Enhancement Description

Many production Kubernetes clusters blend on-demand (higher-SLA) and spot/preemptible (lower-SLA) nodes to optimize costs while maintaining reliability for critical workloads. Platform teams need a safe default that keeps most workloads away from risky capacity, while allowing specific workloads to opt-in with explicit thresholds like "SLA ≥ 95%".

Currently, NodeAffinity supports numeric comparisons (Gt, Lt, etc.) but lacks the operational benefits that taints/tolerations provide:

• Policy orientation: NodeAffinity is per-pod; to keep most pods away from low-SLA nodes requires editing every workload. Taints invert control: nodes declare risk; only pods with matching tolerations may land.
• Eviction semantics: Affinity has no eviction capability. Taints support NoExecute with tolerationSeconds, enabling operators to drain/evict pods when a node's SLA degrades or spot instances are reclaimed.
• Operational ergonomics: Centralized, node-side policy is consistent with other safety taints (e.g., disk-pressure, memory-pressure).

This enhancement extends core/v1 Toleration to support numeric comparison operators (Lt, Gt) when matching Node Taints. This preserves the well-understood safety model of taints/tolerations while enabling threshold-based placement for SLA-aware scheduling.

Benefits for DRA and AI Workloads

• Cost-reliability optimization: Bind resource claims to reliability tiers via taints with opt-in tolerations
• Stage-aware placement: Different pipeline stages can tolerate different risk levels explicitly
• Resilience after preemption: Use NoExecute/tolerationSeconds for graceful drain and controlled failover
• Multi-tenant fairness: Prevent monopolization of high-SLA resources by requiring explicit tolerations
• Smooth burst handling: Allow bursts to land on low-SLA pools with clear safety boundaries

The scheduler impact is limited to the existing TaintToleration Filter; no new scheduling stages or algorithms are required.

/sig scheduling
/sig apps
/stage alpha

/cc @ahg-g @alculquicondor @johnbelamaric @sanposhiho @kubernetes/sig-scheduling-misc

• One-line enhancement description (can be used as a release note): Add numeric comparison operators (Lt, Gt) to Tolerations for SLA-based scheduling with threshold-based taint matching.
• Kubernetes Enhancement Proposal: https://github.com/kubernetes/enhancements/tree/master/keps/sig-scheduling/5471-enable-sla-based-scheduling
• Discussion Link: Allow nodes to declare failure probability/SLA kubernetes#118669
• PRs by stage and milestone:
  • Alpha - v1.35
    • KEP (k/enhancements) update PR(s): KEP-5471 Extended Toleration Operators for Threshold-Based Placement #5473
    • Code (k/k) update PR(s):
      • KEP-5471: Extend tolerations operators kubernetes#134665
    • Docs (k/website) update PR(s):
      • KEP-5471: Add docs for extend tolerations for alpha website#52824
      • Blog: KEP-5471: [Feature blog]Add Blog Post for Extended Toleration Operators Alpha in 1.35 website#52897