Skip to content
Kueue on EKS

KEDA vs Kueue

KEDA
Kubernetes Event-driven Autoscaling — scales Deployment or StatefulSet replicas from external metrics such as SQS queue depth. Complements Kueue; does not replace batch admission. and Kueue
Kubernetes-native job admission controller — decides when batch Jobs may create pods based on quota, priority, and resource flavors. often land on the same EKS cluster. They answer different questions and usually sit next to each other — they do not replace each other.

Side-by-side: KEDA scales Deployments from Kafka, SQS, or Prometheus; Kueue admits batch Jobs into Admitted or Queued by quota
KEDA Kueue
Question How many replicas from external demand? Which batch jobs may run given quota?
Unit Deployment, StatefulSet, ScaledJob Job, Workload
Trigger SQS, Kafka, Prometheus, cron Internal quota and priority
Outcome Replica count goes up or down Workload is Admitted or stays Queued
EKS add-on Available No (install via GitOps)
Web apps Common use case Not for always-on serving

Use together

Section titled “Use together”

Event depth can grow workers freely; those workers still submit Jobs that must pass Kueue quota before pods appear:

Pipeline: events to KEDA to worker Deployment to Jobs to Kueue admission to Auto Mode nodes 
KEDA scales workers from queue depth
   → workers submit training Jobs
   → Kueue admits Jobs by GPU or CPU quota

Use Kueue alone when

Section titled “Use Kueue alone when”
  • Batch Jobs compete for a fixed or capped pool
  • Multiple projects need fair share
  • You cap concurrent Jobs on Spot

Use KEDA alone when

Section titled “Use KEDA alone when”
  • Workloads scale with events and cluster capacity is not contested
  • No batch admission policy is required

Both deploy through Argo CD on this stack — neither appears in the EKS add-ons list for Auto Mode. This lab installs Kueue only; add KEDA the same GitOps way when you need event-driven replica scaling.