/port_server

A super simple framework for extending Erlang supervision to external apps

Primary LanguagePython

===========
port_server
===========

port_server is a simple set of example code that illustrates how an Erlang OTP
application can be used to start, supervise and stop a set of related
external applications.

For background on this code, refer to doc/ceug-dec-2010.svg.

The motivation for the presentation to the Chicago Erlang User Group was to
provide a tangible benefit for using Erlang in non-Erlang environments.

The gist:

- Use Erlang ports to run external applications from a single Erlang node

- Use OTP application supervisors to start, supervise, and shutdown the
  applicable services

- Refactor external applications written in other languages to be smaller,
  more service oriented, and use a "fail fast and hard" pattern for error
  handling

Building the Code
=================

From the directory containing this README::

 $ make

Running the Example
===================

You need Python 2.x installed. The sample Python services assume the Python
interpreter is named /usr/bin/python2. If this is not the case for your system,
you can either create a symlink or (preferably) just modify the shebang in the
two service scripts in priv.

The example can be run from the directory containing this README file::

 $ ./run

This will start the "example" application, which starts two Python based
services: service_a and service_b.

To exit the Erlang process, type ^C ^C. This will shut down the Erlang process,
which will in turn shut down the two services.

From another console, test the services by running the following from the
same directory::

 $ ./client M N

where M and N are two positive integers. These are passed via XML-RPC to
service_b, which will "crunch" them into a single, utterly meaningless number.

To simulate a crash, provide a negative number for either M or N.

You can also manually kill either service_a or service_b:

 $ pgrep service_a
 $ pkill service_a
 $ pgrep service_a

You should see a different PID on the second call to pgrep.

Note the Erlang console log for SASL messages when either service crashes.

Refer to the presentation for background on why the example is designed the
way it is.

This pattern can be extended to other languages and communication protocols.

Benefits
========

This pattern implements a very simple, light weight service oriented
architecture (SOA) using just a few lines of Erlang. This is not just a
theoretical example -- this same code/pattern, which has been running in
production for over a year, is proven to work very well at providing robust,
fault tolerant applications in non-Erlang languages.

If an organization is considering Erlang for building fault tolerant
applications, this scheme is a practical starting point. It employs the central
Erlang princples of fault tolerance, but does so with existing application code
and requires only increment change to a target environment.