Beyond the Limits: Scaling Kubernetes Controllers Horizontally
Talkat KubeCon + CloudNativeCon Europe 2025 · on Thursday, 2025-04-03· at 11:45· Level 1 | Hall Entrance S10 | Room D·
summary
Demonstrates a working sharding system that partitions controller work by injecting a shard label via a mutating webhook and tracking membership with per-instance lease objects and a controller ring CRD. Implements a separate sharder that builds a consistent hash ring, a training handover label to avoid concurrent reconciliations, and a demo web hosting operator. Tests run up to 9,000 objects and 300 changes/s across one to five instances and show near‑linear scaling with 99th‑percentile queue latency ~<1s and reconciliation <5s, with small constant overhead.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:24Why scaling controllers matters
- 3:11Controller basics and demo
- 5:21Leader election limitations
- 8:33Introducing sharding
- 10:24Sharding architecture
- 12:11Demo: sharding in action
- 18:57Implementing sharding
- 22:57Testing scalability metrics
- 26:10Summary and call to action
- 27:13Q&A
related talks
21mOur journey to horizontally scalabe controllers
17mBreaking GitOps Bottlenecks: The Power of Sharding
28mScaling in Kubernetes Safely on On-Prem KaaS Across 1,300+ Clusters and 40,000+ Nodes
46mOperators Live in the Past: Designing Reliable Kubernetes Controllers
36mBuilding a Fault-Tolerant Resources Controller on Kubernetes
22mScaling a Financial Institution with Kubernetes
21m07 Lightning talks
37mKubernetesception: How to run a 1000 control planes