使用0mq（ZeroMQ）同步两个简单python3脚本时出现死锁

当我尝试使用0mq（

ZeroMQ

）同步两个python3脚本时，我遇到了这种奇怪的死锁。这些脚本在数千次迭代中运行良好，但迟早它们都会停止并等待对方。我在Windows7上从不同的CMD窗口运行这两个脚本

我不明白为什么会出现这样的僵局
这里会出什么问题？

脚本A:

while (1):
   context = zmq.Context()
   socket = context.socket(zmq.REP)
   socket.bind('tcp://127.0.0.1:10001')
   msg = socket.recv()                        # Waiting for script B to send done
   # ............................................................................
   # ... do something useful (takes only a few millisecs)
   # ............................................................................     
   context = zmq.Context()
   socket = context.socket(zmq.REQ)
   socket.connect('tcp://127.0.0.1:10002')
   socket.send_string("done")                 # Tell script B we are done

脚本B

while (1):
   # ............................................................................
   # ... do something useful (takes only a few millisecs)
   # ............................................................................
   context = zmq.Context()
   socket = context.socket(zmq.REQ)
   socket.connect('tcp://127.0.0.1:10001')
   socket.send_string("done")               # Tell script A we are done

   context = zmq.Context()
   socket = context.socket(zmq.REP)
   socket.bind('tcp://127.0.0.1:10002')
   msg = socket.recv()                      # Waiting for script A to send done

---

您应该只处理一次

上下文

和

套接字

创建，而不是每次迭代

此外，您还应该重用上下文（除非您打算从另一个上下文使用它） 线程）

这同样适用于第二个脚本

try:
    context = zmq.Context()
    rq_sck = context.socket(zmq.REQ)
    rq_sck.connect('tcp://127.0.0.1:10001')

    rep_sck = context.socket(zmq.REP)
    rep_sck.bind('tcp://127.0.0.1:10002')

    while (1):
        do something useful (takes only a few millisecs)

        rq_sck.send_string("done") # Tell script A we are done

        msg = rep_sck.recv() # Waiting for script A to send done
finally:
    rq_sck.close()
    rep_sck.close()

编辑：更新代码以调用

Socket.close（）

自pyzmq版本14.3.0以来，

Socket.close（）

和

Context.term（）

不会自动调用 在垃圾收集期间，添加了正确关闭套接字的操作。

这不是死锁情况 当然，代码仍然需要一些注意

消除歧义：您的场景不会进入资源互锁状态，即死锁。是的，当然，您的代码会崩溃，但很可能不是由于

REQ/REP

死锁造成的（在有损网络

tcp:

传输类中可能会出现死锁）。发布的代码崩溃是由于非托管资源处理，而不是由于达到死锁/活锁的相互阻塞状态

---

如何修复它？ 首先，假设您的超低延迟激励系统不允许重复实例化任何内容。这方面也有例外，但让我们从中受益

将

.Context（）

资源设置（或从外部调用继承）移出循环

查看，您是否需要并且您的延迟限制允许您在每次循环运行中设置/拆除

.socket（）

资源两次

决定，一旦第一条消息在传输路径中丢失，您是否可以接受real

REQ/REP

死锁

强制执行优雅的资源使用终止（

.socket（）

-s，O/s

端口#

s，

.Context（）

-s）。不要让它们永远挂在没有终端的地方，而要创造无限多的其他系统，这会破坏任何“故障恢复”系统。资源永远不是无限的

设计以非阻塞方式发送信号和传输行为。这允许您检测和处理远程进程超时，并为本地补救/响应操作提供机会

重新设计代码，使其达到您所需的安全代码级别（以下示例在分布式处理框架中24/7/365软实时控制无止境循环中工作几年，使用远程键盘和其他一些本地和远程诊断工具）

---

生产等级代码缺少什么？ 您的代码必须“预见”分布式系统的任何部分中可能出现的错误。是的，这很难，但却是必要的。您的远程节点（一个通信对手）停止响应、丢失消息、重新启动、由于O/S崩溃而暂停，这一切都是可以想象得到的（再加上一些您只会在运行中发现的令人不快的惊喜…）。这是另一个潘多拉的盒子，以涵盖在这个小帖子，这并不意味着它是没有必要的。这是你的救生衣

