Add the "group_by" attribute for AFD alarm rules

[OpenStack Infra]

https://review.openstack.org/358448
Dear bug triager. This bug was created since a commit was marked with DOCIMPACT.

commit f86b49893e25500127884218dd06aa94021619cf
Author: Swann Croiset <email address hidden>
Date: Mon Aug 22 09:14:57 2016 +0200

    Add the "group_by" attribute for AFD alarm rules

    The 'group_by' attribute defines an array of fields specifying on which
    field(s) the received metrics should be grouped.

    One alarm definition can now dynamically generate as many alarms that
    there are different values for a field/dimension:

    Say that the following alarm is defined:

      name: 'filesystem-warning'
      severity: 'warning'
      trigger:
        rules:
          - metric: 'fs_space_percent_free'
            window: 30
            periods: 1
            function: 'avg',
            relational_operator: '<='
            threshold: 5
            group_by: ['fs']

    If the AFD filter receives metrics for 3 different 'fs' values in the
    defined window:

      - fs_space_percent_free[fs='/']=1
      - fs_space_percent_free[fs='/var/log']=4
      - fs_space_percent_free[fs='/var/lib/nova']=95

    When evaluating the alarm, the filter will emit one AFD metric and the
    Payload will contain the 2 alerts:

      - fs='/' (severity=WARNING)
      - fs='/var/log' (severity=WARNING)

    Implements-blueprint: afd-alarm-fields-matching

    DocImpact

    Change-Id: I4ec4fa0c4627462bf045487eea475f5f81c33acd