System and Cluster Critical Priority Classes
What You'll Learn
You'll learn how to use Kubernetes' two highest-priority built-in PriorityClasses to protect essential system components from eviction or scheduling failures.
Why System Components Need Special Treatment
Imagine a hospital during a crisis: doctors and nurses get first access to resources, not visitors. Similarly, Kubernetes clusters need a way to guarantee that critical system components—like DNS, metrics servers, and the API server itself—always run, even when resources are tight.
Kubernetes provides two pre-configured, highest-priority PriorityClasses specifically for this purpose:
system-cluster-critical
This PriorityClass (priority value: 2000000000) is reserved for components essential to the entire cluster's operation. Think cluster-wide services that, if removed, would make the cluster unusable.
Common uses:
- CoreDNS pods (cluster DNS)
- Metrics server
- CNI network plugins
- Cluster autoscaler
system-node-critical
This PriorityClass (also priority value: 2000000000, but evaluated differently) protects components critical to individual node functionality. These pods must run on specific nodes to keep those nodes operational.
Common uses:
- kube-proxy (node networking)
- Node-level logging/monitoring agents
- Storage plugins required for node operation
How They Work
These PriorityClasses are created automatically by Kubernetes. When you assign them to system pods, the scheduler ensures they:
- Always schedule first when nodes have capacity
- Can preempt lower-priority pods if necessary
- Are protected from eviction during resource pressure (especially with
system-node-critical)
You typically won't assign these priorities to your own application pods—they're reserved for infrastructure maintained by cluster operators.
Key Takeaway:
system-cluster-critical and system-node-critical are Kubernetes' highest-priority classes that guarantee essential cluster and node components always have resources, protecting infrastructure stability even under extreme resource pressure.