当corba服务器进程停止时，客户端调用被阻止

Question

这里的问题是corba调用不会返回，并且corba服务器停止时不会引发异常。 在我的情况下，只有一个多线程corba代理（Window），监视一个后端corba服务器。 corba服务器的IDL为：

  void run() 
      void echo(); 

代理通过echo心跳调用检查后端的健康状况。如果在echo中抛出corba异常，代理会将后端分类为DOWN状态。此过程在大多数时间都有效，但后端关闭。

1）如果我关闭后端机器，立即回声抛出异常。

2）如果我停止后端corba进程，echo调用挂起并且不返回，在客户端没有异常。客户不能再走向未来了。

以上两种情况都不会发生运行调用。

具有'ORBDebugLevel 10'的日志显示代理完成回应请求发送，而netstat showns在代理和后端机器之间执行一个TCP连接，尽管后端corba服务器进程停止（我承认后端服务器是无序的或编程不良）。但是作为代理，它是如何避免由于个别调用失败而被阻塞，如果它既不返回也不抛出异常？

下面是两个日志，默认策略：

TAO (276|1592) - Invocation_Adapter::invoke_i, making a TAO_CS_REMOTE_STRATEGY i 
nvocation 
TAO (276|1592) - Transport_Cache_Manager_T::is_entry_available_i[828], true, sta 
te is ENTRY_IDLE_AND_PURGABLE 
TAO (276|1592) - Cache_IntId_T::recycle_state, ENTRY_IDLE_AND_PURGABLE->ENTRY_BU 
SY Transport[828] IntId=00A64ABC 
TAO (276|1592) - Transport_Cache_Manager_T::find_i, Found available Transport[82 
8] @hash:index {-1062676757:0} 
TAO (276|1592) - Transport_Connector::connect, got an existing connected Transpo 
rt[828] in role TAO_CLIENT_ROLE 
TAO (276|1592) - Muxed_TMS[828]::request_id, <4> 
TAO (276|1592) - GIOP_Message_Base::dump_msg, send GIOP message v1.2, 60 data by 
tes, my endian, Type Request[4] 
GIOP message - HEXDUMP 72 bytes 
47 49 4f 50 01 02 01 00 3c 00 00 00 04 00 00 00 GIOP....<....... 
03 00 00 00 00 00 cd cd 1b 00 00 00 14 01 0f 00 ................ 
52 53 54 00 00 00 6c 00 06 9c b5 00 00 00 00 00 RST...l......... 
00 00 01 00 00 00 01 cd 05 00 00 00 65 63 68 6f ............echo 
00 cd cd cd 00 00 00 00       ........ 
TAO (276|1592) - Transport[828]::drain_queue_helper, sending 1 buffers 
TAO (276|1592) - Transport[828]::drain_queue_helper, buffer 0/1 has 72 bytes 
TAO - Transport[828]::drain_queue_helper (0/72) - HEXDUMP 72 bytes 
47 49 4f 50 01 02 01 00 3c 00 00 00 04 00 00 00 GIOP....<....... 
03 00 00 00 00 00 cd cd 1b 00 00 00 14 01 0f 00 ................ 
52 53 54 00 00 00 6c 00 06 9c b5 00 00 00 00 00 RST...l......... 
00 00 01 00 00 00 01 cd 05 00 00 00 65 63 68 6f ............echo 
00 cd cd cd 00 00 00 00       ........ 
TAO (276|1592) - Transport[828]::drain_queue_helper, end of data 
TAO (276|1592) - Transport[828]::cleanup_queue, byte_count = 72 
TAO (276|1592) - Transport[828]::cleanup_queue, after transfer, bc = 0, all_sent 
= 1, ml = 0 
TAO (276|1592) - Transport[828]::drain_queue_helper, byte_count = 72, head_is_em 
pty = 1 
TAO (276|1592) - Transport[828]::drain_queue_i, helper retval = 1 
TAO (276|1592) - Transport[828]::make_idle 
TAO (276|1592) - Cache_IntId_T::recycle_state, ENTRY_BUSY->ENTRY_IDLE_AND_PURGAB 
LE Transport[828] IntId=00A64ABC 
TAO (276|1592) - Leader_Follower[828]::wait_for_event, (follower), cond <00B10DD 
8> 

