Pod Disruption Budgets¶
OpenStack-Helm leverages PodDisruptionBudgets to enforce quotas that ensure that a certain number of replicas of a pod are available at any given time. This is particularly important in the case when a Kubernetes node needs to be drained.
These quotas are configurable by modifying the minAvailable field
within each PodDisruptionBudget manifest, which is conveniently mapped
to a templated variable inside the values.yaml file.
The min_available within each service’s values.yaml file can be
represented by either a whole number, such as 1, or a percentage,
such as 80%.  For example, when deploying 5 replicas of a pod (such as
keystone-api), using min_available: 3 would enforce policy to ensure at
least 3 replicas were running, whereas using min_available: 80% would ensure
that 4 replicas of that pod are running.
Note: The values defined in a PodDisruptionBudget may
conflict with other values that have been provided if an operator chooses to
leverage Rolling Updates for deployments.  In the case where an
operator defines a maxUnavailable and maxSurge within an update strategy
that is higher than a minAvailable within a pod disruption budget,
a scenario may occur where pods fail to be evicted from a deployment.