Platform Perseverance: Taming 1,000 Kubernetes Clusters
Talkat Platform Engineering Day 2025 - Europe · on Tuesday, 2025-04-01· at 14:40· Level 1 | Hall Entrance N10 | Room H·
summary
Operates 1,000 upstream Kubernetes clusters with a three‑person platform team, using an in‑house cluster lifecycle CLI and a custom cluster CRD/operator to manage clusters declaratively, Argo CD with GitHub‑based deployments for app delivery, and a homemade managed Kubernetes service across VMware and AWS. Migration from one big cluster to one cluster per product takes two years, upgrades roll out per cluster so the platform can update 1,000 clusters within two days, improving isolation and reliability.
AI-written summary · about
chapters (13)
- 0:00Kubernetes journey at Pik Bank
- 1:24Single bare-metal cluster start
- 3:52Shared cluster challenges
- 5:421,000 dedicated clusters decision
- 7:08Scalability engineering principles
- 11:25Upstream Kubernetes choice
- 12:51KCO for cluster lifecycle
- 14:38Argo CD declarative clusters
- 15:22Migrating applications
- 16:48New platform benefits
- 17:50Lessons from transformation
- 18:33Rolling cluster upgrades
- 19:37Conclusion
related talks
32mDon't Start With 500 Clusters: Patterns in Scaling Multi-Cluster Kubernetes…
03mFrom 5 to 1,300+ Clusters: Declarative Scaling for Private Kubernetes
35mTo Infinite Scale and Beyond: Operating Kubernetes Past the Steady State
27mArgoCD as the Engine of a Brownfield Migration to Kubernetes
26mMulti-Tenant vs Micro-Clusters in Kubernetes-Driven Platform Engineering
47mPostmortem: A Modern Approach to Managing Kubernetes Clusters
23mScaling Kubernetes in Regulated Environments Through Platform APIs
56mManaging from 1 to 1 million Kubernetes Clusters