[gridengine users] qlogin, interactive sessions, and cgroups

Alex Chekholko chekh at stanford.edu
Fri Jun 5 21:52:15 UTC 2015


It should "just work" with something like:


qlogin_wrapper.sh:
#!/bin/sh

HOST=$1
PORT=$2

if [ "x$DISPLAY" = "x" ]; then
   exec /usr/bin/ssh -p $PORT $HOST
else
   exec /usr/bin/ssh -X -Y -p $PORT $HOST
fi



qconf -mconf : qlogin_command: /path/to/qlogin_wrapper.sh
                qlogin_daemon: /usr/sbin/sshd -i



FreeIPA shouldn't matter, as sge_shepherd will just run the command in 
qlogin_daemon and somehow pass the port?  That sshd will just use your 
regular sshd_config.

The h_vmem behavior is independent of qlogin behavior.  Do your h_vmem 
settings already work as you expect?

On 06/05/2015 01:50 PM, Chris Tobey wrote:
> Hi Everyone,
>
> I am looking to get some help in setting up my cluster to use
> interactive sessions.
>
> My cluster currently looks like this:
>
>    qmaster: “GridMaster” - CentOS 6.5 VM running Sons of Grid Engine 8.1.6.
>
>    execution hosts: ~40 CentOS 6.6 physical servers of varying
> capability. All execd installed from GridMaster ‘start_gui_installer’.
>
>    Users: Managed by FreeIPA 3.x authentication server. Each user has
> passwordless SSH between hosts via public keys.
>
> My goals:
>
> 1.A user will call something like ‘qlogin -l
> mf=20G,h_vmem=20G,h_rt=1:00:00’ and they will be presented with a
> terminal to a system with their requested resources / slots reserved.
>
> 2.If a user goes over their h_vmem or h_rt their shell will be
> terminated and any processes launched from it will be terminated /
> cleaned up.
>
> 3.Shell must be capable of X-forwarding.
>
> I think my goals are achievable through some configuration of SoGE using
> the qlogin wrapper script and the cgroups functions, but I am not sure how.
>
> I was able to set up the wrapper script, but could not use the port
> number. I fear it has to do with FreeIPA and their control over SSH.
>
> --
>
>    #!/bin/sh
>
>    HOST=$1
>
>    PORT=$2
>
>    /usr/bin/ssh -Y $HOST
>
> --
>
> I was able to get the qlogin command working:
>
> --
>
>    $ hostname
>
>    charmander.server.com
>
>    $ qlogin -m beas -N "grid-gui" -l mf=20G -l h_vmem=20G -l
> h_rt=1:00:00 -w e
>
>    Your job 140263 ("grid-gui") has been submitted
>
>    waiting for interactive job to be scheduled ...
>
>    Your interactive job 140263 has been successfully scheduled.
>
>    Establishing /import/stargate/grid/scripts/qlogin_ssh_wrapper.sh
> session to host bulbasaur.server.com ...
>
>    Warning: Permanently added 'bulbasaur.server.com' (RSA) to the list
> of known hosts.
>
>    Last login: Fri Jun  5 15:59:37 2015 from charmander.server.com
>
>    -bash-4.1$ hostname
>
>    bulbasaur.server.com
>
> --
>
> For some reason Grid Engine seems to think the job is aborted after 60
> seconds (maybe due to the port not being used?). I still have an open
> terminal, and can still run anything I want, but the Grid has released
> the mem_free and slots for others to use.
>
> If anyone has advice on how I can set this up to accomplish my three
> goals, it would be very much appreciated. I can post any configuration /
> logs / details required.
>
> Thanks,
>
> -Chris Tobey
>
-- 
Alex Chekholko chekh at stanford.edu




More information about the users mailing list