Solve the grpc connection problem. Dial succeeds with transientfailure

If help , Welcome to leave your footprints ！

Details are as follows

code = Unavailable desc = connection error: desc = "transport: Error while dialing dial tcp 192.168.31.33:9001: 
connectex: No connection could be made because the target machine actively refused it."

At this time, the connection status is  TransientFailure

Catalog

background

Phenomenon summary

Analyze and solve

Summary and appendix

---

background

GRPC When establishing a connection between the client and the server, we all know that Dial, When reporting the error, in fact Dial The operation was successful , here err by nil, But the call rpc Interface failed , Reported the wrong .

Phenomenon summary

1,  Put both the client and the server on this computer , And when the address is written in the address of this machine, it will also report an error ,Dial success , call rpc Interface failed ;
Dial The connection status when successful is Connecting, call rpc The connection state before interface failure changes to TransientFailure;
    
   What do these two states mean ？ Look at the source ：

    const (
    // Idle indicates the ClientConn is idle. Idle  Express  ClientConn  At rest .
    Idle State = iota
    // Connecting indicates the ClientConn is connecting.  Connecting  Express  ClientConn  on connection .
    Connecting
    // Ready indicates the ClientConn is ready for work. Ready  Express  ClientConn  Ready to work .
    Ready
    // TransientFailure indicates the ClientConn has seen a failure but expects to recover.  TransientFailure  Express  ClientConn  I have seen the failure but hope to recover .
    TransientFailure
    // Shutdown indicates the ClientConn has started shutting down.  Shutdown  Express  ClientConn  Has begun to close .
    Shutdown
)

2, Put both the client and the server on this computer , And the address is 127.0.0.1 When it comes to success , call rpc The interface is also successful , Everything is all right ;

3, Put the client and server on different hosts , And the starting service address is written 127.0.0.1、 Write the client connection address to the server ip If you fail, you will make a mistake ,Dial success , call rpc Interface failed ;

It makes people wonder , What is the problem ？

Analyze and solve

When the above phenomena occur , The host of the server has opened the port number 9001 Of tcp port , Therefore, the problem of port opening is eliminated ;

in addition , The client and server can succeed when both are on the machine , But it doesn't succeed on different hosts , It shows that the probability is still the problem of network failure ;

At the same time, grab bags , It is found that there are packets flowing from the client to the server host and the ports are correct , But back and forth are R. and S The marks of are alternating , The server keeps replying to the client R. Indicates that the connection exception has been reset , At the same time, such a reply is actually consistent with the situation when the server does not start the service ;

According to the above phenomenon , It is basically determined that the client does not recognize the start of the server （ In fact, the server starts ）

Finally, I checked the code and learned , When starting the service, the listening address is  

net.Listen("tcp", fmt.Sprintf("%s:%s", "127.0.0.1", 9001))

The following expressions can be used ：

net.Listen("tcp", fmt.Sprintf(":%s", 9001))
net.Listen("tcp", fmt.Sprintf("%s:%s", " This machine IP", 9001))
net.Listen("tcp", fmt.Sprintf("%s:%s", "0.0.0.0", 9001))

Can succeed ！

First of all 、 The three service addresses printed after startup are [::]:9001
The second is 192.168.31.33:9001

Summary and appendix

Then sum up the reasons , It's actually 127.0.0.1 Only this machine can be connected successfully at startup , It only works on this machine , Because it is the local loopback address ,
Normal network packets are from 3 Layer into 2 layer , Then send it out , And the packet sent to the loopback address , Directly in IP It's short circuited , That is, to IP Layers are directly IP Layer receiving 、 Don't send it down any more ;
That is to say, you can directly use 127.0.0.1.

1、 If the listening address is 127.0.0.1：port, It means only listening to the data of the loopback interface , That is, the machine sends it to the machine port Port data , Data sent by other hosts cannot be received .
2、 If the listening address is 0.0.0.0:port, It means to monitor the data of all network interfaces , Including loopback interface and other network interfaces , All data sent by this machine and other hosts can be received .
3、 If the listening address is local ip:port, Means to listen to the local computer only ip Corresponding network interface . Only the destination address is local ip The data of .（ For the case that this machine is sent to this machine , If the destination host address is 127.0.0.1 be , Don't get , Only if the destination host address is local ip Can receive ）.

actually 0.0.0.0 Risk is bigger , It is recommended to use this machine IP Appropriate .