python线程卡死问题解决

1. top命令和日志方式判定卡死的位置

python代码忽然卡死，日志不输出，通过如下方式可以确定线程确实已经死掉了：

# top 命令

top命令可以看到机器上所有线程的执行情况，%CPU和%MEM可以看出线程消耗的资源情况

由于机器上线程数量太多，可能要查看的线程的信息在top命令当前屏幕上显示不出来可以通过如下方式查看

在top命令下输入:u

接下来会提示输入用户名，就可以查看该用户所执行的所有线程

Which user (blank for all): denglinjie

这样就可以看到degnlinjie用户的所有线程

可以看到那几个卡死线程的%CPU和%MEM都为0，说明线程根本没有消耗资源，那么可以看出线程已经卡死了

接下来通过打日志的方式来确定线程究竟是卡死在哪里了，线程卡死的地方大多数都是在io或者http请求那，所以以后遇到线程卡死的情况，就通过打日志的方式来确定卡死的位置，最终定位到问题确实是一个http服务挂掉了，而且此时requests.get()我虽然设置了超时，但是竟然无效

 

2 . 服务进程数量不足导致的客户端进程卡死

服务端代码：

handler = SimilarityService()
transport = TSocket.TServerSocket('10.134.113.75', 1234)
factory = TBinaryProtocol.TBinaryProtocolFactory()
processor = Processor(handler)
server = TProcessPoolServer.TProcessPoolServer(processor, transport)
server.setNumWorkers(10)
server.serve()
server.setNumWorkers(10)
server.serve()

客户端代码

docQue = queues.Queue(maxsize=1000)
pCount = 15


class ParseSaveEsProcess(multiprocessing.Process):

    def __init__(self, threadId):
        self.threadId = threadId
        multiprocessing.Process.__init__(self)

    def run(self):
        global docQue
        f = open('recall_match_file_all_simi.lst.%s' % self.threadId, 'w')
        try:
            transport = TSocket.TSocket('10.134.113.75', 1234)
            transport = TTransport.TBufferedTransport(transport)
            protocol = TBinaryProtocol.TBinaryProtocol(transport)
            client = Client(protocol)
            transport.open()

            while True:
                line = docQue.get(block=True)
                if not line:
                    print 'thread%d run over' % self.threadId    
                    break

                p = line.split('\t')
                if len(p) >= 6 and p[5] == 'simi_high':
                    simi_str = client.calculate_similarity_by_itemurl(p[0])
                    f.write(line + '\t' + simi_str + '\n')
                else:
                    f.write(line + '\n')    
            transport.close()
        except Thrift.TException as e:
            print str(e)
            pass    

class PutUrlProcess(multiprocessing.Process):

    def __init__(self):
        multiprocessing.Process.__init__(self)

    def run(self):
        global docQue
        for line in open('recall_match_file.lst', 'r'):
            baikeid = line.strip()
            docQue.put(baikeid, block=True)

        for i in range(pCount):
            docQue.put(None, block=True)


if __name__ == '__main__':
    putProcess = PutUrlProcess()
    putProcess.start()

    for i in range(pCount):
        parseProcess = ParseSaveEsProcess(i)
        parseProcess.start()transport = TSocket.TSocket('10.134.113.75', 1234)
            transport = TTransport.TBufferedTransport(transport)
            protocol = TBinaryProtocol.TBinaryProtocol(transport)
            client = Client(protocol)
            transport.open()

            while True:
                line = docQue.get(block=True)
                if not line:
                    print 'thread%d run over' % self.threadId    
                    break

                p = line.split('\t')
                if len(p) >= 6 and p[5] == 'simi_high':
                    simi_str = client.calculate_similarity_by_itemurl(p[0])
                    f.write(line + '\t' + simi_str + '\n')
                else:
                    f.write(line + '\n')    
            transport.close()
        except Thrift.TException as e:
            print str(e)
            pass    

class PutUrlProcess(multiprocessing.Process):

    def __init__(self):
        multiprocessing.Process.__init__(self)

    def run(self):
        global docQue
        for line in open('recall_match_file.lst', 'r'):
            baikeid = line.strip()
            docQue.put(baikeid, block=True)

        for i in range(pCount):
            docQue.put(None, block=True)


if __name__ == '__main__':
    putProcess = PutUrlProcess()
    putProcess.start()

    for i in range(pCount):
        parseProcess = ParseSaveEsProcess(i)
        parseProcess.start()

可以看到，进程ParseSaveEsProcess进程总共开启了15个，每个进程会打开一个thrift连接，打开后一直发送请求，并将处理的结果写文件，全部执行完成后才关闭thrift连接。

可是，发现从启动到执行了很长时间后，只有10个文件里面有内容写入，其中5个一直没有写入：

111965 recall_match_file_all_simi.lst.0
   111878 recall_match_file_all_simi.lst.1
        0 recall_match_file_all_simi.lst.10
        0 recall_match_file_all_simi.lst.11
        0 recall_match_file_all_simi.lst.12
        0 recall_match_file_all_simi.lst.13
        0 recall_match_file_all_simi.lst.14
   113429 recall_match_file_all_simi.lst.2
   110720 recall_match_file_all_simi.lst.3
   111993 recall_match_file_all_simi.lst.4
   113691 recall_match_file_all_simi.lst.5
   113360 recall_match_file_all_simi.lst.6
   113953 recall_match_file_all_simi.lst.7
   112007 recall_match_file_all_simi.lst.8
   113818 recall_match_file_all_simi.lst.9

原因是因为thrift服务端只启动了10个服务进程，所以只能同时处理10个请求，而我客户端打开的thrift连接一直没有关闭，所以10个服务进程被10个客户端连接霸占了，另外5个进程获取不到连接，自然就一直卡住了