Using R’s Socket Interface for Interprocess Communication

Often it is necessary to communicate runtime execution details from one process to another process or monitoring utility. The information can be used for logging, debugging, tracing or simply to relay the status or progress of a running program. In this post, I’ll demonstrate how to perform interprocess communication between two separate R processes using functions available in base R.

In order to setup a communication channel between separately running R instances, it is first necessary to initialize the socket server. A very simple example of a socket server that echoes the client’s messages is provided below. I’ll save it in a file named server.R:

# server.R: Example of a socket server in listening mode.
serverSocket = make.socket(host="localhost", port=6000, server=TRUE)

while(TRUE) {
    statusMsg = read.socket(serverSocket)
    print(statusMsg)
    if (nchar(statusMsg)==0) break
} 

close.socket(serverSocket)

We first call make.socket with server=TRUE, and port=6000. The port number can be any integer between 0-65535 (port numbers are unsigned 16-bit integers), but should always be set to something greater than 1024, since ports 0-1024 are reserved for privileged services. Next a looping construct is setup and the socket is read from until a message is received having character length 0. Once the zero-length character message is received, iteration ceases and serverSocket is closed by calling close.socket.

Next we implement the client socket and the logic to facilitate message passing between two R processes. In this example, a random poisson number is generated, the client sleeps for 1 second, then writes the current time along with the randomly generated Poisson number to the server listening on port 6000. We save the code that follows in a file named client.R:

# client.R: Example of socket client and message generating logic.
clientSocket = make.socket(host="localhost", port=6000, server=FALSE)

for (ii in 1:10) {
    rp = rpois(1,10)
    Sys.sleep(1)
    msg = paste0("[",Sys.time(), "] - Random Poisson: ", rp)
    write.socket(clientSocket, msg)
}

close.socket(clientSocket)

The call to make.socket is similar to that in server.R, except in client.R, server=FALSE (note that server=FALSE is the default for make.socket, but listing it explicitly helps to clearly indicate the purpose of each socket). All that remains is to kick them off. As mentioned earlier, it’s imperative to first run server.R. The socket server will enter listening mode, waiting on messages from client.R. If you attempt to run client.R without server.R already running, an exception will be generated, along the lines of:

Error in make.socket(host="localhost", port=6000, server=FALSE) : 

  socket not established

From RStudio, a separate R process can be established by selecting Session > New Session. First kickoff server.R. Once running, start client.R. The image below shows client.R on the left and server.R on the right:

The output written to the server.R console confirms that our socket setup successfully communicated information between separate invocations of R.