Java线程创建流程

Question

我正在阅读 Brian Goetz 的 ThreadPools 部分 6.2.3 java concurrency in practice。我遇到的一个说法是 “重用现有线程而不是创建新线程可以分摊线程创建和拆卸成本。”

1) 我想获得 java 线程创建过程中涉及的一些指标，我们知道这将涉及堆栈的创建/分配和程序计数器寄存器到创建的线程。是否有我可以使用的工具/实用程序/可视化虚拟机跟踪器/jmx bean，它可以为我提供一些关于线程创建的内存和时间使用情况的指标。有人可以指导我做同样的事情吗？

2) 是否有文本可以详细指导我创建 java 线程的整个过程，其中应该涵盖对 windows 的各个操作系统调用？

Why is creating a Thread said to be expensive?给了我一些信息，但我想详细研究java线程创建的内部结构

谢谢

Answer 1

关于你的第二个问题:

2) Is there a text which can guide me to the whole process of java thread creation in detail which should cover the respective OS calls to windows ?

本指南虽然有点偏离主题，但很棒:http://blog.jamesdbloom.com/JVMInternals.html

和下面的书The Java Virtual Machine Specification, Java SE 7 Edition (google book link)深入解释 JVM 内部结构，因此这将是我最好的选择(免责声明:我只是浏览可见的部分)

如果这还不够好，您可以随时 download open jdk (7) 的源代码并爬取代码...

打开的 jdk 代码的摘录是这样的(openjdk\hotspot\src\share\vm\runtime\thread.cpp - c'tor):

// Base class for all threads: VMThread, WatcherThread, ConcurrentMarkSweepThread,
// JavaThread


Thread::Thread() {
  // stack and get_thread
  set_stack_base(NULL);
  set_stack_size(0);
  set_self_raw_id(0);
  set_lgrp_id(-1);

  // allocated data structures
  set_osthread(NULL);
  set_resource_area(new ResourceArea());
  set_handle_area(new HandleArea(NULL));
  set_active_handles(NULL);
  set_free_handle_block(NULL);
  set_last_handle_mark(NULL);

  // This initial value ==> never claimed.
  _oops_do_parity = 0;

  // the handle mark links itself to last_handle_mark
  new HandleMark(this);

  // plain initialization
  debug_only(_owned_locks = NULL;)
  debug_only(_allow_allocation_count = 0;)
  NOT_PRODUCT(_allow_safepoint_count = 0;)
  NOT_PRODUCT(_skip_gcalot = false;)
  CHECK_UNHANDLED_OOPS_ONLY(_gc_locked_out_count = 0;)
  _jvmti_env_iteration_count = 0;
  set_allocated_bytes(0);
  _vm_operation_started_count = 0;
  _vm_operation_completed_count = 0;
  _current_pending_monitor = NULL;
  _current_pending_monitor_is_from_java = true;
  _current_waiting_monitor = NULL;
  _num_nested_signal = 0;
  omFreeList = NULL ;
  omFreeCount = 0 ;
  omFreeProvision = 32 ;
  omInUseList = NULL ;
  omInUseCount = 0 ;

  _SR_lock = new Monitor(Mutex::suspend_resume, "SR_lock", true);
  _suspend_flags = 0;

  // thread-specific hashCode stream generator state - Marsaglia shift-xor form
  _hashStateX = os::random() ;
  _hashStateY = 842502087 ;
  _hashStateZ = 0x8767 ;    // (int)(3579807591LL & 0xffff) ;
  _hashStateW = 273326509 ;

  _OnTrap   = 0 ;
  _schedctl = NULL ;
  _Stalled  = 0 ;
  _TypeTag  = 0x2BAD ;

  // Many of the following fields are effectively final - immutable
  // Note that nascent threads can't use the Native Monitor-Mutex
  // construct until the _MutexEvent is initialized ...
  // CONSIDER: instead of using a fixed set of purpose-dedicated ParkEvents
  // we might instead use a stack of ParkEvents that we could provision on-demand.
  // The stack would act as a cache to avoid calls to ParkEvent::Allocate()
  // and ::Release()
  _ParkEvent   = ParkEvent::Allocate (this) ;
  _SleepEvent  = ParkEvent::Allocate (this) ;
  _MutexEvent  = ParkEvent::Allocate (this) ;
  _MuxEvent    = ParkEvent::Allocate (this) ;

#ifdef CHECK_UNHANDLED_OOPS
  if (CheckUnhandledOops) {
    _unhandled_oops = new UnhandledOops(this);
  }
#endif // CHECK_UNHANDLED_OOPS
#ifdef ASSERT
  if (UseBiasedLocking) {
    assert((((uintptr_t) this) & (markOopDesc::biased_lock_alignment - 1)) == 0, "forced alignment of thread object failed");
    assert(this == _real_malloc_address ||
           this == (void*) align_size_up((intptr_t) _real_malloc_address, markOopDesc::biased_lock_alignment),
           "bug in forced alignment of thread objects");
  }
#endif /* ASSERT */
}