jstat通常是作为使用查看gc 状态,编译信息等具体信息,加载class量等,笔者在前面的博客里谈了jmap,jstack工具的如何实现可具体参考(http://blog.csdn.net/raintungli/article/details/7023092) ,谈到通常的方法必须让jvm的所有线程进入一个safe point的状态,也就是在使用jmap, jstack工具的时候会有线程的停顿感,并且影响到当时jvm所正在处理的事情,如果在产线上这是会影响客户体验的。
在调用jstat 的时候,并没有感觉到线程的停顿,也就是jstat 使用与jmap,jstack完全不同的方式,接着还是源码分析来解释。
在jstat.java中初始化LocalMonitoredVM中,初始化了PerfDataBuffer, 在PerfDataBuffer是通过Perf attach vm的进程生成byte buffer对象
private ByteBuffer attachImpl(String user, int lvmid, int mode)
throws IllegalArgumentException, IOException
{
final ByteBuffer b = attach(user, lvmid, mode);
if (lvmid == 0) {
// The native instrumentation buffer for this Java virtual
// machine is never unmapped.
return b;
}
else {
// This is an instrumentation buffer for another Java virtual
// machine with native resources that need to be managed. We
// create a duplicate of the native ByteBuffer and manage it
// with a Cleaner object (PhantomReference). When the duplicate
// becomes only phantomly reachable, the native resources will
// be released.
final ByteBuffer dup = b.duplicate();
Cleaner.create(dup, new Runnable() {
public void run() {
try {
instance.detach(b);
}
catch (Throwable th) {
// avoid crashing the reference handler thread,
// but provide for some diagnosability
assert false : th.toString();
}
}
});
return dup;
}
}
private native ByteBuffer attach(String user, int lvmid, int mode)
throws IllegalArgumentException, IOException;
关键就在native程序中, 查看perf.cpp 中对应的方法
PerfMemory::attach(user_utf, vmid, (PerfMemory::PerfMemoryMode) mode,
&address, &capacity, CHECK_NULL);
也就是类PerfMemory的attach方法
在perfMemory_linux.cpp中,我们看到了attach 的实现
void PerfMemory::attach(const char* user, int vmid, PerfMemoryMode mode, char** addrp, size_t* sizep, TRAPS) {
if (vmid == 0 || vmid == os::current_process_id()) {
*addrp = start();
*sizep = capacity();
return;
}
mmap_attach_shared(user, vmid, mode, addrp, sizep, CHECK);
}
在mmap_attach_shared函数中,
char* filename = get_sharedmem_filename(dirname, vmid);
// copy heap memory to resource memory. the open_sharedmem_file
// method below need to use the filename, but could throw an
// exception. using a resource array prevents the leak that
// would otherwise occur.
char* rfilename = NEW_RESOURCE_ARRAY(char, strlen(filename) + 1);
strcpy(rfilename, filename);
// free the c heap resources that are no longer needed
if (luser != user) FREE_C_HEAP_ARRAY(char, luser);
FREE_C_HEAP_ARRAY(char, dirname);
FREE_C_HEAP_ARRAY(char, filename);
// open the shared memory file for the give vmid
fd = open_sharedmem_file(rfilename, file_flags, CHECK);
assert(fd != OS_ERR, "unexpected value");
if (*sizep == 0) {
size = sharedmem_filesize(fd, CHECK);
assert(size != 0, "unexpected size");
}
mapAddress = (char*)::mmap((char*)0, size, mmap_prot, MAP_SHARED, fd, 0);
该函数把一个文件mmap成内存,然后通过读取内存来读取该文件的内容,而文件的路径就是/tmp/hsperfdata_{userid}/{pid}
而在create_shared_memory函数中,也就是jvm 在vm_init_globals初始化后会创建这个文件并且用mmap映射成内存,接着虚拟机会把gc 或者其他需要的状态写入这个mmap文件中,jstat工具通过访问该文件,读取内容,显示在屏幕上,该文件通过内存映射实现,没有什么性能损耗,完整的jstat 的流程就全部走通了。