thompson@PAN.SSEC.HONEYWELL.COM (John Thompson) (02/18/91)
OK -- I've had it. I'm trying to reset TCP/IP services on a gateway (ether/ring, dn10k) in our network. The node USED to run name-server lookup (it is a secondary name- server for the net too), and have the same name for both of its addresses. The name->address lookup gave the ethernet address, since that was closer to the outside world. I want the node to NOT use name-server lookup (although it'll still offer named services). It will now have 2 different names, so that there's a 1:1 and onto correspondence between name and address. Why? (I just had to answer). <FLAME ON> Cadence, in their finite wisdom, decided to write their own gethostname routine. Thus, name-servers tend to confuse/break their license/data manager. Not only did they do that, but they also broke the routine! It 'sometimes' fails in the case where there are multiple addresses with the same name. It also doesn't recognize the full domain name, so thinks that 'bregor' and 'bregor.ssec.honeywell.com' are 2 different nodes. Now that we're putting data-files on the dn10k, it needs to have the data mgr running (provides file locking for brain-dead Sun workstations, and they insist on it on the wonderful Apollos too). This means that I have to break TCP/IP so that it will work. <FLAME OFF> The problem is that the Apollos seem to keep some internal host information as long as they're up. Here's what I did: After making the necessary changes to /etc/rc.local, /etc/daemons/*, /etc/hosts, and /etc/named.data/*, I aborted all TCP services on the node. I then deleted the `node_data/systmp/tcp_data and my_inet_addr files. Then, I restarted TCP services (they start up fine). From any other node in the network, if I PING one of the 2 names, I get a response coming back from the correct address. However, if I'm on the node itself, and PING the ORIGINAL name, I get a response with the full domain name, AND from the original (now wrong) interface. SOMEWHERE in its kernel, the node is clinging tightly to the fact that 'bregor' is really bregor.ssec.honeywell.com and has an ethernet address (rather than a ring one). When I had to do this to our other gateway, I ended up rebooting the node to get it cleared out. Is there an easier way? (Is there any way?) I really don't want to reboot the node. -- jt -- John Thompson Honeywell, SSEC Plymouth, MN 55441 thompson@pan.ssec.honeywell.com Me? Represent Honeywell? You've GOT to be kidding!!!
ericb@srvr1 (Eric Bratton,235 Chry,9363561,) (02/19/91)
From article <9102181504.AA03886@pan.ssec.honeywell.com>, by thompson@PAN.SSEC.HONEYWELL.COM (John Thompson): > Why? (I just had to answer). <FLAME ON> > Cadence, in their finite wisdom, decided to write their own gethostname > routine. Thus, name-servers tend to confuse/break their license/data > manager. Not only did they do that, but they also broke the routine! > It 'sometimes' fails in the case where there are multiple addresses with > the same name. It also doesn't recognize the full domain name, so thinks > that 'bregor' and 'bregor.ssec.honeywell.com' are 2 different nodes. Now > that we're putting data-files on the dn10k, it needs to have the data mgr > running (provides file locking for brain-dead Sun workstations, and they > insist on it on the wonderful Apollos too). This means that I have to break > TCP/IP so that it will work. <FLAME OFF> We have the same problem. I hope Cadence fixes the problems with their software as soon as possible! Might I suggest another workaround? As far as I can tell (we don't run the data mgr), Dracula will still work as long as the /bin/hostname is set to the non-domain hostname without terrorizing name service for a network. Running "/bin/hostname bregor" in the /etc/rc file on the node should make the standalone programs run. It's a pretty ugly thing to do, tho. If you need a daemon to run with the contents of the /etc/hosts file, then try out the sequence: # sequence to run a program using the /etc/hosts file. /bin/hostname foo # cadence bug won't allow "foo.acme.com" /etc/nmconfig -h hostent_ascii # start daemon here /etc/server -p /idea/...datamgr... & # now start up everything else to use bind /etc/nmconfig -h hostent_bind # everything from here on out will use the name server. The real solution, of course, it to make sure Cadence fixes their software. Their node-lock scheme uses the node id, I don't see anything to be gained from fiddling with the hostnames. -- *-----------------------------------------------------------------* | Eric Bratton | The University of Michigan, Inc | | ericb@caen.engin.umich.edu | #include <std.disclaimers> | *-----------------------------------------------------------------*
randall@bcstec.boeing.com (Michael Randall) (02/19/91)
In article <9102181504.AA03886@pan.ssec.honeywell.com>, thompson@PAN.SSEC.HONEYWELL.COM (John Thompson) writes: > ...... However, if I'm on the node itself, and PING > the ORIGINAL name, I get a response with the full domain name, AND from the > original (now wrong) interface. SOMEWHERE in its kernel, the node is clinging > tightly to the fact that 'bregor' is really bregor.ssec.honeywell.com and has > an ethernet address (rather than a ring one). When I had to do this to our > other gateway, I ended up rebooting the node to get it cleared out. Is there > an easier way? (Is there any way?) I really don't want to reboot the node. > John Thompson > Honeywell, SSEC To clear the "memory" copy of hosts information use "/etc/route -f". This will "Flush" the hosts information. I'm not sure if it will solve all your problems. But it will get rid of the hosts info in memory. To check the host info in memory use "/bsd4.3/usr/bin/netstat -r". Just to see that the "route -f" works you may want to run this command before and after trying it. -- --------------------------------------------------------------------------- Michael W. Randall Work phone: (206)965-9557 randall@bcstec.boeing.com or mwr6182@skagway@bcsaic@uw-beaver ...!uunet!bcstec!randall