Kueue: Save Some QPS for the Rest of Us! How To Manage 100k Updates Pe...
Lightningat KubeCon + CloudNativeCon Europe 2025 · on Thursday, 2025-04-03· at 16:50· Level 0 | ICC Auditorium·
summary
Kueue faces updating position for up to 100,000 jobs and avoiding saturating the API server, so it rejects per-workload position CRD updates as too many etcd writes and size-limited single-object queues. The implemented solution uses CRDs for semantics but moves dynamic positioning into an API Aggregation Layer backed by Kueue storage: workloads live in a heap snapshot that is lazily converted to a ranked list on demand, cutting QPS to etcd and keeping scheduling fast.
AI-written summary · about
chapters (5)
- 0:00Introduction to Kueue
- 1:02Challenges of workload scaling
- 1:46Alternative storage strategies
- 2:29Implementing aggregation layer
- 4:03Balancing performance and ease
related talks
35mKueue: A Kubernetes-native Job Queueing
28mMultitenancy and Fairness at Scale with Kueue: A Case Study
29mBuilding a Batch System for the Cloud with Kueue
33mBatch Systems in Production with Kueue: Multi-Tenancy and Fungibility
24mAccelerate Your AI/ML Workloads With Topology-Aware Scheduling in Kueue
05mScaling Kubescape to Thousands of Nodes
29mScale Your Batch / Big Data / AI Workloads Beyond the K8s...
31mGPU Hunter: Architecting Global GPU Availability With MultiKueue