尽可能以非阻塞方式进行设计，这样您就可以控制事件

无论如何，总是以优雅的方式发布系统资源和

.term（）

所有ZeroMQ

.Context（）

实例--“整理”是一种公平的做法--无论是在现实生活中还是在代码帝国中

# /\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
#NONSTOP RESPONDER RAW EXAMPLE:
def aMiniRESPONDER( aTarget2Bind2_URL             = "tcp://A.B.C.D:8889",
                    anExternalPREDICTOR           = None,
                    anExternallyManagedZmqCONTEXT = None,
                    aSpreadMinSafetyMUL           = 3.0,
                    aSilentMODE                   = True
                    ):
   try: # RESOURCES LAYER
        # ... SETUP
        # ------------------------------------------------- .Context()
        # can setup a locally-managed context or re-use
        # anExternallyManagedZmqCONTEXT obtained upon a func Call
        aZmqCONTEXT   = anExternallyManagedZmqCONTEXT or zmq.Context( 1 )   

        # localhost:8887 [REP] ... remote [REQ] peer  .connect() + .send()
        aCtrlPORT_URL = "tcp://*:8887"                                      

        # localhost:8890 [PUB] ... remote [SUB] peers .connect() +
        # .subscribe + .recv( zmq.NOBLOCK ) ( MQL4 cannot .poll() so far ...)
        aSIGsPORT_URL = "tcp://*:8890"                                      
        aXmitPORT_URL = aTarget2Bind2_URL

        aListOfSOCKETs = []

        pass # -------------------------------------------------------------# ZMQ
        try: # -------------------------------------------------------------#
            # try: XmitPORT
            aXmitSOCKET = aZmqCONTEXT.socket( zmq.PAIR )

            # XmitPORT
            aXmitSOCKET.bind(      aXmitPORT_URL )                          
            aListOfSOCKETs.append( aXmitSOCKET )
        except:                                                             
            #    EXC: XmitPORT on Failure: GRACEFUL CLEARING XmitPORT

            msg =  "\nEXC. ZmqError({0:s}) on aXmitSOCKET setup / .bind( {1:s} )"
            print msg.format( repr( zmq.ZMQError() ), aTarget2Bind2_URL )
            raise ValueError( "ZMQ_EXC_EXIT @ XmitPORT SETUP" )
        pass # -------------------------------------------------------------# ZMQ
        try: # -------------------------------------------------------------#
            # try: CtrlPORT    
            # CtrlSOCKET [REP] .recv()s<--[REQ] + .send()s--> [REQ]
            aCtrlSOCKET = aZmqCONTEXT.socket( zmq.REP )                     

            # CtrlPORT <-REQ/REP means a remote peer [REQ] has to
            # .send()+.recv() before sending another CtrlCMD
            aCtrlSOCKET.bind(      aCtrlPORT_URL )                          
            aListOfSOCKETs.append( aCtrlSOCKET )
        except:                                                             
            # EXC: CtrlPORT on Failure: GRACEFUL CLEARING both CtrlPORT
            # and XmitPORT
            msg =  "\nEXC. ZmqError({0:s}) on aCtrlSOCKET setup / .bind( {1:s} )"
            print msg.format( repr( zmq.ZMQError() ), aCtrlPORT_URL )
            raise ValueError( "ZMQ_EXC_EXIT @ CtrlPORT SETUP" )
        pass # -------------------------------------------------------------# ZMQ
        try: # -------------------------------------------------------------#
            # try: SIGsPORT

            # SIGsPORT [PUB] .send()s--> [SUB]s
            aSIGsSOCKET= aZmqCONTEXT.socket( zmq.PUB  )                     

            # SIGsPORT -->  PUB/SUB means a remote peer(s) [SUB] .subscribe() + .recv()
            aSIGsSOCKET.bind(      aSIGsPORT_URL )                          
            aListOfSOCKETs.append( aSIGsSOCKET )
        except:                                                             
            # EXC: SIGsPORT on Failure: GRACEFUL CLEARING both CtrlPORT
            # and XmitPORT and SIGsPORT
            msg =  "\nEXC. ZmqError({0:s}) on aSIGsSOCKET setup / .bind( {1:s} )"
            print msg.format( repr( zmq.ZMQError() ), aSIGsPORT_URL )
            raise ValueError( "ZMQ_EXC_EXIT @ SIGsPORT SETUP" )
        pass # -------------------------------------------------------------# ZMQ

        # vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv
        # ... SETUP YOUR APPLICATION CODE

        try:     # APP LAYER ___________________________________________
           #           what you want to do
           #           here you go ...

        except:  # APP LAYER ___________________________________________
           #           handle EXCs

        finally: # APP LAYER ___________________________________________
           #           your own application post-mortem / pre-exit code

        # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

   except:  # RESOURCES LAYER .............................................
        # ... code shall handle it's own exceptions + externally caused events

   finally: # RESOURCES LAYER .............................................
        # ... always, ALWAYS gracefully exit ( avoid leakages and dirty things )

        [ allSOCKETs.setsockopt( zmq.LINGER, 0 ) for allSOCKETs in aListOfSOCKETs ]
        [ allSOCKETs.close( )                    for allSOCKETs in aListOfSOCKETs ]

        # --------------------------------------------------------------#
        # RESOURCES dismantled, may .term()

        # .TERM(), NOP otherwise
        if not ( aZmqCONTEXT is anExternallyManagedZmqCONTEXT ):        #
                 aZmqCONTEXT.term()                                     #
        return

#/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
#不间断响应程序原始示例：
定义aMiniRESPONDER（aTarget2Bind2_URL=”tcp://A.B.C.D:8889",
anExternalPREDICTOR=无，
anExternallyManagedZmqCONTEXT=None，
aSpreadMinSafetyMUL=3.0，
aSilentMODE=True
):
try:#资源层
# ... 设置
#-------------------------------------------------------.Context（）
#可以设置本地管理的上下文或重复使用
#在func调用时获得的外部ymanagedzmqcontext
aZmqCONTEXT=anexternalymanagedzmqcontext或zmq.Context（1）
#本地主机：8887[代表]。。。远程[REQ]对等连接（）+发送（）
aCtrlPORT_URL=“tcp://*：8887”
#本地主机：8890[发布]。。。远程[SUB]对等点连接（）+
#.subscribe+.recv（zmq.NOBLOCK）（到目前为止MQL4无法.poll（））
aSIGsPORT_URL=“tcp://*：8890”
aXmitPORT_URL=aTarget2Bind2_URL
aListOfSOCKETs=[]
通过#---------------------------------------------#ZMQ
试试看-------------------------------------------------------------#
#尝试：XmitPORT
aXmitSOCKET=aZmqCONTEXT.socket（zmq.PAIR）
#XmitPORT
aXmitSOCKET.bind（aXmitPORT\u URL）
aListOfSOCKETs.append（aXmitSOCKET）
除：
#EXC:XmitPORT发生故障：正常清除XmitPORT
msg=“\nEXC.ZmqError（{0:s}）在aXmitSOCKET setup/.bind（{1:s}）”上
print msg.format（repr（zmq.ZMQError（）），aTarget2Bind2_URL）
提升值错误（“ZMQ\u EXC\u EXIT@XmitPORT SETUP”）
通过#-
# /\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\
#NONSTOP RESPONDER RAW EXAMPLE:
def aMiniRESPONDER( aTarget2Bind2_URL             = "tcp://A.B.C.D:8889",
                    anExternalPREDICTOR           = None,
                    anExternallyManagedZmqCONTEXT = None,
                    aSpreadMinSafetyMUL           = 3.0,
                    aSilentMODE                   = True
                    ):
   try: # RESOURCES LAYER
        # ... SETUP
        # ------------------------------------------------- .Context()
        # can setup a locally-managed context or re-use
        # anExternallyManagedZmqCONTEXT obtained upon a func Call
        aZmqCONTEXT   = anExternallyManagedZmqCONTEXT or zmq.Context( 1 )   

