[backend] Make Request and limits configurable in KFP Driver and Launcher

[rimolive]

The memory and cpu limit/request ratio for driver and launcher are hard-coded values.

In a multi-tenant cluster with a configured resource limit/request ratio this will block the pods from being started. Pod is not getting created from the resulted argoworkflow of the KFP DSL, since the root-driver and kfp-launcher containers are not even being admitted.

Error node pipeline-step-f9888.root-driver: task 'document-ingestion-f9888.root-driver' errored:pods"pipeline-step-f9888-172800538" is forbidden: memorymax limit to request ratio per Container is 3, butprovidedratio is 8.000000

Environment

• How did you deploy Kubeflow Pipelines (KFP)?
  Manifests, multi-user
• KFP version:
  2.3.0
• KFP SDK version:
  2.10.1

Steps to reproduce

• Deploy KFP
• Specify a LimitRange in kubeflow namespace. Note that maxLimitRequestRatio but have a value higher than 3
• Create a pipeline run

Expected result

Pipeline should run without breaking any restrictions from the LimitRange

Materials and Reference

N/A

---

Impacted by this bug? Give it a 👍.