Zone Placement Policy¶
The zone placement policy is designed to enable the deployment and management resource pools across multiple availability zones. Note that the current design is only concerned with the availability zones configured to Nova compute service. Support to Cinder availability zones and Neutron availability zones may be added in future when we have volume storage specific or network specific profile types.
The current implementation of the zone placement policy works with clusters of Nova virtual machines only.
- Name of an availability zone.
- Weight of the availability zone (default is 100).
A typical spec for a zone placement policy is exemplified in the following sample:
# Sample placement policy for cross-availability-zone placement type: senlin.policy.zone_placement version: 1.0 description: A policy for node placement across availability zones properties: zones: - name: zone1 weight: 100 - name: zone2 weight: 100
In this sample spec, two availability zones are provided, namely “
az_2”. Each availability zone can have an optional “
associated with it.
weight” value is to be interpreted as a relative number. The value
assigned to one zone has to be compared to those assigned to other zones for
an assessment. In the sample shown above,
az_2 are assigned
weights of 100 and 200 respectively. This means that among every 3 nodes
creation, one is expected to be scheduled to
az_1 and the other 2 are
expected to be scheduled to
az_2. In other words, the chance for
receiving a node creation request is twice of that for
weight” value has to be a positive integer, if specified. The default
value is 100 for all zones whose weight is omitted.
When creating a zone placement policy, the Senlin engine validates whether the zone names given are all known to be usable availability zones by the Nova compute service. Do NOT pass in an invalid availability zone name and hope Senlin can create a zone for you.
Later on when the zone placement policy is triggered upon node creation or node deletion actions, it always validates if the provided availability zones are still valid and usable.
After a zone 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. Similarly, a node deletion action will also trigger an evaluation of it because the policy’s goal is to maintain the node distribution based on the one computed from the weight distribution of all zones.
The zone placement policy will favor availability zones with highest weight values when selecting a zone for nodes to be created.