A common problem in k8s is the fragmentation of available resources across nodes. Inadvertently, pods end up scheduled in such a way that for any new pod, ALL of the resources requested by it are collectively unavailable at any single node, making it un-schedulable. This requires a scale-up, even though cluster itself might have much more overall capacity available. This could be simply avoided by carefully re-scheduling the pods across the nodes in a smart manner.

In this session, Yash will outline a custom pod migration strategy derived from classic retro games like Jenga and Tetris that carefully re-organizes and optimizes the cluster to:

1) Fit any new pods stuck waiting for resources, within the existing infrastructure,
2) Avoid such a fragmented scenario by periodically reorganizing the cluster nodes to keep a healthy balance of different resources intensive applications.

He will showcase the production improvements and motivation for further such solutions to this problem.