» affinity Stanza

Placement job -> affinity
job -> group -> affinity
job -> group -> task -> affinity

The affinity stanza allows operators to express placement preference for a set of nodes. Affinities may be expressed on attributes or client metadata. Additionally affinities may be specified at the job, group, or task levels for ultimate flexibility.

job "docs" {
  # Prefer nodes in the us-west1 datacenter
  affinity {
    attribute = "${node.datacenter}"
    value     = "us-west1"
    weight    = 100
  }

  group "example" {
    # Prefer the "r1" rack
    affinity {
      attribute  = "${meta.rack}"
      value     = "r1"
      weight    = 50
    }

    task "server" {
      # Prefer nodes where "my_custom_value" is greater than 5
      affinity {
        attribute = "${meta.my_custom_value}"
        operator  = ">"
        value     = "3"
        weight    = 50
      }
    }
  }
}

Affinities apply to task groups but may be specified within job and task stanzas as well. Job affinities apply to all groups within the job. Task affinities apply to the whole task group that the task is a part of.

Nomad will use affinities when computing scores for placement. Nodes that match affinities will have their scores boosted. Affinity scores are combined with other scoring factors such as bin packing. Operators can use weights to express relative preference across multiple affinities. If no nodes match a given affinity, placement is still successful. This is different from constraints where placement is restricted only to nodes that meet the constraint's criteria.

» affinity Parameters

  • attribute (string: "") - Specifies the name or reference of the attribute to examine for the affinity. This can be any of the Nomad interpolated values.

  • operator (string: "=") - Specifies the comparison operator. The ordering is compared lexically. Possible values include:

    =
    !=
    >
    >=
    <
    <=
    regexp
    set_contains_all
    set_contains_any
    version
    

    For a detailed explanation of these values and their behavior, please see the operator values section.

  • value (string: "") - Specifies the value to compare the attribute against using the specified operation. This can be a literal value, another attribute, or any Nomad interpolated values.

  • weight (integer: 50) - Specifies a weight for the affinity. The weight is used during scoring and must be an integer between -100 to 100. Negative weights act as anti affinities, causing nodes that match them to be scored lower. Weights can be used when there is more than one affinity to express relative preference across them.

» operator Values

This section details the specific values for the "operator" parameter in the Nomad job specification for affinities. The operator is always specified as a string, but the string can take on different values which change the behavior of the overall affinity evaluation.

affinity {
  operator = "..."
}
  • "regexp" - Specifies a regular expression affinity against the attribute. The syntax of the regular expressions accepted is the same general syntax used by Perl, Python, and many other languages. More precisely, it is the syntax accepted by RE2 and described at in the Google RE2 syntax.

    affinity {
      attribute = "..."
      operator  = "regexp"
      value     = "[a-z0-9]"
      weight    = 50
    }
    
  • "set_contains_all" - Specifies a contains affinity against the attribute. The attribute and the list being checked are split using commas. This will check that the given attribute contains all of the specified elements.

    affinity {
      attribute = "..."
      operator  = "set_contains"
      value     = "a,b,c"
      weight    = 50
    }
    
  • "set_contains" - Same as set_contains_all

  • "set_contains_any" - Specifies a contains affinity against the attribute. The attribute and the list being checked are split using commas. This will check that the given attribute contains any of the specified elements.

    affinity {
      attribute = "..."
      operator  = "set_contains"
      value     = "a,b,c"
      weight    = 50
    }
    
  • "version" - Specifies a version affinity against the attribute. This supports a comma-separated list of values, including the pessimistic operator. For more examples please see the go-version repository for more specific examples.

    affinity {
      attribute = "..."
      operator  = "version"
      value     = ">= 0.1.0, < 0.2"
      weight    = 50
    }
    

» affinity Examples

The following examples only show the affinity stanzas. Remember that the affinity stanza is only valid in the placements listed above.

» Kernel Data

This example adds a preference for running on nodes which have a kernel version higher than "3.19".

affinity {
  attribute = "${attr.kernel.version}"
  operator  = "version"
  value     = "> 3.19"
  weight    = 50
}

» Operating Systems

This example adds a preference to running on nodes that are running Ubuntu 14.04

affinity {
  attribute = "${attr.os.name}"
  value     = "ubuntu"
  weight    = 50
}

affinity {
  attribute = "${attr.os.version}"
  value     = "14.04"
  weight    = 100
}

» Meta Data

The following example adds a preference to running on nodes with specific rack metadata

affinity {
  attribute = "${meta.rack}"
  value     = "rack1"
  weight    = 50
}

The following example adds a preference to running on nodes in a specific datacenter.

affinity {
  attribute = "${node.datacenter}"
  value     = "us-west1"
  weight    = 50
}

» Cloud Metadata

When possible, Nomad populates node attributes from the cloud environment. These values are accessible as filters in affinities. This example adds a preference to run this task on nodes that are memory-optimized on AWS.

affinity {
  attribute = "${attr.platform.aws.instance-type}"
  value     = "m4.xlarge"
  weight    = 50
}

» Placement Details

Operators can run nomad alloc status -verbose to get more detailed information on various factors, including affinities that affect the final placement.

» Example Placement Metadata

The following is a snippet from the CLI output of nomad alloc status -verbose <alloc-id> showing scoring metadata.

Placement Metrics
Node                                  binpack  job-anti-affinity  node-reschedule-penalty  node-affinity  final score
30bd48cc-d760-1096-9bab-13caac424af5  0.225    -0.6               0                        1              0.208
f2aa8b59-96b8-202f-2258-d98c93e360ab  0.225    -0.6               0                        1              0.208
86df0f74-15cc-3a0e-23f0-ad7306131e0d  0.0806   0                  0                        0              0.0806
7d6c2e9e-b080-5995-8b9d-ef1695458b52  0.0806   0                  0                        0              0.0806

The placement score is affected by the following factors.

  • bin-packing - Scores nodes according to how well they fit requirements. Optimizes for using minimal number of nodes.
  • job-anti-affinity - A penalty added for additional instances of the same job on a node, used to avoid having too many instances of a job on the same node.
  • node-reschedule-penalty - Used when the job is being rescheduled. Nomad adds a penalty to avoid placing the job on a node where it has failed to run before.
  • node-affinity - Used when the criteria specified in the affinity stanza matches the node.