(Click to open topic with navigation)
|Description:||Specifies job specific flags.|
The job can restart and should only utilize reserved resources.
|Description:||Specifies the list of partitions the object can access. If no partition list is specified, the object is granted default access to all partitions.|
The object can access resources located in the OldSP, Cluster1, and/or O3K partitions.
|Description:||Specifies the default QOS associated with the object.|
The object is assigned the default QOS premium.
|Description:||Specifies the list of QoSs the object can access. If no QOS list is specified, the object is granted access only to its default partition.|
The object can access any of the 3 QoSs listed.
By default, jobs may access QoSs based on the 'logical or' of the access lists associated with all job credentials. For example, a job associated with user "John," group "staff," and class "batch" may utilize QoSs accessible by any of the individual credentials. Thus the job's QOS access list, or QLIST, equals the 'or' of the user, group, and class QLIST's. (i.e., JOBQLIST = USERQLIST | GROUPQLIST | CLASSQLIST). If the ampersand symbol, '&', is associated with any list, this list is logically and'd with the other lists. If the carat symbol, '^', is associated with any object QLIST, this list is exclusively set, regardless of other object access lists using the following order of precedence user, group, account, QOS, and class. These special symbols affect the behavior of both QOS and partition access lists.
|Default:||Use available resources where ever found, whether inside a reservation or not.|
|Description:||Specifies the job may only utilize accessible, reserved resources. If <RESID> is specified, only resources in the specified reservation may be utilized.|
The job may only utilize resources located in the META.1 reservation.
|Description:||Specifies that the job array being submitted should not span across multiple partitions. This locks all sub jobs of the array to a single partition. If you want to lock all job arrays to a single partition, specify the ARRAYJOBPARLOCK parameter in moab.cfg to force this behavior on a global scale.|
> msub -t moab.[1-5]%3 -l walltime=30,flags=arrayjobparlock
|Description:||Specifies that the job array being submitted should span across multiple partitions. This is the default behavior in Moab, unless the ARRAYJOBPARLOCK parameter is specified in moab.cfg. This job flag overrides the ARRAYJOBPARLOCK parameter so that job arrays can be allowed to span multiple partitions at submit time.|
> msub -t moab.[1-5]%3 -l walltime=30,flags=arrayjobparspan
|Description:||Uses no compute resources such as processors, memory, and so forth; uses only generic resources.|
> msub -l gres=matlab,walltime=300
|Description:||Only applies to QOS. IGNIDLEJOBRSV allows jobs to start without a guaranteed walltime. Instead, it overlaps the idle reservations of real jobs and is preempted 2 minutes before the real job starts.|
|Default:||Jobs remain queued until they are able to run|
|Description:||Specifies that the job should be removed it is unable to allocate resources and start execution immediately.|
The job should be removed unless it can start running at submit time.
This functionality is identical to the resource manager extension QUEUEJOB:FALSE.
|Default:||Moab passes jobs to a resource manager to schedule.|
|Description:||Specifies that the job is an internal system job and will not be started via an RM.|
The job begins running in Moab without a corresponding RM job.
|Default||Moab can migrate the VM associated with the job.|
|Description||Specifies that Moab may not migrate the VM that the job sets up.|
msub -l walltime=INFINITY,template=VMTracking,os=linux,nodes=h3,jobflags=novmmigrate
Moab will not migrate the new VM.
|Default:||Jobs may not be preempted by other jobs|
|Description:||Specifies that the job may be preempted by other jobs which have the PREEMPTOR flag set.|
The job may be preempted by other jobs which have the PREEMPTOR flag set.
|Default:||Jobs may not preempt other jobs|
|Description:||Specifies that the job may preempt other jobs which have the PREEMPTEE flag set .|
The job may preempt other jobs which have the PREEMPTEE flag set.
|Default||Completed jobs are sent to a queue for a short period of time before Moab purges them from the system.|
|Description||Specifies that Moab should only purge the job from the completed queue if it completed successfully. If the job failed, Moab will keep it in the queue indefinitely to allow you to restart it at any time. This flag is particularly useful for setup and take down jobs in job workflows. See Creating workflows with job templates and Creating a Cloud Workflow for more information.|
If the job fails, Moab will not purge it from the completed job queue.
|Default:||Jobs may not be restarted if preempted.|
|Description:||Specifies jobs can be requeued and later restarted if preempted.|
The associated job can be preempted and restarted at a later date.
|Default:||Jobs may not be suspended if preempted.|
|Description:||Specifies jobs can be suspended and later resumed if preempted.|
The associated job can be suspended and resumed at a later date.
|Description:||Creates an internal system job that does not require resources.|
|Description:||When Moab is determining when and where a job can run, it either searches for the most resources or the longest range of resources. In almost all cases searching for the longest range is ideal and returns the soonest starttime. In some rare cases, however, a particular job may need to search for the most resources. In those cases this flag can be used to have the job find the soonest starttime. The flag can be specified at submit time, or you can use mjobctl -m to modify the job after it has been submitted. See the RSVSEARCHALGO parameter.|
> msub -l flags=widersvsearchalgo > mjobctl -m flags+=widersvsearchalgo job.1