[GE users] How to move the common directory ?

templedf dan.templeton at sun.com
Wed Mar 17 14:29:51 GMT 2010


On 03/17/10 00:54, pmuller wrote:
> Hello,
>
> This is my first post, and I'm setting up my first Grid Engine
> platform. I would like to run a shadow master to ensure the grid
> availability.
>
>  From what I understand, I have to share the following directories :
> - The Master Spool directory : It must be available for writing from
> both master instances.
> - $SGE_ROOT/$SGE_CELL/common : It must be available for writing from
> both masters, and for reading from all nodes.
> Is it true ?

Yes.

>
> I currently specify the Master Spool directory with the
> QMASTER_SPOOL_DIR variable in my install template file.
> However, I can't find a way to select a specific common directory.

There isn't one.  It's always $SGE_ROOT/$SGE_CELL/common

>
> Since I have to share the common directory between all my hosts, I'm
> would like to move it outside of $SGE_ROOT. Something like
> /mnt/sgecommons would be great.
>
> I figured out several solutions :
> - Install the qmaster, move the directory content elsewhere on my
> filesystem and add a symlink. =>  This way, I can't easily automate my
> grid deployment.

What about installing the master, moving the whole cell directory off to 
somewhere on your remote file system, and then mounting it under 
$SGE_ROOT as default (or whatever the cell name is)?

> - Hack the $SGE_ROOT/$SGE_CELL/common/sgemaster script to change the
> common directory location - could it work ? (I don't know SGE
> internals...)

Probably not something you want to take on.  I believe there's an open 
RFE to implement this behavior, though.

>
> What's the cleaner way to share the common directory between hosts ?
>
> FYI :
> - We don't use NFS for data sharing - just GlusterFS.
> - We plan to use Classic spooling. Berkeley DB server looks
> interesting but seems to be a single point of failure and could open
> security holes.

And here's a really interesting loophole I just found.  I just installed 
a cluster with a cell name of "../../../var/tmp/sgecell", and it worked! 
  (My $SGE_ROOT is /usr/local/sge.)  The cell directory is in 
/var/tmp/sgecell, and the cluster seems to be perfectly happy with that. 
  Obviously I haven't done much testing of it, but it does appear to be 
something you should consider exploring further.

Daniel

> Thanks for your help.
>
> Philippe Muller
>
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=249136
>
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=249174

To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].



More information about the gridengine-users mailing list