使用静态Client_Strategy_Factory " -ORBTransportMuxStrategy EXCLUSIVE "

2014-Sep-03 16:34:26.143024 
TAO (6664|5612) - Invocation_Adapter::invoke_i, making a TAO_CS_REMOTE_STRATEGY 
invocation 
TAO (6664|5612) - Transport_Cache_Manager_T::is_entry_available_i[824], true, st 
ate is ENTRY_IDLE_AND_PURGABLE 
TAO (6664|5612) - Cache_IntId_T::recycle_state, ENTRY_IDLE_AND_PURGABLE->ENTRY_B 
USY Transport[824] IntId=00854ABC 
TAO (6664|5612) - Transport_Cache_Manager_T::find_i, Found available Transport[8 
24] @hash:index {-1062667171:0} 
TAO (6664|5612) - Transport_Connector::connect, got an existing connected Transp 
ort[824] in role TAO_CLIENT_ROLE 
TAO (6664|5612) - Exclusive_TMS::request_id - <3> 
TAO (6664|5612) - GIOP_Message_Base::dump_msg, send GIOP message v1.2, 60 data b 
ytes, my endian, Type Request[3] 
GIOP message - HEXDUMP 72 bytes 
47 49 4f 50 01 02 01 00 3c 00 00 00 03 00 00 00 GIOP....<....... 
03 00 00 00 00 00 cd cd 1b 00 00 00 14 01 0f 00 ................ 
52 53 54 00 00 00 55 00 0d 7a 85 00 00 00 00 00 RST...U..z...... 
00 00 01 00 00 00 01 cd 05 00 00 00 65 63 68 6f ............echo 
00 cd cd cd 00 00 00 00       ........ 
TAO (6664|5612) - Transport[824]::drain_queue_helper, sending 1 buffers 
TAO (6664|5612) - Transport[824]::drain_queue_helper, buffer 0/1 has 72 bytes 
TAO - Transport[824]::drain_queue_helper (0/72) - HEXDUMP 72 bytes 
47 49 4f 50 01 02 01 00 3c 00 00 00 03 00 00 00 GIOP....<....... 
03 00 00 00 00 00 cd cd 1b 00 00 00 14 01 0f 00 ................ 
52 53 54 00 00 00 55 00 0d 7a 85 00 00 00 00 00 RST...U..z...... 
00 00 01 00 00 00 01 cd 05 00 00 00 65 63 68 6f ............echo 
00 cd cd cd 00 00 00 00       ........ 
TAO (6664|5612) - Transport[824]::drain_queue_helper, end of data 
TAO (6664|5612) - Transport[824]::cleanup_queue, byte_count = 72 
TAO (6664|5612) - Transport[824]::cleanup_queue, after transfer, bc = 0, all_sen 
t = 1, ml = 0 
TAO (6664|5612) - Transport[824]::drain_queue_helper, byte_count = 72, head_is_e 
mpty = 1 
TAO (6664|5612) - Transport[824]::drain_queue_i, helper retval = 1 
TAO (6664|5612) - Leader_Follower[824]::wait_for_event, (follower), cond <009009 
10> 

我理解这可能是螺纹和ORB模型的问题。我尝试了一些客户的策略：

静态Client_Strategy_Factory“-ORBTransportMuxStrategy EXCLUSIVE -ORBClientConnectionHandler RW”

这样可以减少问题发生的频率，但解决不了问题完整。

---

这和我6年前的经验很相似。在这种情况下，调用是在客户端的一个线程中发送的。在接收到响应之前，由于反应堆模式，该线程被重新用于发送另一个corba请求。这个案例似乎不同于这里的案例，因为它只是一个corba调用。我的线程堆栈的印象是有点像：

server.anotherInvocation() //the thread is used for another invocation. 
... 
server::echo() //send 1st corba invocation 
.... 
orb-run() 

Answer 1

的问题是，它是依赖于操作系统的，当网络堆栈将检测服务器断开连接，有时它永远不会发生。设置RELATIVE_RT_TIMEOUT_POLICY_TYPE策略以强制调用超时更好也更安全，请参阅ACE_wrappers/TAO/tests/Timeout以获取示例。