In shikokuchuo/nanonext: NNG (Nanomsg Next Gen) Lightweight Messaging Library

#| include: false
knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

nanonext

R binding for NNG (Nanomsg Next Gen), a successor to ZeroMQ. NNG is a socket library for reliable, high-performance messaging over in-process, IPC, TCP, WebSocket and secure TLS transports. Implements 'Scalability Protocols', a standard for common communications patterns including publish/subscribe, request/reply and service discovery.

As its own threaded concurrency framework, provides a toolkit for asynchronous programming and distributed computing. Intuitive 'aio' objects resolve automatically when asynchronous operations complete, and synchronisation primitives allow R to wait upon events signalled by concurrent threads.

Designed for performance and reliability, nanonext is a lightweight wrapper around the NNG C library, and is itself implemented almost entirely in C.

Provides the interface for code and processes to communicate with each other - receive data generated in Python, perform analysis in R, and send results to a C++ program – on the same computer or across networks spanning the globe.

Implemented scalability protocols:

Bus (mesh networks)
Pair (two-way radio)
Poly (one-to-one of many)
Push/Pull (one-way pipeline)
Publisher/Subscriber (topics & broadcast)
Request/Reply (RPC)
Surveyor/Respondent (voting & service discovery)

Supported transports:

inproc (intra-process)
IPC (inter-process)
TCP (IPv4 or IPv6)
WebSocket
TLS (over TCP and WebSocket)

Development of the TLS implementation was generously supported by the .

Web utilities:

ncurl - (async) http(s) client
stream - secure websockets client / generic low-level socket interface
ip_addr() - for retrieving all local network IP addresses by interface
messenger() - console-based instant messaging with authentication

Quick Start

nanonext offers 2 equivalent interfaces: a functional interface, and an object-oriented interface.

Functional Interface

The primary object in the functional interface is the Socket. Use socket() to create a socket and dial or listen at an address. The socket is then passed as the first argument of subsequent actions such as send() or recv().

Example using Request/Reply (REQ/REP) protocol with inproc transport:
(The inproc transport uses zero-copy where possible for a much faster solution than alternatives)

Create sockets:

#| label: example2
library(nanonext)

socket1 <- socket("req", listen = "inproc://nanonext")
socket2 <- socket("rep", dial = "inproc://nanonext")

Send message from 'socket1':

#| label: send2
send(socket1, "hello world!")

Receive message using 'socket2':

#| label: recv2
recv(socket2)

Object-oriented Interface

The primary object in the object-oriented interface is the nano object. Use nano() to create a nano object which encapsulates a Socket and Dialer/Listener. Methods such as $send() or $recv() can then be accessed directly from the object.

Example using Pipeline (Push/Pull) protocol with TCP/IP transport:

Create nano objects:

#| label: example
library(nanonext)

nano1 <- nano("push", listen = "tcp://127.0.0.1:5555")
nano2 <- nano("pull", dial = "tcp://127.0.0.1:5555")

Send message from 'nano1':

#| label: send
nano1$send("hello world!")

Receive message using 'nano2':

#| label: recv
nano2$recv()

Vignette

Please refer to the nanonext vignette for full package functionality.

This may be accessed within R by:

#| label: vignette
#| eval: false
vignette("nanonext", package = "nanonext")

Installation

Install the latest release from CRAN:

#| label: cran
#| eval: false
install.packages("nanonext")

Or the current development version from R-universe:

#| label: universe
#| eval: false
install.packages("nanonext", repos = "https://r-lib.r-universe.dev")

Building from Source

Linux / Mac / Solaris

Installation from source requires 'libnng' >= v1.9.0 and 'libmbedtls' >= 2.5.0 (suitable installations are automatically detected), or else 'cmake' to compile 'libnng' v1.11.0 and 'libmbedtls' v3.6.2 included within the package sources.

It is recommended for optimal performance and stability to let the package automatically compile bundled versions of 'libmbedtls' and 'libnng' during installation. To ensure the libraries are compiled from source even if system installations are present, set the NANONEXT_LIBS environment variable prior to installation e.g. by Sys.setenv(NANONEXT_LIBS = 1).

As system libraries, 'libnng' is available as libnng-dev (deb) or nng-devel (rpm), and 'libmbedtls' as libmbedtls-dev (deb) or libmbedtls-devel (rpm). The INCLUDE_DIR and LIB_DIR environment variables may be set prior to package installation to specify a custom location for 'libmbedtls' or 'libnng' other than the standard filesystem locations.

Additional requirements for Solaris: (i) the 'xz' package - available on OpenCSW, and (ii) a more recent version of 'cmake' than available on OpenCSW - refer to the 'cmake' website for the latest source file.

Windows

On Windows, 'libnng' v1.11.0 and 'libmbedtls' v3.6.2 will be compiled from the package sources during installation and hence requires the 'Rtools' toolchain.

For R >= 4.2 using the 'Rtools42' or newer toolchains, the prerequisite 'cmake' is included. For previous R versions using 'Rtools40' or earlier, it may be necessary to separately install a version of 'cmake' in Windows and ensure that it is added to your system's PATH.

Acknowledgements and Links

We would like to acknowledge in particular:

Garrett D'Amore, author of the NNG library, for generous advice and for implementing a feature request specifically for a more efficient 'aio' implementation in nanonext.
The R Consortium for funding the development of the secure TLS capabilities in the package, and Henrik Bengtsson and Will Landau's roles in making this possible.
Joe Cheng for prototyping the integration of nanonext with later to support the next generation of completely event-driven 'promises'.
Luke Tierney (R Core) and Mike Cheng for meticulous documentation of the R serialization mechanism, which led to the package's own implementation of a low-level interface to R serialization.
Travers Ching for a novel idea in extending the original custom serialization support in the package.
Jeroen Ooms - for his 'Anticonf (tm)' configure script, on which our original 'configure' was based, although much modified since.