R

The VRPC R agent provides remote access to the entire R language functionality. Using a VRPC client, existing R code can be called by any other (VRPC-supported) programming language and from any remote location - e.g. directly from the browser within a React web application.

Technology

The VRPC R package implements a fully fledged MQTT client in pure dependency-free C++ (thanks to the great work of Takatoshi Kondo). Based on VRPC's asynchronous RPC protocol, the MQTT client calls R functions in an asynchronous and isolated fashion.

Teaser

Agent code: hello-world.R

library(vrpc)

greet <- function(whom) {
  paste("Hello,", whom)
}

start_vrpc_agent(domain = "public.vrpc")

Run in terminal:

Rscript hello-world.R # displays generated agent name

Client code: hello-world.js

const { VrpcRemote } = require('vrpc')

;(async () => {
  const client = new VrpcRemote({ domain: 'public.vrpc', agent: process.argv[2] })
  await client.connect()
  console.log(await client.callStatic({
    className: 'Session',
    functionName: 'greet',
    args: ['world!']
  }))
})()

Run in another terminal:

node index.js <agentName> # will print: "Hello, world!"

DISCLAIMER

At this very early time of this brand-new repository the VRPC package is not (yet) available via CRAN. You may still run the example utilizing the all-inclusive docker image heisenware/vrpc-r (see shipped Dockerfile) or by building the vrpc package your own (R CMD INSTALL --preclean vrpc).

Features

• lightning fast using plain C++-based MQTT with zero protocol overhead

• inherently parallel execution of R code

• support for stateless calls (mapped to VRPC static functions)

• support for stateful calls (R sessions mapped to VRPC instances)

• forwarding of graphical R output (plots) as high-quality SVGs

• forwarding of R errors as regular exceptions

• auto discovery of all available R functionality

• uncomplicated, client-only network architecture with constant management overhead even for large applications

Differences to OpenCPU

The OpenCPU project is another very nice solution that provides generic remote access to existing R code.

Depending on your use case, this might be the better choice for you. In order to understand the differences, here is a small and very high-level comparison:

VRPC	OpenCPU
transport protocol is MQTT	transport protocol is HTTP
no API but language specific clients (SDK-like)	RESTful API
establishes a single permanent connection as TCP client	establishes a connection per request as TPC server
provides access to code within the same file and packages	provides access to code within packages
allows stateful interaction in form of Session class instances (OOP style)	allows stateful interaction using session keys (functional state style)

Differences to Shiny

Shiny allows building interactive web apps straight from R and solely using R. In effect, reactive web-apps are realized using server-side rendering and server-side business logic for wiring input element updates (such as button clicks or slider moves) to output element updates (such as re-rendered tables or plots). Hence, the client (= browser) receives fully pre-rendered HTML pages for display.

VRPC in contrast, does not deal with HTML nor CSS. It focuses on transporting data in the most performing and simple way. Visualization and reactivity can still be implemented but those concepts are separated and left to technology that is much better suited to do so than R would ever be.

See for example VRPC Live which allows all the reactivity-wiring and visualization using client-side code based on React. It calls the R functions (and if needed any other technology's functions) to retrieve the relevant data when needed.

Or implement your own React application and use react-vrpc for backend connectivity and state management.