        # localhost:8887 [REP] ... remote [REQ] peer  .connect() + .send()
        aCtrlPORT_URL = "tcp://*:8887"                                      

        # localhost:8890 [PUB] ... remote [SUB] peers .connect() +
        # .subscribe + .recv( zmq.NOBLOCK ) ( MQL4 cannot .poll() so far ...)
        aSIGsPORT_URL = "tcp://*:8890"                                      
        aXmitPORT_URL = aTarget2Bind2_URL

        aListOfSOCKETs = []

        pass # -------------------------------------------------------------# ZMQ
        try: # -------------------------------------------------------------#
            # try: XmitPORT
            aXmitSOCKET = aZmqCONTEXT.socket( zmq.PAIR )

            # XmitPORT
            aXmitSOCKET.bind(      aXmitPORT_URL )                          
            aListOfSOCKETs.append( aXmitSOCKET )
        except:                                                             
            #    EXC: XmitPORT on Failure: GRACEFUL CLEARING XmitPORT

            msg =  "\nEXC. ZmqError({0:s}) on aXmitSOCKET setup / .bind( {1:s} )"
            print msg.format( repr( zmq.ZMQError() ), aTarget2Bind2_URL )
            raise ValueError( "ZMQ_EXC_EXIT @ XmitPORT SETUP" )
        pass # -------------------------------------------------------------# ZMQ
        try: # -------------------------------------------------------------#
            # try: CtrlPORT    
            # CtrlSOCKET [REP] .recv()s<--[REQ] + .send()s--> [REQ]
            aCtrlSOCKET = aZmqCONTEXT.socket( zmq.REP )                     

            # CtrlPORT <-REQ/REP means a remote peer [REQ] has to
            # .send()+.recv() before sending another CtrlCMD
            aCtrlSOCKET.bind(      aCtrlPORT_URL )                          
            aListOfSOCKETs.append( aCtrlSOCKET )
        except:                                                             
            # EXC: CtrlPORT on Failure: GRACEFUL CLEARING both CtrlPORT
            # and XmitPORT
            msg =  "\nEXC. ZmqError({0:s}) on aCtrlSOCKET setup / .bind( {1:s} )"
            print msg.format( repr( zmq.ZMQError() ), aCtrlPORT_URL )
            raise ValueError( "ZMQ_EXC_EXIT @ CtrlPORT SETUP" )
        pass # -------------------------------------------------------------# ZMQ
        try: # -------------------------------------------------------------#
            # try: SIGsPORT

            # SIGsPORT [PUB] .send()s--> [SUB]s
            aSIGsSOCKET= aZmqCONTEXT.socket( zmq.PUB  )                     

            # SIGsPORT -->  PUB/SUB means a remote peer(s) [SUB] .subscribe() + .recv()
            aSIGsSOCKET.bind(      aSIGsPORT_URL )                          
            aListOfSOCKETs.append( aSIGsSOCKET )
        except:                                                             
            # EXC: SIGsPORT on Failure: GRACEFUL CLEARING both CtrlPORT
            # and XmitPORT and SIGsPORT
            msg =  "\nEXC. ZmqError({0:s}) on aSIGsSOCKET setup / .bind( {1:s} )"
            print msg.format( repr( zmq.ZMQError() ), aSIGsPORT_URL )
            raise ValueError( "ZMQ_EXC_EXIT @ SIGsPORT SETUP" )
        pass # -------------------------------------------------------------# ZMQ

        # vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv
        # ... SETUP YOUR APPLICATION CODE

        try:     # APP LAYER ___________________________________________
           #           what you want to do
           #           here you go ...

        except:  # APP LAYER ___________________________________________
           #           handle EXCs

        finally: # APP LAYER ___________________________________________
           #           your own application post-mortem / pre-exit code

        # ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

   except:  # RESOURCES LAYER .............................................
        # ... code shall handle it's own exceptions + externally caused events

   finally: # RESOURCES LAYER .............................................
        # ... always, ALWAYS gracefully exit ( avoid leakages and dirty things )

        [ allSOCKETs.setsockopt( zmq.LINGER, 0 ) for allSOCKETs in aListOfSOCKETs ]
        [ allSOCKETs.close( )                    for allSOCKETs in aListOfSOCKETs ]

        # --------------------------------------------------------------#
        # RESOURCES dismantled, may .term()

        # .TERM(), NOP otherwise
        if not ( aZmqCONTEXT is anExternallyManagedZmqCONTEXT ):        #
                 aZmqCONTEXT.term()                                     #
        return