Path: utzoo!utgpu!news-server.csri.toronto.edu!clyde.concordia.ca!uunet!mcsun!ukc!mucs!cns!umaida!rn
From: rn@ap.co.umist.ac.uk (bob nutter)
Newsgroups: comp.sys.apollo
Subject: NCS problems (was-Re: create remote process)
Keywords: NCS,glbd
Message-ID: <1990Aug10.093611.11456@cns.umist.ac.uk>
Date: 10 Aug 90 09:36:11 GMT
References: <1174@fang.dsto.oz>
Sender: usenet@cns.umist.ac.uk (Usenet news user)
Reply-To: rn@ap.co.umist.ac.uk (bob nutter)
Organization: UMIST Computation, Manchester, UK.
Lines: 49

Hi!

I've suffered the same problems with crp from time to time, but my worst
problem has been with NCS, which has caused the printer not found/can't
bind socket type errors. This might be different, then again...

I've been getting 'registry server unavailable' messages on our gateway
to a large campus backbone. Merging registries doesn't work, and does
rebooting (full or phase II) *sometimes* worked. I finally traced the
problem to being due to the llbd on the node thinking that the glb was
somewhere off our local network and on the ethernet backbone. Thus I was
getting totally unknown apollos registered (with various addresses, ip
and dds) as objects such as the glb, registry updates/queries, etc. As
we had *Domain* routing switched off we weren't expecting this. I
haven't found a way to stop this yet, so if anyone has any ideas...

However a temporary cure, which lasts a random amount of time (usually
in days) is to disconnect from the offending network, retstart the llbd
after removing it's database and then reconnecting the network. 

eg: kill <llbd_pid> 
    rm /tmp/llbdbase.dat
    /etc/ifconfig eth0 down #or whatever interface
    /etc/server -p /etc/ncs/llbd &
    #then check using lb_admin that the correct glb is being used
    /etc/ifconfig eth0 up
    #restart lpd, etc if needed


-on bootup roughly the same thing happens, except we put a sleep on the
ether ifconfig to give the llbd chance to get the right glb. BTW, we
have a domain network number unique to our campus (at least) as well now.

It took a long time to realise this, so I hope it's of some help to
anyone suffering the same/similar problem. Glancing at the sr10.2
release notes, I notice that you can restrict glb to certain address
families to limit a glb to a subset of hosts on an internet. Is this
'new feature' a bug fix, perchance? ;-) If anyone else has had this
problem (especially if they've found a permanent fix for it!) I'd be
glad to hear from them....

bob

--------------------------------------------------------------------
bob nutter, computer officer     |  I'm happy, I'm happy, I'm happy,
UMIST dept of computation        | -And I'll punch the man who says 
PO box 88 Manchester m60 1qd uk  |                           I'm not.
tel:+44 61 200 3386              |         
email:rn@ap.co.umist.ac.uk       |          -Ivor Cutler