Re: [Condor-users] New installation of Condor : questions about configuration files and other general things

[Jan Ploski <Jan.Ploski@xxxxxxxx>]

Thomas Pegeot wrote:
> Hello,
>
> I'm doing a work placement related to Condor, but as i'm a newbie in 
> Condor, i need your help. Actually, i have to put in place a desktop 
> grid composed of virtual machines running on dual-core computers 
> (Windows hosts running VirtualBox or VMWare Server virtual machines - 
> Debian Etch) and i want to use Condor to deal with these virtual machines.
>
> I set up Condor-6.9.2

Condor-6.9.2 is a development release, maybe you should stick with the 
stable 6.8.4?

> as a Central Manager (cm) on a linux server
> (Debian Etch) running a NFS Server, which allows me to share the release 
> directory (/usr/local/condor in my case) and a unique home 
> (/home/condor/hosts/$hostname/) with the virtual machines (node1,node2 
> ...).
>
> To install the central manager, i used the Older Unix Installation 
> Procedure (condor_install).
>
> Unfortunately, there is a lot of things i don't know how to deal with, 
> even with the Condor Manual.
>
> First, about files sharing. Do you think my setup is correct ?

It is ok. AFAIK, apart from the lock files (which reside in local /tmp 
by default), you can share everything over NFS.

> On each virtual machine, i mount cm:/home/condor/ on /home/condor (i 
> have a condor user on each vm) but is it correct ?

Yes.

> Same thing for the 
> release directory : is it good to share this same folder between the 
> central manager and the nodes ?

Yes.

> After the installation, i tried to run condor_status but i only got the 
> status of my central manager, but there was nothing about my nodes. :s

How long have you waited? (A few minutes should be enough.)

> http://pastebin.ca/448387
>
> When i run "condor_status -direct node1", "condor_status -direct node2" 
> ..., i get the following error : "condor_status:can't find address for 
> startd ...".
> So i checked which processes were running with ps -ef | egrep condor_ :
> - on my cm : 
> condor_master,condor_procd,condor_negociator,condor_collector,condor_startd,condor_schedd.
> - on my nodes : condor_master,condor_procd,startd,condor_sched
>
> So, i think that my problem must be related to my configuration files.

It could also be your network configuration (firewall, routing etc.)

> Here is my condor_config : http://pastebin.ca/448390
> My cm condor_config.local : http://pastebin.ca/448397

I don't see any errors at first glance.

> But the nodes condor_config.local are empty. I wonder whether it is 
> correct or not to have empty configurations files. :s

It is okay for the local configuration files to be empty. It simply 
means that you do not override any option from the global configuration 
file.

> Is another script to run on my virtual machines ?

No.

> I ran condor_init but it didn't fix my problem.

Next thing to do is inspect the logs of the cm and a chosen node:
$(TILDE)/hosts/$(HOSTNAME)/log/*

Shut down Condor (e.g. by killing condor_master on cm and the node), 
manually remove files from the log directory, then start the 
condor_master process again and look at or paste messages that appear in 
log files.

Regards,
Jan Ploski