Plugin_slurm_sampler 

Date:: 30 Sep 2019

NAME 

Plugin_slurm_sampler - man page for the LDMSD slurm_sampler plugin

Within ldmsd_controller or a configuration file: config name=slurm_sampler producer=PRODUCER instance=INSTANCE [ component_id=COMP_ID ] [ stream=STREAM ] [ job_count=MAX_JOBS ] [ task_count=MAX_TASKS ]

DESCRIPTION 

slurm_sampler is a sampler plugin that collects the information of the Slurm jobs running on the node. It subscribes to the specified stream to which the slurm_notifier SPANK plugin (see Plugin_slurm_notifier(7)) publish Slurm job events (default stream: slurm). The sampler supports multi-tenant jobs.

The job_count option is the number of slots in the LDMS set allocated for concurrent jobs. If the number of concurrent jobs on the node is greater than job_count, the new job will occupy the slot of the oldest job. If job_count is not specified, the default value is 8.

The task_count is the maximum number of tasks per job on the node. If not specified, it is CPU_COUNT. In the event of the sampler failed to obtain CPU_COUNT, the default value is 64.

CONFIG OPTIONS 

name=slurm_sampler: This MUST be slurm_sampler (the name of the plugin).
producer=PRODUCER: The name of the data producer (e.g. hostname).
instance=INSTANCE: The name of the set produced by this plugin. This option is required.
component_id=COMPONENT_ID: An integer identifying the component (default: 0).
stream=STREAM: The name of the LDMSD stream to get the job event data.
job_count=MAX_JOBS: The number of slots to hold job information. If all slots are occupied at the time the new job arrived, the oldest slot is reused. The default value is 8.
task_count=MAX_TASKS: The number of slots for tasks information per job. If not specified, the sampler will try to obtain system CPU_COUNT and use it as task_count. If it failed, the default value is 64.

BUGS 

No known bugs.

EXAMPLES 

Plugin configuration example:

load name=slurm_sampler
config name=slurm_sampler producer=${HOSTNAME} instance=${HOSTNAME}/slurm \
       component_id=2 stream=slurm job_count=8 task_count=8
start name=slurm_sampler interval=1000000 offset=0