In kubernetes, I want to restrict my pod scheduling on worker node if memory utilisation of worker node goes above 50% of its capacity?

Question

For eg: worker node(W) have 64GB of memory and 16vCPUs

resources assigned to a pod : cpu - 4 , memory- 8GB

Now only 4 pods can be scheduled on worker node(W).

How can i achieve this ?

Node affinity , node selectors or taint and tolerations do not work in this case.

Answer 1

Use a custom scheduler or a scheduler extender. First, ensure the Kubernetes Metrics Server is running to collect real-time memory data from the nodes. Then, either modify the default scheduler / create a custom one that checks the node’s memory usage before placing new pods. If the memory usage is over 50%, the scheduler should block new pods from being scheduled, even if there are still available resources based on the pod's requests. Also set up Prometheus to monitor memory usage and trigger alerts or actions when a node hits the threshold, preventing new pod placements. This approach should ig.