policy.mdx

layout	page_title	description
docs	Scaling Policies	Scaling policies describe the target resource desired state and how to perform calculations to ensure the current state reaches the desired.

Nomad Autoscaler Scaling Policies

Nomad Autoscaler scaling policies can be configured via the scaling block or by configuration files stored on disk. The options available differ whether you are performing horizontal application/cluster scaling or Dynamic Application Sizing.

Top Level Options

• enabled - A boolean flag that allows operators to administratively disable a policy from active evaluation.

• min - The minimum running count of the targeted resource. This can be 0 or any positive integer.

• max - The maximum running count of the targeted resource. This can be 0 or any positive integer.

Task Group and Cluster Scaling policy Options

The following options are available when using the Nomad Autoscaler to perform horizontal application scaling or horizontal cluster scaling.

• cooldown - A time interval after a scaling action during which no additional scaling will be performed on the resource. It should be provided as a duration (e.g.: "5s", "1m"). If omitted the configuration value policy_default_cooldown from the agent will be used.

• evaluation_interval - Defines how often the policy is evaluated by the Autoscaler. It should be provided as a duration (e.g.: "5s", "1m"). If omitted the configuration value default_evaluation_interval from the agent will be used.

• on_check_error - Defines how to handle errors during check evaluation. Possible values are "fail" or "ignore". If set to "fail" the policy evaluation will stop if any check returns an error and no scaling action will take place. If set to "ignore" any errors returned by a check will be ignored when computing the scaling action. This value may be overridden individually by setting on_error. Defaults to "ignore".

• target - Defines where the autoscaling target is running. Detailed information on the configuration options can be found on the Target Plugins page.

• check - Specifies one or more checks to be executed when determining if a scaling action is required.

check Options

• source - The APM plugin that should handle the metric query. If omitted, this defaults to using the Nomad APM.

• query - The query to run against the specified APM. Currently this query should return a single value. Detailed information on the configuration options can be found on the APM Plugins page.

• query_window - Defines how far back to query the APM for metrics. It should be provided as a duration (e.g.: "5s", "1m"). Defaults to 1m. This value may not be supported by all APM plugins. Refer to the documentation of the specific APM plugin being used for more information.

• query_window_offset - Defines a time offset from the current time to apply to the query window. A short offset (for example, 1m) can be used to avoid reading unstable and incomplete data from the APM. A long offset (168h, representing 7 days ago) can be used for predictive scaling based on past data.

• group - Specifies which checks should treated as correlated when the policy is evaluated. Refer to Check Grouping for more information.

• on_error - Defines how to handle errors during the check evaluation. Possible values are "fail" or "ignore". If set to "fail" the policy evaluation will stop in case an error occurs and not scaling action will take place. If set to "ignore" the result of this check will not be taken into considation when computing the scaling action. If not set the value of on_check_error will be used.

• strategy - The strategy to use, and it's configuration when calculating the desired state based on the current count and the metric returned by the APM. Detailed information on the configuration options can be found on the Strategy Plugins page. Strategies for Dynamic Application Sizing are not allowed in this case.

Example in a Job

A full example of a policy document that can be written into the Nomad task group scaling block can be seen below.

job "example" {
  group "app" {
    scaling {
      min     = 2
      max     = 10
      enabled = true

      policy {
        evaluation_interval = "5s"
        cooldown            = "1m"

        check "active_connections" {
          source = "prometheus"
          query  = "scalar(open_connections_example_cache)"

          strategy "target-value" {
            target = 10
          }
        }
      }
    }
  }
}

Example in a File

An example of a policy document that can be placed in a file within the policy_dir can be seen below. Multiple policies can be defined in the same file using multiple scaling blocks.

scaling "aws_cluster_policy" {
  enabled = true
  min     = 1
  max     = 2

  policy {
    cooldown            = "2m"
    evaluation_interval = "1m"

    check "cpu_allocated_percentage" {
      source = "prometheus"
      query  = "..."

      strategy "target-value" {
        target = 70
      }
    }

    check "mem_allocated_percentage" {
      source = "prometheus"
      query  = "..."

      strategy "target-value" {
        target = 70
      }
    }

    target "aws-asg" {
      dry-run             = "false"
      aws_asg_name        = "hashistack-nomad_client"
      node_class          = "hashistack"
      node_drain_deadline = "5m"
    }
  }
}

scaling "azure_cluster_policy" {
  enabled = true
  min     = 1
  max     = 2

  policy {
    ...
    target "azure-vmss" {
      resource_group      = "hashistack"
      vm_scale_set        = "clients"
      node_class          = "hashistack"
      node_drain_deadline = "5m"
    }
  }
}

Task (DAS) policy Options

This functionality only exists in Nomad Autoscaler Enterprise. This is not present in the open source version of Nomad Autoscaler.

The following options are available when using the Nomad Autoscaler Enterprise to perform Dynamic Application Sizing recommendations for task resources. When using the scaling block for Dynamic Application Sizing, the block requires a label to identify which resource it relates to. It currently supports cpu and mem labels, examples of which can be seen below.

• cooldown - A time interval after a scaling action during which no additional scaling will be performed on the resource. It should be provided as a duration (e.g.: "5s", "1m"). If omitted the configuration value policy_default_cooldown from the agent will be used.

• evaluation_interval - Defines how often the policy is evaluated by the Autoscaler. It should be provided as a duration (e.g.: "5s", "1m"). If omitted the configuration value default_evaluation_interval from the agent will be used.

• target - Defines where the autoscaling target is running. Detailed information on the configuration options can be found on the Target Plugins page.

• check - Specifies one check to be executed when determining if a recommendation is required. Only one check is permitted per scaling block within Dynamic Application Sizing.

check Options

• strategy - The strategy to use, and it's configuration when calculating the desired state based on the current value and the available historic data. Detailed information on the configuration options can be found on the Strategy Plugins page. Only Dynamic Application Sizing strategies are allowed.

Example

The following examples are minimal blocks which can be used to configure CPU and Memory based sizing recommendations for a Nomad job task.

scaling "cpu" {
  policy {
    check "96pct" {
      strategy "app-sizing-percentile" {
        percentile = "96"
      }
    }
  }
}

scaling "mem" {
  policy {
    check "max" {
      strategy "app-sizing-max" {}
    }
  }
}