[SOLVED] Which algorithim does Kubernetes use to assign pods to nodes?

Which algorithim does Kubernetes use to assign pods to nodes?

This is more of cost estimation question than how to use features like node affinity.

So basically there are m pods with some constraints like:

each pod of specific Deployments / StatefulSets should be on a different kubernetes node
pods of specific Deployments / StatefulSets should be balanced over 3 availability zones

Now, I want to find how many nodes (all same types) I will need to host given set of Deployments / StatefulSets.

I first thought this of more like an assignment problem to be solved using Hungarian Algorithim but this seems much more complex in term like multi dimensional constraints.

Solution

By my knowledge the algorithm used by default by the kube-scheduler is described on github here.

It explains how it works. It first filters nodes that do not meet the requirements of the pods, e.g. resource requests > available resources on nodes, affinity etc.

Then uses a ranking algorithm to determine the best fitting node. For deeper insights on the