Configure the DevZero Scheduler plugins -- target utilization, overprovision limits, cost scoring, and checkpoint restore.

Configuration

The Scheduler is configured via a KubeSchedulerConfiguration stored in a ConfigMap (dz-scheduler-config in kube-system). Each plugin has its own arguments section.

Plugin Weights

Scoring plugins are weighted to control their influence on the final scheduling decision:

Plugin	Weight	Purpose
TaintToleration	6	Kubernetes built-in
NodeAffinity	5	Kubernetes built-in
PodTopologySpread	5	Kubernetes built-in
InterPodAffinity	5	Kubernetes built-in
NodeCost	4	Cost optimization
TargetAllocated	3	Utilization targeting
LimitOverProvisionControl	2	Overprovision safety

Higher weights mean the plugin has more influence. Kubernetes built-in plugins keep the highest weights to respect standard scheduling constraints first.

TargetAllocated

Controls the target resource utilization for node bin-packing.

Field	Type	Range	Default	Description
`cpuAllocatedPercent`	int	1--100	80	Target CPU allocation percentage
`memoryAllocatedPercent`	int	1--100	80	Target memory allocation percentage

pluginConfig:
  - name: TargetAllocated
    args:
      cpuAllocatedPercent: 80
      memoryAllocatedPercent: 80

Setting both values to 99 maximizes bin-packing density. This is useful for batch workloads but may leave less headroom for burst traffic.

LimitOverProvisionControl

Prevents scheduling pods on nodes where total pod limits far exceed the node's capacity.

Field	Type	Range	Default	Description
`cpuOverProvisionPenalizeRatio`	int	>100	180	CPU limits-to-capacity ratio above which nodes are penalized
`memoryOverProvisionPenalizeRatio`	int	>100	180	Memory limits-to-capacity ratio above which nodes are penalized

pluginConfig:
  - name: LimitOverProvisionControl
    args:
      cpuOverProvisionPenalizeRatio: 180
      memoryOverProvisionPenalizeRatio: 180

A ratio of 180 means nodes are penalized when the sum of all pod CPU (or memory) limits exceeds 1.8x the node's total capacity. Lower values are more conservative; higher values allow more overcommit.

NodeCost

Scores nodes based on pricing data from the DevZero control plane.

Field	Type	Required	Description
`controlPlaneAddress`	string	Yes	URL of the metrics collector (e.g., `https://dakr.devzero.dev`)
`controlPlaneToken`	string	No	Direct bearer token (highest priority)
`tokenFromSecret`	object	No	Reference to a Kubernetes Secret containing the token
`tokenFromConfigMap`	object	No	Reference to a ConfigMap containing the token

Token resolution priority: controlPlaneToken > tokenFromSecret > tokenFromConfigMap.

pluginConfig:
  - name: NodeCost
    args:
      controlPlaneAddress: "https://dakr.devzero.dev"
      tokenFromSecret:
        name: "devzero-zxporter-token"
        namespace: "kube-system"
        key: "CLUSTER_TOKEN"

At least one token source must be configured. The Scheduler validates this at startup and will fail to start if no token is available.

Token from Secret

tokenFromSecret:
  name: "devzero-zxporter-token"
  namespace: "kube-system"
  key: "CLUSTER_TOKEN"

Token from ConfigMap

tokenFromConfigMap:
  name: "devzero-zxporter-env-config"
  namespace: "devzero-zxporter"
  key: "CLUSTER_TOKEN"

CheckpointRestore

The CheckpointRestore plugin has no configuration. It operates based on annotations and labels:

Resource	Key	Value	Purpose
Pod annotation	`checkpoint-shim.io/try-restore`	`"true"`	Marks a pod for checkpoint restore
Node label	`dakr.devzero.io/checkpoint-node`	`"true"`	Marks a node as checkpoint-capable

Full Configuration Example

apiVersion: kubescheduler.config.k8s.io/v1
kind: KubeSchedulerConfiguration
profiles:
  - schedulerName: dz-scheduler
    plugins:
      filter:
        enabled:
          - name: CheckpointRestore
      preScore:
        enabled:
          - name: TaintToleration
            weight: 6
          - name: NodeAffinity
            weight: 5
          - name: PodTopologySpread
            weight: 5
          - name: InterPodAffinity
            weight: 5
          - name: NodeCost
            weight: 4
          - name: TargetAllocated
            weight: 3
          - name: LimitOverProvisionControl
            weight: 2
      score:
        enabled:
          - name: TaintToleration
            weight: 6
          - name: NodeAffinity
            weight: 5
          - name: PodTopologySpread
            weight: 5
          - name: InterPodAffinity
            weight: 5
          - name: NodeCost
            weight: 4
          - name: TargetAllocated
            weight: 3
          - name: LimitOverProvisionControl
            weight: 2
        disabled:
          - name: NodeResourcesFit
          - name: NodeResourcesBalancedAllocation
    pluginConfig:
      - name: TargetAllocated
        args:
          cpuAllocatedPercent: 80
          memoryAllocatedPercent: 80
      - name: LimitOverProvisionControl
        args:
          cpuOverProvisionPenalizeRatio: 180
          memoryOverProvisionPenalizeRatio: 180
      - name: NodeCost
        args:
          controlPlaneAddress: "https://dakr.devzero.dev"
          tokenFromSecret:
            name: "devzero-zxporter-token"
            namespace: "kube-system"
            key: "CLUSTER_TOKEN"
      - name: PodTopologySpread
        args:
          defaultingType: "List"

Deployment Resources

The Scheduler deployment defaults:

Setting	Value
Replicas	1
CPU request	200m
Node selector	`kubernetes.io/os: linux`
Priority class	`system-cluster-critical`
Health check	HTTPS on port 10259
Tolerations	All taints

Configuration

On this page