Describe the bug

We are seeing issues where the scheduler is scheduling pods to nodes that have cpu/memory availability but don't have pod count availability and those pods just sit in a pending state and get lease returned multiple times.

Expected behavior

I expect Armada to consider node pod limits