Re: [Condor-users] Jobs are executed only on the submitting machines

[Jaime Frey <jfrey@xxxxxxxxxxx>]

On Oct 2, 2006, at 3:10 AM, Dr. Raffaele Montella wrote:

> I'm new in Condor installing and configuring so I don't now if  
> there is some
> else to do to avoid this behaviour:
> I installed the Condor 6.8.1 on a cluster build by 9 P4@xxxxxx with  
> 1 master
> host and 9 working nodes (basically on a Beowulf system) with  
> Fedora Core 4
> Linux. All nodes share /home and /opt. I followed the install  
> procedure
> choosing a full installation on the master node configuring it as  
> the condor
> control manager. All daemons on the controller starts up correctly.
> In the CONTROL_CONFIG file I choose:
> LOCAL_DIR               = /home/condor/hosts/$(HOSTNAME)
> LOCAL_CONFIG_FILE       = $(RELEASE_DIR)/etc/$(HOSTNAME).local
> REQUIRE_LOCAL_CONFIG_FILE = FALSE
> HOSTALLOW_WRITE = *
>
> Then I configured each working node defining the
> CONDOR_HOME=/opt/condor-6.8.1, the
> CODOR_CONFIG=/opt/condor-6.8.1/etc/condor_config. Condor starts up  
> correctly
> on each working node. The condor_status command shows all machines  
> in the
> pool:
>
> Name          OpSys       Arch   State      Activity   LoadAv Mem
> ActvtyTime
>
> vm1@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000  1012
> 0+20:51:56
> vm2@xxxxxxxxx LINUX       INTEL  Owner      Idle       0.070  1012
> 0+00:10:05
> vm1@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+03:05:04
> vm2@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+20:50:50
> vm1@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+03:05:04
> vm2@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+20:50:53
> vm1@xxxxxxxxx LINUX       INTEL  Owner      Idle       1.000   250
> 0+22:20:50
> vm2@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+02:50:08
> vm1@xxxxxxxxx LINUX       INTEL  Owner      Idle       1.000   250
> 0+22:20:42
> vm2@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+02:55:05
> vm1@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+03:05:06
> vm2@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+20:50:54
> vm1@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   504
> 0+03:05:04
> vm2@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   504
> 0+20:50:55
> vm1@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+03:05:05
> vm2@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+20:50:53
> vm1@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.010   250
> 0+03:05:05
> vm2@xxxxxxxxx LINUX       INTEL  Unclaimed  Idle       0.000   250
> 0+20:50:54
>
>                      Total Owner Claimed Unclaimed Matched Preempting
> Backfill
>
>          INTEL/LINUX    18     3       0        15       0          0
> 0
>
>                Total    18     3       0        15       0          0
> 0
>
>
> It's appear working correctly, but if I submit a using the  
> following script
> with the command condor_submit -a "log = out.log" -a "error =  
> error.log"
> ex02.submit:
>
> Executable     = /bin/hostname
> Universe       = vanilla
> Requirements   = OpSys == "LINUX" && Arch =="INTEL"
>              Error   = err.$(Process)
>              Output  = out.$(Process)
>              Log = foo.log
>
> Queue 50
>
> The jobs are queued but executed only on the submitting machine.
> I tried with more jobs, for example 500, with all machines  
> unclaimed, but
> nothing! If I submit from the master node all jobs are executed on the
> master node, if I submit from the node01 all jobs are executed on  
> the node01
> and so on.
>
> What is wrong?

This is probably a shared filesystem problem. On unix, if you don't  
say otherwise, Condor assumes your job's data files are on a shared  
filesystem. If your machines don't have a share filesystem, then  
Condor will only run the job on the submit machine.

You can tell Condor to not rely on a shared filesystem and transfer  
the job's file itself by including the follow in your submit file:
should_transfer_files = YES
when_to_transfer_output = ON_EXIT
Then you can also use transfer_input_files to say what input files  
need to be transferred in addition to the executable.

+--------------------------------+-----------------------------------+
|           Jaime Frey           | I used to be a heavy gambler.     |
|       jfrey@xxxxxxxxxxx        | But now I just make mental bets.  |
| http://www.cs.wisc.edu/~jfrey/ | That's how I lost my mind.        |
+--------------------------------+-----------------------------------+