Splunk® Enterprise

Admin Manual

Splunk Enterprise version 7.2 is no longer supported as of April 30, 2021. See the Splunk Software Support Policy for details. For information about upgrading to a supported version, see How to upgrade Splunk Enterprise.
This documentation does not apply to the most recent version of Splunk® Enterprise. For documentation on the most recent version, go to the latest release.

workload_pools.conf

The following are the spec and example files for workload_pools.conf.

workload_pools.conf.spec

#   Version 7.2.8
#

OVERVIEW


# This file contains descriptions of the settings that you can use to
# configure workloads for splunk.
#
# There is a workload_pools.conf file in the $SPLUNK_HOME/etc/system/default/ directory.
# Never change or copy the configuration files in the default directory.
# The files in the default directory must remain intact and in their original
# location.
#
# To set custom configurations, create a new file with the name workload_pools.conf in
# the $SPLUNK_HOME/etc/system/local/ directory. Then add the specific settings
# that you want to customize to the local configuration file.
# For examples, see workload_pools.conf.example. You may need to restart the Splunk instance
# to enable configuration changes.
#
# To learn more about configuration files (including file precedence) see the
# documentation located at
# http://docs.splunk.com/Documentation/Splunk/latest/Admin/Aboutconfigurationfiles
#

GLOBAL SETTINGS


# Use the [default] stanza to define any global settings.
#   * You can also define global settings outside of any stanza, at the top of
#     the file.
#   * Each .conf file should have at most one default stanza. If there are
#     multiple default stanzas, settings are combined. In the case of
#     multiple definitions of the same setting, the last definition in the
#     file takes precedence.
#   * If a setting is defined at both the global level and in a specific
#     stanza, the value in the specific stanza takes precedence.
#
# CAUTION: Do not alter the settings in the workload_pools.conf file unless you know
#     what you are doing.  Improperly configured worloads might result in
#     splunkd crashes, memory overuse, or both.

[general]

enabled = <bool>
* Specifies whether workload management has been enabled on the system or not.
* This setting only applies to the default stanza as a global setting.
* Default: false

default_pool = <string>
* Specifies the default workload pool to be used at runtime for search workloads.
* Admin users could specify workload pools associated with roles. If no workload
  pool can be found, then we fall back to this default_pool that is defined in
  the general stanza in workload.conf.
* This setting is only applicable when workload management has been enabled in
  the system. If workload management has been enabled, this is a mandatory setting.

ingest_pool = <string>
* Specifies the workload pool for the splunkd process that controls ingestion
  and other actions in the Splunk deployment.
* Use this setting to guarantee a minimum lower-bound for resources for tasks
  controlled and managed by splunkd.
* This setting is only applicable when workload management has been enabled in
  the system. If workload management has been enabled, this is a mandatory setting.

workload_pool_base_dir_name = <string>
* Specifies the base controller directory name for Splunk cgroups on Linux to be used by a Splunk deployment.
* Workload pools created from the workload management page are all created relative
  to this base directory.
* This setting is only applicable when workload management has been enabled in
  the system. If workload management has been enabled, this is a mandatory setting.
* Default: splunk

[workload_pool:<pool_name>]

cpu_weight = <number>
* Specifies the cpu weight to be used by this workload pool.
* This is effectively a relative ratio or fraction of the total weights assigned
  across all the workload pools.
* Note that this is not a percentage and instead a relative weight as a fraction
  of the total weight calculated by summing all workload pool weights.
* This is a mandatory parameter for the creation of a workload pool and only
  allows positive integral values.
* Default is unset

mem_weight = <number>
* Specifies the memory weight to be used by this workload pool.
* This is effectively a ratio or fraction of the total weights assigned
  across all the workload pools.
* Note that this is not a percentage and instead a relative weight as a fraction
  of the total weight calculated by summing all workload pool weights.
* This is a mandatory parameter for the creation of a workload pool and only
  allows positive integral values.
* Default is unset

workload_pools.conf.example

#   Version 7.2.8
# CAUTION: Do not alter the settings in workload_pools.conf unless you know what you are doing.
# Improperly configured workloads may result in splunkd crashes and/or memory overuse.

[general]
enabled = false
default_pool = pool_1
ingest_pool = pool_2
workload_pool_base_dir_name = splunk

[workload_pool:pool_1]
cpu_weight = 40
mem_weight = 40

[workload_pool:pool_2]
cpu_weight = 30
mem_weight = 30

[workload_pool:pool_3]
cpu_weight = 20
mem_weight = 20

[workload_pool:pool_4]
cpu_weight = 10
mem_weight = 10

Last modified on 09 September, 2019
workflow_actions.conf   workload_rules.conf

This documentation applies to the following versions of Splunk® Enterprise: 7.2.8


Was this topic useful?







You must be logged into splunk.com in order to post comments. Log in now.

Please try to keep this discussion focused on the content covered in this documentation topic. If you have a more general question about Splunk functionality or are experiencing a difficulty with Splunk, consider posting a question to Splunkbase Answers.

0 out of 1000 Characters