Hi, everyone.
I’m reading the kubernetes source code now to figure out scheduler framework plugin interpodaffinity performance problem under the large scale cluster (over thousand of nodes).
I read the framework process PreFilter → Filter → PreScore → Score, and cannot understand some strange code cases. The plugin interpodaffinity PreScore function accept the score nodes []*framework.NodeInfo as parameters, but compute all allNodes from pl.sharedLister.NodeInfos()
Here is code link PreScore .
Hope someone could answer the question. Thanks!