[GE users] stupid software queues for license, was: weird "orders queue version is not uptodate" messages in qmaster log
xavier.machenaud at st.com
Thu Mar 10 15:28:18 GMT 2005
[ The following text is in the "ISO-8859-1" character set. ]
[ Your display is set for the "ISO-8859-10" character set. ]
[ Some special characters may be displayed incorrectly. ]
There are few reasons I proceed this way :
* The code which is calling the licensed tool is embeded in in house
developped code which has been thru a validication cycle and I don't
want to change the code.
* the code is also part of a makefile and thus, submitted using qrsh.
As far as I known, qrsh doesn't rerun jobs doing an exit 99.
Magnus Söderberg wrote:
> Xavier MACHENAUD wrote:
>> I was able to progress a little bit on my problem.
>> Here is the context :
>> My goal is to avoid taking too many licenses. I first implemented the
>> solution described in the Flexlm HOWTO, but, because the complex is
>> not consumable, this solution result in oversubmitting jobs using the
>> license. And as my licenses are supported the flexlm queue feature,
>> I'm getting lots of licenses requests queued in flexlm, which is
>> penalizing non grid users using these licenses.
>> So I though about using a consumable resource, but I wasn't able to
>> make it work using the Flexlm HOWTO method. The only way I found was
>> to have a script deamon doing the following : * query flexlm to get
>> the number of free licenses + number of licenses used by grid's jobs.
>> * set the complex value by "doing qconf -mattr exechost
>> complex_values lic_XXX=<used by grid+free>
>> I've got one script deamon per flexlm vendor running in parallel (3
>> scripts monitoring 5 licenses).
>> This looked to work fine first but : * I'm seeing lots of "orders
>> queue version is not uptodate" messages in my qmater log * after a
>> while, the scheduler do not work anymore (no jobs get dispatched).
>> This is a really serious issue as my grid is not working properly.
>> Do you know if I'm doing something wrong or if it's a bug?
> Isn't this kind of a lot of trouble for nothing?
> Perhaps you could turn off queueing for licenses instead, at least for
> jobs through SGE?
> We have the software (in our case Modelsim) die if it doesn't get the
> license(s) it needs right away.
> We then detect it in the jobscript and exit(99), thereby telling SGE
> to rerun the job.
> We never have any problems with licenses.
> If your software doesn't support turning off license-queueing, tell
> your vendor to fix it.
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