entry so that the time can be calculated on return:
dtrace -n 'syscall::write:entry { self->s = timestamp; } syscall::write:return /self->s/
syscall Provider
dtrace -n 'syscall:::entry { @[probefunc] = count(); }'
Which processes are executing the most system calls?
dtrace -n 'syscall:::entry { @[pid, execname] = count(); }'
What system calls are a given process name executing (for example, firefox-bin)?
dtrace -n 'syscall:::entry /execname == "firefox"/ { @[probefunc] = count(); }'
dtrace -qn 'syscall::read:entry,syscall::write:entry /fds[arg0].fi_fs == "sockfs"/ { @[probefunc] = sum(arg2); } tick-1sec { printa(@); trunc(@); }' #暂无打印
dtrace -n 'syscall::read:entry,syscall::write:entry /execname == "firefox" && fds[arg0].fi_fs == "sockfs"/ { @[execname,pid] = count(); }' #暂无打印
/
指定进程的MALLOC调用情况
dtrace -n 'pid$target::malloc:entry { @[ustack()] = quantize(arg0); }' -p 513
//
Disk I/O
dtrace -n 'io:::start { @[execname, pid] = count(); }'
bash-3.2# dtrace -qn 'syscall:::entry /execname == "firefox"/
{ @[pid, probefunc] = count(); } END { trunc(@, 10); printa(@); }'
dtrace -n 'syscall::pread*:entry,syscall::pwrite*:entry /execname == "java"/
{ @[fds[arg0].fi_fs] = count(); }'
dtrace -n 'syscall::pread*:entry,syscall::pwrite*:entry /execname == "java"/
{ @[fds[arg0].fi_pathname] = count(); }'
/
Memory
Tracking process user stack sizes:
Tracking which processes are growing their address space heap segment:
Tracking memory page faults by process name:
dtrace -n 'vminfo:::as_fault { @mem[execname] = sum(arg0); }'
Tracking pages paged in by process name:
dtrace -n 'vminfo:::pgpgin { @pg[execname] = sum(arg0); }'
Tracking pages paged out by process name:
dtrace -n 'vminfo:::pgpgout { @pg[execname] = sum(arg0); }'
sched Provider
dtrace -n 'sched:::on-cpu { @[pid, execname] = count(); }'
Tracking process user stack sizes:
dtrace -n 'sched:::on-cpu { @[execname] = max(curthread->t_procp->p_stksize);}'
Tracking which processes are growing their address space heap segment:
dtrace -n 'fbt::brk:entry { @mem[execname] = count(); }'
fbt Provider
Tracking which processes are growing their address space stack segment:
dtrace -n 'fbt::grow:entry { @mem[execname] = count(); }'
///
I/O
Which processes are executing common I/O system calls?
dtrace -n 'syscall::*read:entry,syscall::*write:entry { @rw[execname,probefunc] =
count(); }'
Which file system types are targeted for reads and writes?
dtrace -n 'syscall::*read:entry,syscall::*write:entry { @fs[execname, probefunc,fds[arg0].fi_fs] = count(); }'
Which files are being read, and by which processes?
dtrace -n 'syscall::*read:entry { @f[execname, fds[arg0].fi_pathname] = count(); }'
Which files are being written, and by which processes?
dtrace -n 'syscall::*write:entry { @f[execname, fds[arg0].fi_pathname] = count(); }'
Which processes are generating network I/O (Solaris)?
dtrace -n 'fbt:sockfs::entry { @[execname, probefunc] = count(); }' #暂时无法使用
What is the rate of disk I/O being issued?
dtrace -n 'io:::start { @io = count(); } tick-1sec { printa("Disk I/Os per second: %@d \n", @io); trunc(@io); }'
检查socket 调用情况
dtrace -n 'syscall::socket:entry { @[execname] = quantize(arg0); }'
dtrace -n 'syscall::write:entry /execname=="VineSample"/ { @[execname] = quantize(arg0); }'
//
lquantize解释:https://blogs.oracle.com/swan/entry/dtrace%E7%AE%80%E4%BB%8B_3
使用lquantize(所指定表达式的值的线性频率分布),我们了解需要调查的表达式的分布情况。比如,我们想知道系统调用write打开的文件描述符(file descriptor)的线性分布情况。
'syscall::write:entry{@fds[execname]=lquantize(arg0,0,100,1)}' #参数说明:标量表达式,下限,上限,步长值
dtrace: description 'syscall::write:entry' matched 1 probe
\^C
dtrace
value ------------- Distribution ------------- count
0 | 0
1 |@@@@@@@@@@@@@@@@@@@@ 1
2 | 0
sshd
value ------------- Distribution ------------- count
3 | 0
4 |@@@@@@@@@@@@@@@@@@@@ 1
5 | 0
6 | 0
7 | 0
8 |@@@@@@@@@@@@@@@@@@@@ 1
9 | 0
在上例中,我们可以看到,在该时间内,sshd进程对文件描述符4操作了1次,对文件描述符8操作了1次。虽然不具有实际意义,但可以帮助我们理解lquantize的作用。
如果要聚合的表达式的值非常大,使用lquantize可能会输出太多信息,这种情况下可以使用quantize(所指定表达式的值的二次方幂频率分布)来聚合。
下面是一个统计执行程序系统调用的时间分布的D脚本: time.d
#!/usr/sbin/dtrace -s
syscall:::entry
{
self->ts=timestamp;
}
syscall:::return
/self->ts/
{
@time[execname]=quantize(timestamp-self->ts);
}
执行一段时间,按Ctrl+C中断。限于篇幅,下面只列出部分信息。
# ./time.d
dtrace: script './time.d' matched 462 probes
\^C
sendmail
value ------------- Distribution ------------- count
1024 | 0
2048 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ 7
4096 |@@@@ 1
8192 |@@@@ 1
16384 | 0
sshd
value ------------- Distribution ------------- count
1024 | 0
2048 |@@@@@@@@@@@@@@@@@@@ 7
4096 |@@@@@ 2
8192 |@@@@@ 2
16384 |@@@@@ 2
32768 | 0
65536 |@@@@@ 2
131072 | 0
以sendmail程序为例:
系统调用执行时间(从entry到return)在大于等于2048纳秒并小于4096纳秒区间共有7次,在大于等于4096纳秒小于8192纳秒区间共有1次,在大于等于8192纳秒小于16384纳秒区间共有1次。