Moab Workload Manager > Appendices > Appendix W: Moab Resource Manager Language Interface Overview > Moab Resource Manager Language Data Format

W.1 Moab Resource Manager Language Data Format

W.1.1 Query Resources Data Format

NAME FORMAT DEFAULT DESCRIPTION
ADISK <INTEGER> 0 Available local disk on node (in MB)
AFS <fs id="X" size="X" io="Y" rcount="X" wcount="X" ocount="X"></fs>[...] 0 Available filesystem state
AMEMORY <INTEGER> 0 Available/free RAM on node (in MB)
APROC <INTEGER> 1 Available processors on node
ARCH <STRING> --- Compute architecture of node
ARES one or more comma delimited <NAME>:<VALUE> pairs (ie, MATLAB:6,COMPILER:100) --- Arbitrary consumable resources currently available on the node
ASWAP <INTEGER> 0 Available swap on node (in MB)
CCLASS one or more bracket enclosed <NAME>:<COUNT> pairs (ie, [batch:5][sge:3]) --- Run classes supported by node. Typically, one class is 'consumed' per task. Thus, an 8 processor node may have 8 instances of each class it supports present, ie [batch:8][interactive:8]
CDISK <INTEGER> 0 Configured local disk on node (in MB)
CFS <STRING> 0 Configured filesystem state
CMEMORY <INTEGER> 0 Configured RAM on node (in MB)
CONTAINERNODE <STRING> --- The physical machine that is hosting the virtual machine. Only valid on VMs.
CPROC <INTEGER> 1 Configured processors on node
CPULOAD <DOUBLE> 0.0 One minute BSD load average
CPUSPEED <INTEGER> --- The node's processor speed in MHz
CRES one or more comma delimited <NAME>:<VALUE> pairs (ie, MATLAB:6,COMPILER:100) --- Arbitrary consumable resources supported and tracked on the node, ie software licenses or tape drives
CSWAP <INTEGER> 0 Configured swap on node (in MB)
FEATURE one or more colon delimited <STRING>'s (ie, WIDE:HSM) --- Generic attributes, often describing hardware or software features, associated with the node
GEVENT GEVENT[<EVENTNAME>]=<STRING> --- Generic event occurrence and context data
GMETRIC GMETRIC[<METRICNAME>]=<DOUBLE> --- Current value of generic metric, i.e., 'GMETRIC[temp]=103.5'.
IDLETIME <INTEGER> --- Number of seconds since last detected keyboard or mouse activity (often used with desktop harvesting)
MAXTASK <INTEGER> <CPROC> Maximum number of tasks allowed on the node at any given time
NETADDR <STRING> --- The IP address of the machine
NODEINDEX <INTEGER> --- The node's index
OS <STRING> --- Operating system running on node
OSLIST One or more comma delimited <STRING>'s with quotes if the string has spaces (ie. "SAS7 AS3 Core Baseline Build v0.1.0","RedHat AS3-U5Development Build v0.2"). --- Operating systems accepted by node
OTHER <ATTR>=<VALUE>[,<ATTR>=<VALUE>]... --- Opaque node attributes assigned to node
PARTITION <STRING> DEFAULT Partition to which node belongs
POWER <BOOLEAN>   Whether the machine is on or off
PRIORITY <INTEGER> --- Node allocation priority
RACK <INTEGER> 0 Rack location of the node
SLOT <INTEGER> 0 Slot location of the node
STATE* one of the following: Idle, Running, Busy, Unknown, Drained, Draining, or Down Down State of the node
UPDATETIME* <EPOCHTIME> 0 Time node information was last updated
VARATTR <ATTR1>=<VAL1>[=<displayName1>][+<ATTR2>=<VAL2>[=<displayName2>]]... ---

Plus-delimited (+) list of <ATTR>=<VAL>[=<displayName>] pairs that jobs can request. You can replace any of the equals signs with colons if desired.

Specifying a display name allows you to choose a name that will be displayed in the Mongo database instead of the unique ID (the <VALUE>).

If you give two different attributes the same value and one of them also has a display name specified, both attributes will appear with the same display name.

VARIABLE <ATTR>=<VAL> --- Generic variables to be associated with node
VMOSLIST <STRING> --- Comma-delimited list (,) of supported virtual machine operating systems for this node
XRES one or more comma delimited <NAME>:<VALUE> pairs (ie, MATLAB:6,COMPILER:100) --- Amount of external usage of a particular generic resource

* indicates required field

Node states have the following definitions:

State Description
Busy Node is running some jobs and will not accept additional jobs
Down Resource Manager problems have been detected. Node is incapable of running jobs.
Draining Node is responding but will not accept new jobs
Idle Node is ready to run jobs but currently is not running any.
Running Node is running some jobs and will accept additional jobs
Unknown Node is capable of running jobs but the scheduler will need to determine if the node state is actually Idle, Running, or Busy.

