[GE users] SGE 6.1 and qsh failure

Reuti reuti at staff.uni-marburg.de
Wed Aug 8 13:22:15 BST 2007


Am 27.07.2007 um 20:16 schrieb Barry McInnes:

> I seem to have qlogin and qrsh working on the Macs, but not the way  
> its
> advertised in the ql.sh in the docs.
>
> It seems to be a problem with Macs or the Version of ssh/sshd.
> On Suns using Solaris 2.8 SGE 6.1 and OpenSSL 3.7.1p2 the ql.sh  
> works if
> I differentiate between qlogin and qrsh
> # qlogin
> if [ "$#" -le 2 ] ; then
> HOST=$1
> PORT=$2
> echo "Sun:params $PORT $HOST" >> /data/bmcinnes/macopenmpi.log
> /usr/local/bin/ssh -X -p $PORT $HOST
> else
> #qrsh
> PORT=$2
> HOST=$3
> echo "Sun:params $PORT $HOST" >> /data/bmcinnes/macopenmpi.log
> /usr/local/bin/ssh -X $*
> fi

The Howto doesn't mention a script for qrsh at all. You tried it on  
the SUN by specifying a script only for qlogin?

The -X I would suggest to put in /etc/ssh/ssh_config like:

Host *
     ForwardX11 yes
     ForwardX11Trusted yes


> On the Macs using 10.4.10 and SGE 6.1 and OpenSSL 4.5.p1

On Macs, sshd is not run as a daemon, but will be launched on  
incoming requests by launchd AFAIK. And in addition: maybe the TCP- 
wrapper is checking the used port then against the allowed/defined  
ports in /etc/services. Using ssh on a Mac I got for now qrsh  
<command> working by setting:

ssh             22,50000-65535/tcp
ssh             22,50000-65535/udp

And the first port found there will be used for outgoing connections  
as default - so 22 should stay there.

# ps aux -w -o pid,ppid,pgid,command
USER       PID %CPU %MEM      VSZ    RSS  TT  STAT STARTED      TIME   
PPID  PGID COMMAND
reuti     4496   0.1  0.1    28768   1240  ??  S<   12:52PM    
0:00.77     1  4496 /Users/reuti/sge/bin/darwin-x86/sge_execd
reuti     4458   0.0  0.4    43992   4052  ??  S    12:52PM    
0:01.48     1  4457 /Users/reuti/sge/bin/darwin-x86/sge_qmaster
reuti     4463   0.0  0.2    30840   1616  ??  S    12:52PM    
0:00.89     1  4463 /Users/reuti/sge/bin/darwin-x86/sge_schedd
reuti     4795   0.0  0.1    28444   1188  p3  S+    1:44PM    
0:00.01  4092  4795 qrsh sleep 120
reuti     4796   0.0  0.1    28524    620  ??  S<    1:44PM    
0:00.01  4496  4796 sge_shepherd-76 -bg
root      4797   0.0  0.1    87660   1220  ??  Ss    1:44PM    
0:00.13  4796  4797 /usr/sbin/sshd -i
reuti     4798   0.0  0.2    28304   1776  p3  S+    1:44PM    
0:00.07  4795  4795 /usr/bin/ssh -p 63057 node.any.where.com exec '/ 
Users/reuti/sge/utilbin/darwin-x86/qrsh_starter'
reuti     4800   0.0  0.1    87600    596  ??  S     1:44PM    
0:00.00  4797  4797 /usr/sbin/sshd -i
reuti     4801   0.0  0.0    84680    372  ??  Ss    1:44PM    
0:00.02  4800  4801 /Users/reuti/sge/utilbin/darwin-x86/qrsh_starter / 
Users/reuti/sge/default/spool/node/active_jobs/76.1
reuti     4802   0.0  0.0    84592    372  ??  S     1:44PM    
0:00.02  4801  4802 sleep 120

This implies of course:

qrsh /usr/X11R6/bin/xterm

is working and it will quit after typing exit.

-- Reuti


> qrsh will not startup it gives a library error, unless I use port  
> 22 via
> a tunnel. If I use $PORT I get library incompatibilty error, if I use
> port 22 no error and a connection, which is why I setup the tunnel to
> forward the SGE port number.
>
> #qlogin
> if [ "$#" -le 2 ] ; then
> HOST=$1
> PORT=$2
> PORT=22
> echo "Mac:params $PORT $HOST" >> /data/bmcinnes/macopenmpi.log
> /usr/bin/ssh -Y -p $PORT $HOST
> else
> # qrsh
> PORT=$2
> HOST=$3
> echo "Mac:params $PORT $HOST" >> /data/bmcinnes/macopenmpi.log
> shift 3
> /usr/bin/ssh -L $PORT:localhost:22 $HOST $*
> fi
>
>
> sshd -i is used on both platforms. The only other difference is Mac  
> use
> passwordless access for ssh, the Suns require a password.
>
>
> One problem left is the SGE job does not terminate, after the xterm is
> started and exited from eg.
>
> [mac27:/usr/local/sge] bmcinnes% qrsh xterm
> [mac27:/usr/local/sge] bmcinnes% qstat
> job-ID  prior   name       user         state submit/start at      
> queue
>                          slots ja-task-ID
> ---------------------------------------------------------------------- 
> -------------------------------------------
>   35085 0.55500 xterm      bmcinnes     Eqw   07/27/2007 12:11:09
>                              1
> [mac27:/usr/local/sge] bmcinnes% qdel -f 35085
> bmcinnes has deleted job 35085
> [mac27:/usr/local/sge] bmcinnes%
>
>
> I really want to get this to work, so we can start using openmpi ?
>
>
>
> On 7/24/07 11:11 PM, Reuti wrote:
>> Am 24.07.2007 um 18:39 schrieb Barry McInnes:
>>
>>> On 7/23/07 11:04 AM, Reuti wrote:
>>>
>>>> qrsh xterm
>>>>
>>>> will work with your port 22 setup, but will bypass the SGE  
>>>> supplied port
>>>> each job will use otherwise. I.e. the systemwide sshd will be  
>>>> used, not
>>>> the one setup for each job. Hence job control is limited. This is
>>>> completely different from the "qsh" command.
>>>>
>>>>
>>>
>>> What is the way to make sshd use ports from 49000-65535 that SGE  
>>> wants
>>> to use ? I can see how to add single ports to sshd_config, but not a
>>> range. I cant find any documentation within SGE/google.
>>
>> There isn't such an option in ssh. SGE will select a random port  
>> number
>> and start sshd and ssh with it as an option argument.
>>
>> -- Reuti
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
>> For additional commands, e-mail: users-help at gridengine.sunsource.net
>>
>
> -- 
> ---
> Barry McInnes
> 325 Broadway
> Boulder CO 80304
> (303)4976231
> barry.j.mcinnes at noaa.gov
> ---
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list