Wednesday, February 2, 2011

Advanced SGE Commands

Ranger Virtual Workshop
Advanced SGE Commands

At this point, you should have all of the basic information on tools that you need to submit and query jobs on Ranger. In this final section on SGE, we'll go through a few more commands that you may find useful as you go along. At the bottom of the page, you'll find included a table that also provides a conversion chart for SGE commands to other batch processors.
qhold/qrls

Once a job has been submitted, there are times when you may want to suspend or stop a job outright. Perhaps the job depends on data that hasn't finished uploading yet, whatever the reason, SGE provides you with the qhold tool for that.

qhold 28780

Specifically, when you issue a hold on a job you are setting a user hold on the job. There are also system and operator holds, but you'll only be setting user holds, which you implicitely set. To explicitely set a user hold, specify:

login3% qhold -h u 28780
job-ID prior name user state submit/start at queue slots ja-task-ID
-----------------------------------------------------------------------------------------------------------------
28780 0.00000 testJob username hqw 02/25/2008 07:56:57

We can see that the state has changed from the queued state (qw) to a held state (hqw). At this point, we have two options, either we are ready for the job to continue, or we want delete the job. For these two sitations, we use the qrls (release) and qdel commands.

login3% qrls -h u 28780
modified hold of job 28780
login3% qdel 28785
userName has deleted job 28785

There are situations where qdel won't work, and won't delete the job. For instance, if a node lit on fire while running your job, you won't be able to delete that job. qdel will send a signal to the daemon (ge_execd) to delete the job, but won't get a response. In this case, the only way to delete the job will be to add the -f (force) option to qdel. This tells the scheduler to not wait for confirmation from the node, you should only use this option when you have to!