Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!linus!decvax!decwrl!glacier!reid
From: reid@glacier.ARPA (Brian Reid)
Newsgroups: net.news.adm,net.news.b
Subject: Re: great glacier news flood problem solved
Message-ID: <5933@glacier.ARPA>
Date: Mon, 31-Mar-86 22:44:01 EST
Article-I.D.: glacier.5933
Posted: Mon Mar 31 22:44:01 1986
Date-Received: Wed, 2-Apr-86 07:46:08 EST
References: <5819@glacier.ARPA> <1786@seismo.CSS.GOV>
Reply-To: reid@glacier.UUCP (Brian Reid)
Organization: Stanford University, Computer Systems Lab
Lines: 27
Xref: linus net.news.adm:557 net.news.b:1014

>On seismo, I put the shell script that cats to a file name in /bin, not
>/usr/bin. I then mad sure that uuxqt searched /bin first.
>This way I didn't have to screw with any of the source files.

The problem with this--or any scheme that involves the conditional presence
or absence of a particular executable file--is that if rdist should happen
to run while /bin/rnews exists, or while /usr/bin/rnews is the wrong
version, then the bogus setup is distributed to all of the client machines,
and disaster ensues. Maybe the right thing is never to run news on an rdist
master machine, but I use rdist to propagate the news software to client
machines, so that wouldn't work either.

I think that the only solution that will meet all of the criteria is to
modify the c code of rdist so that it does the copying to a spool directory
after checking for the presence/absence of some flag file. It's OK to mark
flag files so that they won't be rdisted.

I'm beginning to get the idea that most people out there don't try to run
big clusters of machines with centralized software support and rdist.

Brian



-- 
	Brian Reid	decwrl!glacier!reid
	Stanford	reid@SU-Glacier.ARPA