jobid

Man page for the LDMS jobid plugin

Date:: 03 Dec 2016
Manual section:: 7
Manual group:: LDMS sampler

SYNOPSIS

Within ldmsd_controller or in a configuration file

config name=jobid [ <attr>=<value> ]

DESCRIPTION

With LDMS (Lightweight Distributed Metric Service), plugins for the ldmsd (ldms daemon) are configured via ldmsd_controller or a configuration file. The jobid plugin provides jobid info from /var/run/ldms.jobinfo or similar files replaced periodically by resource managers. When files are missing, the value 0 or equivalent is reported.

CONFIGURATION ATTRIBUTE SYNTAX

config

name=<plugin_name> producer=<pname> instance=<set_name> [component_id=<compid> schema=<sname>] [with_jobid=<bool>] file=<filepath>

configuration line

name=<plugin_name>: This MUST be jobid.
producer=<pname>: The producer name value.
instance=<set_name>: The name of the metric set.
schema=<schema>: Optional schema name. It is intended that the same sampler on different nodes with different metrics have a different schema. If not specified, will default to `vmstat`.
component_id=<compid>: Optional component identifier. Defaults to zero.
with_jobid=<bool>: Option to lookup job_id with set or 0 if not. The job_id column will always appear, but populated witn zero.

BUGS

No known implementation bugs. Design features you may not like: Relies on site-specific resource manager configuration to produce the file read. Does not query local or remote manager daemons. May be slow to sample and generate undesirable filesystem events if filepath is on a networked filesystem instead of a node-local RAM partition as is usual in clusters.

NOTES

The colname option from LDMS v2 slurmjobid plugin is no longer supported. The sampler offset for the jobid plugin should be slightly less than all other plugins to ensure consistency in the job information reported for a given time interval across all other plugins. The time interval for the jobid plugin need only be approximately the clock granularity of the resource manager.

Other samplers use the jobid plugin as the jobid data source. If the jobid sampler is not loaded, these samplers will report 0 jobid values.

FILE FORMAT

The file consists of key-value pairs, one per line, separated by an equals-sign. The recognized keys are:

JOBID: An unsigned integer (up to 64-bit) identifying the job. The number zero is reserved to mean that no job is currently running.
UID: An unsigned integer (up to 64-bit) representing the User ID associated with the job.
APPID: An unsigned integer (up to 64-bit) representing the an application ID for the job.
USER: A string representing the username associated with the job.

Only the JOBID field is required. The other fields are optional, and will default to zero.

EXAMPLES

Within ldmsd_controller or in a configuration file
load name=jobid
config name=jobid component_id=1 producer=vm1_1 instance=vm1_1/jobid
start name=jobid interval=1000000 offset=-100000


Within ldmsd_controller or in a configuration file
load name=jobid
config name=jobid component_id=1 producer=vm1_1 instance=vm1_1/jobid file=/var/run/rman/node/jobinfo
start name=jobid interval=1000000 offset=-100000

Slurm 2.x installations can populate /var/run/ldms.jobid by adding the following lines to slurm.epilog and slurm.prolog, respectively.

echo "JOBID=0" > /var/run/ldms.jobinfo

and

echo JOBID=$SLURM_JOBID > /var/run/ldms.jobinfo
echo UID=$SLURM_UID >> /var/run/ldms.jobinfo
echo USER=$SLURM_JOB_USER >> /var/run/ldms.jobinfo