Binpacking support for DoEKS

[hitsub2]

Community Note

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request
• Please do not leave "+1" or other comments that do not add relevant new information or questions, they generate extra noise for issue followers and do not help prioritize the request
• If you are interested in working on this issue or have submitted a pull request, please leave a comment

What is the outcome that you are trying to reach?

For running batch jobs, like Spark workloads, we want to have higer resource utilization and cost efficiency to avoid spreading pods across nodes leading nodes can be scaled-in in time.

By default, the scheduling-plugin NodeResourcesFit use the LeastAllocated for score strategies. For the long running workloads, that is good because of high availability. But for batch jobs, like Spark workloads, this would have high cost. By changing the from LeastAllocated to MostAllocated, it avoids spreading pods across all running nodes, leading to higher resource utilization and better cost efficiency.

Describe the solution you would like

Support binpacking strategy.

Describe alternatives you have considered

Using the custom scheduler like Yunikorn or Volcano. But it is complicated in some scenarios that we only need the binpacking.

Additional context

[github-actions]

This issue has been automatically marked as stale because it has been open 30 days
with no activity. Remove stale label or comment or this issue will be closed in 10 days