PBS has certain sensitivity to hostname and hosts file relationships. Make sure the first entry in /etc/hosts which matches whatever "hostname" returns, is associated w/ the cluster interface. It may also need to be the first alias on that line (I believe there are several aliases).
Jeremy
At 02:51 AM 10/17/2003, Chiu, PCM (Peter) wrote:
Dear Paul,
I have a dilemma though.
The head node has a DNS entry on the first ethernet NIC, and a different name (master.gome.xxx.yyy) is assigned to the second interface. This second interface is connected is a different switch with which all other compute nodes share. It also uses a non-routing private subnet (192.168.0.x).
For PBS, I take it that the second interface name (master.gome.xxx.yyy) will be used to set up batch queues with the compute nodes. I don't suppose we can register this name/private address in DNS?
Still, I am much interested in knowing the way you get round this with BIND.
Thanks. Peter
-----Original Message----- From: Paul Gray [mailto:[EMAIL PROTECTED] Sent: 16 October 2003 18:02 To: Chiu, PCM (Peter) Cc: '[EMAIL PROTECTED]' Subject: Re: RE: [Oscar-users] Queries on Oscar
On Thu, 16 Oct 2003, Chiu, PCM (Peter) wrote:
> Still got /opt/pbs/bin/pbsnodes: Server has no node list > qmr obj=node2.gome.xxx.yyy.zzz svr=default; Unauthorzed Request > Creating pbs workq queue... qmgr obj=workq svr=default: Unauthorized > Request > > Attached is the log. > > Any further ideas?
I've seen the above errors several times before. The symptoms that seem to be present when it occurs is the lack of a valid dns entry for the entry in /var/spool/pbs/server_name, which is pbs_oscar by default (for example, type 'host pbs_oscar' and getting a host not found).
Often times (but not 100%), I've been able to correct this by changing the entry in /var/spool/pbs/server_name to match the dns entry for the head node. If the nodes can resolve the IP for the head node, you're in luck.
I've always (100% of the time) been able to correct this by writing my own DNS authority zones for the cluster, and resolving all of the oscar-specific hosts in /etc/hosts manually through bind (eg. pbs_oscar*). Afterwards, rerun the "complete cluster setup" and the test-cluster should pull through without a complete reinstall.
I'd provide some easy zone examples here if you're interested, but hopefully someone will interject here about why this isn't the "Right Thing to Do(tm)".
Anecdotal observations: Most of the time I've seen this issue when the cluster is being installed without an external network connection or when there is no DNS entry on the local LAN's dns for the IP of the head node.
I'd appreciate it if anyone could chime in here with steps that I could take so that this step isn't necessary.
ADVthanksANCE
(*) Keep in mind though, that underscores are not valid characters for hostnames (ref: RFC 952 and 1123), so that alternate DNS servers (pdnsd, for example, which is more light weight and secure than bind) will refuse to honor these hosts.
Paul Gray -o) 323 Wright Hall /\\ University of Northern Iowa _\_V Message void if penguin violated ... Don't mess with the penguin No one ever says "Hey, I can't read that ASCII e-mail ya sent me."
------------------------------------------------------- This SF.net email is sponsored by: SF.net Giveback Program. SourceForge.net hosts over 70,000 Open Source Projects. See the people who have HELPED US provide better services: Click here: http://sourceforge.net/supporters.php _______________________________________________ Oscar-users mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/oscar-users
-------------------------------------------------------
This SF.net email sponsored by: Enterprise Linux Forum Conference & Expo
The Event For Linux Datacenter Solutions & Strategies in The Enterprise Linux in the Boardroom; in the Front Office; & in the Server Room http://www.enterpriselinuxforum.com
_______________________________________________
Oscar-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/oscar-users