W.1.2 Query Workload Data Format

NAME FORMAT DEFAULT DESCRIPTION
ACCOUNT <STRING> --- AccountID associated with job
ARGS <STRING> --- job command-line arguments
COMMENT <STRING> 0 job resource manager extension arguments including qos, dependencies, reservation constraints, etc
COMPLETETIME* <EPOCHTIME> 0 time job completed execution
DDISK <INTEGER> 0 quantity of local disk space (in MB) which must be dedicated to each task of the job
DGRES name:value[,name:value] --- Dedicated generic resources per task.
DPROCS <INTEGER> 1 number of processors dedicated per task
DSWAP <INTEGER> 0 quantity of virtual memory (swap, in MB) which must be dedicated to each task of the job
ENDDATE <EPOCHTIME> [ANY] time by which job must complete
ENV <STRING> --- job environment variables
ERROR <STRING> --- file to contain STDERR
EVENT <EVENT> --- event or exception experienced by job
EXEC <STRING> --- job executable command
EXITCODE <INTEGER> --- job exit code
FLAGS <STRING> --- job flags
GEOMETRY <STRING> --- String describing task geometry required by job
GNAME* <STRING> --- GroupID under which job will run
HOSTLIST comma or colon delimited list of hostnames -
suffix the hostlist with a carat (^) to mean superset; suffix with an asterisk (*) to mean subset; otherwise, the hostlist is interpreted as an exact set
[ANY]

list of required hosts on which job must run. (see TASKLIST)

A subset means the specified hostlist is used first to select hosts for the job. If the job requires more hosts than are in the hostlist, they will be obtained from elsewhere if possible. If the job does not require all of the jobs in the hostlist, it will use only the ones it needs.

A superset means the hostlist is the only source of hosts that should be considered for running the job. If the job can't find the necessary resources in the hosts in this list it should not run. No other hosts should be considered in allocating the job.

INPUT <STRING> --- file containing STDIN
IWD <STRING> --- job's initial working directory
NAME <STRING> --- User specified name of job
NODES <INTEGER> 1 Number of nodes required by job (See Node Definition for more info)
OUTPUT <STRING> --- file to contain STDOUT
PARTITIONMASK one or more colon delimited <STRING>s [ANY] list of partitions in which job can run
PREF colon delimited list of <STRING>s --- List of preferred node features or variables. (See PREF for more information.)
PRIORITY <INTEGER> --- system priority (absolute or relative - use '+' and '-' to specify relative)
QOS <INTEGER> 0 quality of service requested
QUEUETIME* <EPOCHTIME> 0 time job was submitted to resource manager
RARCH <STRING> --- architecture required by job
RCLASS list of bracket enclosed <STRING>:<INTEGER> pairs --- list of <CLASSNAME>:<COUNT> pairs indicating type and number of class instances required per task. (ie, [batch:1] or [batch:2][tape:1])
RDISK <INTEGER> 0 local disk space (in MB) required to be configured on nodes allocated to the job
RDISKCMP one of >=, >, ==, <, or <= >= local disk comparison (ie, node must have > 2048 MB local disk)
REJCODE <INTEGER> 0 reason job was rejected
REJCOUNT <INTEGER> 0 number of times job was rejected
REJMESSAGE <STRING> --- text description of reason job was rejected
REQRSV <STRING> --- Name of reservation in which job must run
RESACCESS <STRING> --- List of reservations in which job can run
RFEATURES colon delimited list <STRING>'s --- List of features required on nodes
RMEM <INTEGER> 0 real memory (RAM, in MB) required to be configured on nodes allocated to the job
RMEMCMP one of '>=', '>', '==', '<', or '<=' >= real memory comparison (ie, node must have >= 512MB RAM)
ROPSYS <STRING> --- operating system required by job
RSOFTWARE <RESTYPE>[{+|:}<COUNT>]
[@<TIMEFRAME>]
--- software required by job
RSWAP <INTEGER> 0 virtual memory (swap, in MB) required to be configured on nodes allocated to the job
RSWAPCMP one of '>=', '>', '==', '<', or '<=' >= virtual memory comparison (ie, node must have ==4096 MB virtual memory)
SID <STRING> --- system id (global job system owner)
STARTDATE <EPOCHTIME> 0 earliest time job should be allowed to start
STARTTIME* <EPOCHTIME> 0 time job was started by the resource manager
STATE* one of Idle, Running, Hold, Suspended, Completed, or Removed Idle State of job
SUSPENDTIME <INTEGER> 0 Number of seconds job has been suspended
TASKLIST one or more comma-delimited <STRING>'s --- list of allocated tasks, or in other words, comma-delimited list of node ID's associated with each active task of job (i.e., cl01, cl02, cl01, cl02, cl03) The tasklist is initially selected by the scheduler at the time the StartJob command is issued. The resource manager is then responsible for starting the job on these nodes and maintaining this task distribution information throughout the life of the job. (see HOSTLIST)
TASKS* <INTEGER> 1 Number of tasks required by job (See Task Definition for more info)
TASKPERNODE <INTEGER> 0 exact number of tasks required per node
UNAME* <STRING> --- UserID under which job will run
UPDATETIME* <EPOCHTIME> 0 Time job was last updated
WCLIMIT* [[HH:]MM:]SS 864000 walltime required by job

* indicates required field

Job states have the following definitions:

State Definition
Completed Job has completed
Hold Job is in the queue but is not allowed to run
Idle Job is ready to run
Removed Job has been canceled or otherwise terminated externally
Running Job is currently executing
Suspended job has started but execution has temporarily been suspended

Completed and canceled jobs should be maintained by the resource manager for a brief time, perhaps 1 to 5 minutes, before being purged. This provides the scheduler time to obtain all final job state information for scheduler statistics.

Related Topics 

© 2015 Adaptive Computing