gRPC：集合点终止于（StatusCode.INTERNAL，收到错误代码为2的RST_STREAM）

我正在用python实现gRPC客户端和服务器。服务器成功从客户端接收数据，但是客户端接收回“ RST_STREAM，错误代码为2 ”。

它实际上是什么意思，我该如何解决？

这是我的原始文件：

service MyApi {
    rpc SelectModelForDataset (Dataset) returns (SelectedModel) {
    }
}
message Dataset {
    // ...
}
message SelectedModel {
    // ...
}

我的服务实现如下所示：

class MyApiServicer(my_api_pb2_grpc.MyApiServicer):
def SelectModelForDataset(self, request, context):
    print("Processing started.")
    selectedModel = ModelSelectionModule.run(request, context)  
    print("Processing Completed.")
    return selectedModel

我用以下代码启动服务器：

import grpc
from concurrent import futures
#...
server = grpc.server(futures.ThreadPoolExecutor(max_workers=100))
my_api_pb2_grpc.add_MyApiServicer_to_server(MyApiServicer(), server)
server.add_insecure_port('[::]:50051')
server.start()

我的客户看起来像这样：

channel = grpc.insecure_channel(target='localhost:50051')
stub = my_api_pb2_grpc.MyApiStub(channel)
dataset = my_api_pb2.Dataset() 
# fill the object ...
model = stub.SelectModelForDataset(dataset)  # call server

客户端调用后，服务器开始处理直到完成（大约需要一分钟），但是客户端立即返回并显示以下错误：

Traceback (most recent call last):                                                                   
File "Client.py", line 32, in                                                                
    run()                                                                                            
File "Client.py", line 26, in run                                                                    
    model = stub.SelectModelForDataset(dataset)  # call server                                       
File "/usr/local/lib/python3.5/dist-packages/grpc/_channel.py", line 484, in __call__
    return _end_unary_response_blocking(state, call, False, deadline)                                
File "/usr/local/lib/python3.5/dist-packages/grpc/_channel.py", line 434, in _end_unary_response_blocking                                                                                               
    raise _Rendezvous(state, None, None, deadline)                                                 
grpc._channel._Rendezvous: <_Rendezvous of RPC that terminated with (StatusCode.INTERNAL, Received RST_STREAM with error code 2)>

如果我异步执行请求并等待将来，

model_future = stub.SelectModelForDataset.future(dataset)  # call server
model = model_future.result()

客户端等待直到完成，但是之后仍然返回错误：

Traceback (most recent call last):                                                                   
File "AsyncClient.py", line 35, in                                                           
    run()                                                                                            
File "AsyncClient.py", line 29, in run                                                               
    model = model_future.result()                                                                    
File "/usr/local/lib/python3.5/dist-packages/grpc/_channel.py", line 276, in result                  
    raise self                                                                                     
grpc._channel._Rendezvous: <_Rendezvous of RPC that terminated with (StatusCode.INTERNAL, Received RST_STREAM with error code 2)>

UPD：启用跟踪后，GRPC_TRACE=all我发现了以下内容：

客户，在请求后立即：

E0109 17:59:42.248727600    1981 channel_connectivity.cc:126] watch_completion_error: {"created":"@1515520782.248638500","description":"GOAWAY received","file":"src/core/ext/transport/chttp2/transport/chttp2_transport.cc","file_line":1137,"http2_error":0,"raw_bytes":"Server shutdown"}            
E0109 17:59:42.451048100    1979 channel_connectivity.cc:126] watch_completion_error: "Cancelled"  
E0109 17:59:42.451160000    1979 completion_queue.cc:659]    Operation failed: tag=0x7f6e5cd1caf8, error={"created":"@1515520782.451034300","description":"Timed out waiting for connection state change","file":"src/core/ext/filters/client_channel/channel_connectivity.cc","file_line":133}
...(last two messages keep repeating 5 times every second)

服务器：

E0109 17:59:42.248201000    1985 completion_queue.cc:659]    Operation failed: tag=0x7f3f74febee8, error={"created":"@1515520782.248170000","description":"Server Shutdown","file":"src/core/lib/surface/server.cc","file_line":1249}                                                                    
E0109 17:59:42.248541100    1975 tcp_server_posix.cc:231]    Failed accept4: Invalid argument                                                                             
E0109 17:59:47.362868700    1994 completion_queue.cc:659]    Operation failed: tag=0x7f3f74febee8, error={"created":"@1515520787.362853500","description":"Server Shutdown","file":"src/core/lib/surface/server.cc","file_line":1249}                                                                                                                                             
E0109 17:59:52.430612500    2000 completion_queue.cc:659]    Operation failed: tag=0x7f3f74febee8, error={"created":"@1515520792.430598800","description":"Server Shutdown","file":"src/core/lib/surface/server.cc","file_line":1249}
... (last message kept repeating every few seconds)                                                             

UPD2：

我Server.py文件的全部内容：

import ModelSelectionModule
import my_api_pb2_grpc
import my_api_pb2
import grpc
from concurrent import futures
import time

class MyApiServicer(my_api_pb2_grpc.MyApiServicer):
    def SelectModelForDataset(self, request, context):
        print("Processing started.")
        selectedModel = ModelSelectionModule.run(request, context)
        print("Processing Completed.")
        return selectedModel


# TODO(shalamov): what is the best way to run a python server?
def serve():
    server = grpc.server(futures.ThreadPoolExecutor(max_workers=100))
    my_api_pb2_grpc.add_MyApiServicer_to_server(MyApiServicer(), server)
    server.add_insecure_port('[::]:50051')
    server.start()

    print("gRPC server started\n")
    try:
        while True:
            time.sleep(24 * 60 * 60)  # run for 24h
    except KeyboardInterrupt:
        server.stop(0)


if __name__ == '__main__':
    serve()

UPD3：似乎ModelSelectionModule.run是造成问题的原因。我试图将其隔离到单独的线程中，但没有帮助。在selectedModel最终计算，但客户已经消失在那个时候。如何防止此呼叫与grpc混淆？

pool = ThreadPool(processes=1)
async_result = pool.apply_async(ModelSelectionModule.run(request, context))
selectedModel = async_result.get()

呼叫比较复杂，它产生并加入大量的线程，调用不同的库，例如scikit-learn和smac等。如果我将所有内容都张贴在这里，那将太多了。

调试时，我发现在客户端请求后，服务器保持2个连接打开（fd 3和fd 8）。如果我手动关闭fd 8或向其中写入一些字节，则在客户端中看到的错误将变为Stream removed（而不是Received RST_STREAM with error code 2）。似乎套接字（fd 8）被子进程破坏了。这怎么可能？如何保护子进程不访问套接字？