Affinity Policy V1.0

Affinity Policy V1.0

This policy is designed for Senlin clusters to exploit the servergroup API exposed by the Nova compute service. The basic policy has been extended to work with vSphere hypervisor when VMware DRS feature is enabled. However, such an extension is only applicable to admin owned server clusters.

Spec

Latest Version

1.0

Available Versions

Version Status Supported Since
1.0 SUPPORTED 2016.10

Applicable Profile Types

This policy is designed to handle the following profile types:
  • os.nova.server-1.0
  • Policy Triggers

    This policy is triggered by the following actions during the respective phases:
    Action Phase
    CLUSTER_RESIZE BEFORE
    CLUSTER_SCALE_OUT BEFORE
    NODE_CREATE BEFORE

    Properties

    availability_zone
    Name of the availability zone to place the nodes.
    enable_drs_extension
    Enable vSphere DRS extension.
    servergroup
    name
    The name of the server group
    policies
    The server group policies.

    Actions Handled

    The policy is capable of handling the following actions:

    • CLUSTER_SCALE_OUT: an action that carries an optional integer value named count in its inputs.
    • CLUSTER_RESIZE: an action that carries various input parameters to resize a cluster. The policy will try to parse the raw inputs if no other policies have done this.
    • NODE_CREATE: an action originated from a node creation RPC request. The policy is capable of processing the node associated with this action.

    The policy will be checked BEFORE any of the above mentioned actions is executed. When the action is CLUSTER_RESIZE, the affinity policy will check if it is about the creation of new nodes. If the resize request is about the removal of existing nodes, the policy won’t block the request.

    Senlin engine respects outputs (i.e. number of nodes to create) from other policies, if any. If no such data exists, it then checks the user-provided “count” input if there is one. The policy is also designed to parse a cluster resize request and see if there are new nodes to be created.

    After validating the count value, the affinity policy proceeds to update the data property of the action with node placement data. For example:

    {
      'placement': {
        'count': 2,
        'placements': [
          {'servergroup': 'XYZ-ABCD'},
          {'servergroup': 'XYZ-ABCD'}
        ]
      }
    }
    

    Scenarios

    S1: Inheriting Server Group from Profile

    When attaching the affinity policy to a cluster that is based on a profile type of os.nova.server-1.0, if the profile contains scheduler_hints property and the property value (a collection) has a group key, the engine will use the value of the group key as a Nova server group name. In this case, the affinity policy will check if the specified server group does exist. If the group doesn’t exist, or the rules specified in the group doesn’t match that specified (or implied) by the affinity policy, you will get an error when attaching the policy to the cluster. If, on the contrary, the group is found and the rules do match that of the current policy, the engine will record the ID of the server group into the policy binding data. The engine also saves a key-value pair inherited_group: True into the policy binding data, so that in future the engine knows that the server group wasn’t created from scratch by the affinity policy. This will lead to the following data stored into the policy binding data:

    {
      'AffinityPolicy': {
        'version': 1.0,
        'data': {
          'servergroup_id': 'XYZ-ABCD',
          'inherited_group': True
        }
      }
    }
    

    When an affinity policy is to be detached from a cluster, the Senlin engine will check and learn the server group was not created by the affinity policy. The engine will not delete the server group.

    Before any of the targeted actions is executed, the affinity policy gets a chance to be checked. It does so by looking into the policy binding data and find out the server group ID to use. For node creation requests, the policy will yield some data into action.data property that looks like:

    {
      'placement': {
        'count': 2,
        'placements': [
          {'servergroup': 'XYZ-ABCD'},
          {'servergroup': 'XYZ-ABCD'}
        ]
      }
    }
    

    S2: Creating A Server Group when Needed

    When attaching an affinity policy to a cluster, if the cluster profile doesn’t contain a scheduler_hints property or there is no group value specified in the scheduler_hints property, the engine will create a new server group by invoking the Nova API, providing it the policies specified (or implied) as inputs. The ID of the newly created server group is then saved into the policy binding data, along with a inherited_group: False key-value pair. For example:

    {
      'AffinityPolicy': {
        'version': 1.0,
        'data': {
          'servergroup_id': 'XYZ-ABCD',
          'inherited_group': False
        }
      }
    }
    

    When such a policy is later detached from the cluster, the Senlin engine will check and learn that the server group should be deleted. It then deletes the server group by invoking Nova API.

    When the targeted actions are about to be executed, the protocol for checking and data saving is identical to that outlined in scenario S1.

    S3: Enabling vSphere DRS Extensions

    When you have vSphere hosts (with DRS feature enabled) serving hypervisors to Nova, a vSphere host is itself a collection of physical nodes. To make better use of the vSphere DRS feature, you can enable the DRS extension by specifying enable_drs_extension: True in your affinity policy.

    When attaching and detaching the affinity policy to/from a cluster, the engine operations are the same as described in scenario S1 and S2. However, when one of the targeted actions is triggered, the affinity policy will first check if the availability_zone property is set and it will use “nova” as the default value if not specified.

    The engine then continues to check the input parameters (as outlined above) to find out the number of nodes to create. It also checks the server group ID to use by looking into the policy binding data.

    After the policy has collected all inputs it needs, it proceeds to check the available vSphere hypervisors with DRS enabled. It does so by looking into the hypervisor_hostname property of each hypervisor reported by Nova (Note: retrieving hypervisor list is an admin-only API, and that is the reason the vSphere extension is only applicable to admin-owned clusters). The policy attempts to find a hypervisor whose host name contains drs. If it fails to find such a hypervisor, the policy check fails with the action’s data field set to:

    {
      'status': 'ERROR',
      'status_reason': 'No suitable vSphere host is available.'
    }
    

    The affinity uses the first matching hypervisor as the target host and it forms a string containing the availability zone name and the hypervisor host name, e.g. “nova:vsphere_drs_1”. This string will later be used as the availability zone name sent to Nova. For example, the following is sample result when applying the affinity policy to a cluster with vSphere DRS enabled.

    {
      'placement': {
        'count': 2,
        'placements': [{
            'zone': 'nova:vsphere_drs_1',
            'servergroup': 'XYZ-ABCD'
          }, {
            'zone': 'nova:vsphere_drs_1',
            'servergroup': 'XYZ-ABCD'
          }
        ]
      }
    }
    

    NOTE: The availability_zone property is effective even when the vSphere DRS extension is not enabled. When availability_zone is explicitly specified, the affinity policy will pass it along with the server group ID to the Senlin engine for further processing, e.g.:

    {
      'placement': {
        'count': 2,
        'placements': [{
            'zone': 'nova_1',
            'servergroup': 'XYZ-ABCD'
          }, {
            'zone': 'nova_1',
            'servergroup': 'XYZ-ABCD'
          }
        ]
      }
    }
    
    Creative Commons Attribution 3.0 License

    Except where otherwise noted, this document is licensed under Creative Commons Attribution 3.0 License. See all OpenStack Legal Documents.