Re: [Gnumed-devel] gnumed ideas 0.1 and post-0.1 (was Time for a major re-think in 2005)

[Ian Haywood]

On Fri, Jan 28, 2005 at 10:53:07PM +1100, catmat wrote:
> Ian Haywood wrote:
> 
> >
> >>#!/usr/local/bin/python2.4
> >>debug = True import sys, cgi
> >>import cgitb
> >>cgitb.enable()
> >
> >Don't use CGI, this won't work well with our middleware layer. (you 
> >would have to re-instantiate
> >clincial objects for every page load -- seriously slow)
> >
> >Use the python HTTP server classes instead to allow persistent objects.
> >
> >Ian
> 
> I was going to suggest have a unix socket server interface for 
> persistent objects for the cgi to call,
> or simple xmlrpc layer between the cgi and backend.
That is also an option.
> Just thought apache gives extra flexibility for later configuration, 
> than say simple http server.
> (I was going to say , more secure too, but that might be debatable ).
Yes, it is probably better at supporting SSL and such then Python
BaseHTTPServer,and may behave better under load.

This means you can write the actual web stuff in whatever you like, PHP,
JSP, mod_python, etc., so long as it can talk the RPC protocol we choose for the
middleware layer. XML-RPC is easy, but I always worry about its slowness
(as all data has to be XMLed then unXMLed.)
One problem is the middleware can't cope with more than one user and
more than one concurrently open patient.
This can be fixed by having the web layer  fork off
a new middleware server instance for each login.
The session cookie then needs to be able to identify the middleware server
instance.

unix sockets are good as they are a) fast and b) easier to secure
against mis-use. However AFAIK they can't be accessed from Java.


Ian

pgpagMirLYeVj.pgp
Description: PGP signature