Feature Overview (aka. Goal Summary)

The current TopologyAndDuplicates descheduler profile in OpenShift is a combined strategy that enforces topology spread and evicts duplicate pods on the same node. However, this bundled behavior limits flexibility for customers who want to apply one strategy but not the other.

This feature will split the TopologyAndDuplicates profile into two independent profiles — Topology and Duplicates — enabling fine-grained control over pod eviction strategies. This is particularly important for managed OpenShift platforms like ROSA and ARO, where customers cannot customize descheduler configurations at a granular level.

Goals (aka. expected user outcomes)

1. Selective Eviction Strategy Enablement
   Users can enable topology-based evictions without triggering duplicate pod evictions, and vice versa.

1. Improved Stability for Stateful or Critical Workloads
   Avoid unintended evictions of duplicate pods when the only desired behavior is to balance pods across topology domains (zones, racks, etc.).

1. Upstream Feature Parity and Configuration Flexibility
   Bring OpenShift Descheduler configuration closer to upstream descheduler capabilities, especially in managed clusters (ROSA/ARO) where direct configuration overrides are not permitted.

1. Reduced Operational Risk and Pod Churn
   By disabling unnecessary duplicate pod evictions, customers can reduce production pod disruptions and improve workload reliability.