Skip to main content

The Descheduler

Key Takeaways for AI & Readers
  • Post-Scheduling Optimization: The Descheduler rebalances an existing cluster by evicting Pods from suboptimal nodes, addressing cluster "skew" that the default scheduler doesn't handle.
  • Addressing Skew Causes: It resolves imbalances caused by node scaling, changes in Pod affinity rules, or uneven resource utilization across nodes.
  • Eviction-Driven Re-scheduling: The Descheduler doesn't schedule directly; instead, it triggers the standard Kubernetes scheduler by evicting Pods, allowing them to be rescheduled onto more appropriate nodes.
  • Periodic Operation: It operates as a periodic job, continuously optimizing resource distribution and ensuring efficient cluster utilization.

The default Kubernetes Scheduler only makes a decision at the moment a pod is created. Once a pod is running on a node, it stays there forever (unless it dies). Over time, this leads to cluster "Skew."

1. Re-balancing the Cluster​

Visualize how the Descheduler evicts pods to fix imbalances.

Node A (Hot)
Node B (Cold)
📥
The standard Scheduler only acts when a pod is created. The Descheduler runs periodically to fix imbalances (like node hotspots or affinity violations) by evicting pods so they land on better nodes.

2. Why do clusters get skewed?​

  • Node Scaling: A new node is added, but existing pods don't move to it.
  • Affinity Changes: Node labels were updated, but pods are still running on nodes that no longer match their constraints.
  • Node Overload: One node is running hot while another is idle.

3. How it works​

The Descheduler is a job that runs periodically. It doesn't "schedule" pods; it evicts them. The standard Kubernetes controllers (Deployments) then notice the pod is gone and recreate it. This time, the scheduler sees the new node and places it there.