I'm getting ready to test my chemical application on a server.  It has
two Durus databases, one for the data (readonly) and one for sessions
(read/write).  I'll be using SCGI with multiple children, so I'll have
to have two Durus daemons running.  What a number of processes to
manage!  Especially if I start using this model for other
applications.

I'm running Linux, so I first tried making a separate init.d script
for each Durus server, with its own PID and Unix domain socket, using
the 'durus' command-line tool and start-stop-daemon.  It refuses to
start and gives no error message, but that could be my fault (file
permission error somewhere?).

But my path configuration is in Python modules, so then I tried
writing a program that called durus.run_durus.start_durus .  But I'd
still have to have start-stop-daemon manage three PID files.

Now I'm trying to do it all in my top-level Python program, doing my
own fork()'s for the Durus servers.  It seems like this would be a
common task, especially at the MEMS Exchange which has lots of Quixote
programs with Durus databases.  So I'm wondering if there's any
existing code for this.  Also, how do you manage logfiles?  I'd like
to get Quixote and the two Durus servers writing to the same logfile
if possible.  What about signals?  Should I just override SIGTERM to
stop the Durus servers on exiting?  Or do I have to override the dozen
other signals like SIGPIPE that might be raised?  Also, stop_durus
just sends a "Q" to the server and hopes it quits, rather than killing
it.  Has this been reliable in practice?

--
Mike Orr