Region Placement Policy¶

Properties¶

Sample¶

A typical spec for a region placement policy is shown in the following sample:

# Sample placement policy for cross-region placement
type: senlin.policy.region_placement
version: 1.0
description: A policy for node placement across regions
properties:
  regions:
    - name: RegionOne
      weight: 100
      cap: 150
    - name: RegionTwo
      weight: 100
      cap: 200

In this sample spec, two regions are provided, namely “region_1” and “region_2”. There are “weight” and “cap” attributes associated with them, both of which are optional.

The “weight” value is to be interpreted as a relative number. The value assigned to one region has to be compared to those assigned to other regions for an assessment. In the sample shown above, region_1 and region_2 are assigned weights with 100 and 200 respectively. This means that among every 3 nodes creation, one is expected to be scheduled to region_1 and the other 2 is expected to be scheduled to region_2. Put it in another way, the chance for region_2 receiving a node creation request is twice of that for region_1.

The “weight” value has to be a positive integer, if specified. The default value is 100 for all regions whose weight is omitted.

There are cases where each region has different amounts of resources provisioned so their capacity for creating and running nodes differ. To deal with these situations, you can assign a “cap” value to such a region. This effectively tells the Senlin engine that a region is not supposed to accommodate nodes more than the specified number.

Validation¶

When creating a region placement policy, the Senlin engine validates whether the region names given are all known to be available regions by the keystone identity service. Do NOT pass in an invalid region name and hope Senlin can create a region for you.

Later on when the policy is triggered by node creation or deletion, it always validates if the provided regions are still valid and usable.

Node Distribution¶

After a region placement policy is attached to a cluster and enabled, all future node creations (by cluster scaling for example) will trigger an evaluation of the policy.

The region placement policy will favor regions with highest weight value when selecting a region for nodes to be created. It will guarantee that no more than the provided cap number of nodes will be allocated to a specific region.

Node distribution is calculated not only when new nodes are created and added to a cluster, it is also calculated when existing nodes are to be removed from the cluster. The policy will strive to maintain a distribution close to the one computed from the weight distribution of all regions.