Created by: flyhighzy

使用环境：自建k8s, 使用paddle镜像：1.0.1-gpu-cuda8.0-cudnn7 启动pserver相关代码 ：

    if training_role == "PSERVER":
        place = fluid.CPUPlace()
    exe = fluid.Executor(place)
    if is_local:
        train_loop(fluid.default_main_program())
    else:
        eplist = []
        for ip in pserver_ips.split(","):
            eplist.append(':'.join([ip, port]))
        pserver_endpoints = ",".join(eplist)  # ip:port,ip:port...
        print(pserver_endpoints)
        print(trainers)
        print(trainer_id)
        t = fluid.DistributeTranspiler()
        t.transpile(
            trainer_id,
            pservers=pserver_endpoints,
            trainers=trainers)

        if training_role == "PSERVER":
            print("pserver")
            print("current_endpoint: %s" % current_endpoint)
            pserver_prog = t.get_pserver_program(current_endpoint)
            pserver_startup = t.get_startup_program(current_endpoint,
                                                    pserver_prog)
            print("start pserver...")
            exe.run(pserver_startup)
            print("pserver start up ok: %s" % pserver_startup)
            exe.run(pserver_prog)
            print("pserver ended")
        elif training_role == "TRAINER":
            print("trainer")
            train_loop(t.get_trainer_program())

输出日志中发现输出了start pserver...之后再无新日志输出，而trainer也一直在等待pserver启动成功, 不停的输出“pserver not ready, wait 3 sec to retry...”，一段时间后trainer也失败退出。 感觉pserver启动过程中在exe.run(pserver_startup)时卡住了，请问是什么原因可能导致这种情况？