cnpg and descheduler #1689
Description
Anyone managed to get descheduler to work with the CNPG operator?
The pods are managed by the operator so rescheduled ignores them, as they are not part of a deployment.
This is the configuration I have tried:
pluginConfig:
- name: DefaultEvictor
args:
ignorePvcPods: false
evictLocalStoragePods: true
labelSelector:
matchExpressions:
- {
key: cnpg.io/instanceRole,
operator: NotIn,
values: [primary],
}
- name: RemoveDuplicates
- name: RemovePodsHavingTooManyRestarts
args:
podRestartThreshold: 100
includingInitContainers: true
- name: RemovePodsViolatingNodeAffinity
args:
nodeAffinityType:
- requiredDuringSchedulingIgnoredDuringExecution
- name: RemovePodsViolatingNodeTaints
- name: RemovePodsViolatingInterPodAntiAffinity
- name: RemovePodsViolatingTopologySpreadConstraint
args:
constraints:
- DoNotSchedule
- ScheduleAnyway
- name: LowNodeUtilization
args:
thresholds:
cpu: 20
memory: 20
pods: 20
targetThresholds:
cpu: 50
memory: 50
pods: 50
plugins:
balance:
enabled:
- RemoveDuplicates
- RemovePodsViolatingTopologySpreadConstraint
- LowNodeUtilization
deschedule:
enabled:
- RemovePodsHavingTooManyRestarts
- RemovePodsViolatingNodeTaints
- RemovePodsViolatingNodeAffinity
- RemovePodsViolatingInterPodAntiAffinity
The issue I am trying to solve:
- drain a node
- some replicas move to other node
- initial node recovers, but replicas remain on another node and I end up with multiple replicas on the same node.