⌂ Home

🎯 Kubernetes Node Selection Process

Repository YAML Files:

k8s/labs/scheduling/nodeselector/nodeselector.yaml — Pod using nodeSelector to target nodes by labels.
k8s/labs/scheduling/nodeselector/notin-nodeselector.yaml — Pod with preferred node affinity using NotIn to avoid nodes labeled env=staging.

PreFilter

Check eligibility

Filter

Eliminate unsuitable nodes

Score

Rank remaining nodes

Bind

Assign pod to node

Scheduler Framework: Kubernetes uses a modular, pluggable framework with extensible phases. The scheduler evaluates pod specifications against cluster state to find the optimal node placement.

📊 Key Decision Factors

💾

Resources

CPU, Memory, Storage availability

🧲

Affinity

Node/Pod preference rules

🚫

Taints

Node restrictions & tolerations

🌐

Topology

Zone/rack distribution

🔽 Node Filtering Process

Available Nodes: 100

All cluster nodes

After Resource Check: 60

Nodes with sufficient CPU/Memory

After Taints Filter: 40

Nodes with matching tolerations

After Affinity: 15

Nodes matching labels/affinity

Best Scored Node: 1

Optimal node selected

🌳 Example: Scheduling Decision

Passed

Failed

❌ Node1: Insufficient memory (has 1GB, needs 2GB)

❌ Node2: Taint key=gpu:NoSchedule (pod has no toleration)

✅ Node3: All checks passed, score: 85

✅ Node4: All checks passed, score: 92 → Selected!

❌ Node5: Missing label disktype=ssd (required by nodeSelector)

📄 Pod with Node Selection Constraints

apiVersion: v1
kind: Pod
metadata:
  name: node-selection-demo
spec:
  # Simple node selector (filter phase)
  nodeSelector:
    disktype: ssd

  # Tolerations allow scheduling on tainted nodes
  tolerations:
  - key: "dedicated"
    operator: "Equal"
    value: "database"
    effect: "NoSchedule"

  # Node affinity for advanced selection
  affinity:
    nodeAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
        nodeSelectorTerms:
        - matchExpressions:
          - key: "topology.kubernetes.io/zone"
            operator: "In"
            values:
            - "us-west-1a"
            - "us-west-1b"

  containers:
  - name: app
    image: nginx:1.21
    resources:
      requests:
        memory: "2Gi"
        cpu: "1000m"
      limits:
        memory: "4Gi"
        cpu: "2000m"

💡 Key Insight: The scheduler ensures pods are placed on nodes that meet ALL constraints. Resource requests are used for scheduling decisions (not limits). Node affinity and nodeSelector filter nodes, while scoring determines the best match among eligible nodes.