1 views collected around this technical thread.
This article shares JD.com’s extensive experience redesigning Kubernetes for massive production use, covering custom DNS and load‑balancing, scaling clusters to ten‑thousand nodes, adapting controllers, building the Archimedes scheduler, and practical insights on resource isolation, deployment, and high‑traffic elasticity.
The article describes JD’s experience of redesigning Kubernetes at massive scale, detailing the JDOS2.0 platform, custom DNS and load‑balancing, the Archimedes scheduler, API and controller optimizations, and operational lessons learned from running tens of thousands of nodes in production.
The article presents ContainerLB, a software‑defined L4 load‑balancing service built on Intel DPDK and X86 servers for JD.com, detailing its architecture, FULLNAT mode, high‑availability design, performance optimizations, and test results that demonstrate near‑line‑rate packet processing and scalable deployment in a container‑native data‑center.
Since the end of 2016, JD has been transitioning its infrastructure from OpenStack to Kubernetes, completing 20% of the migration and aiming for full conversion by Q2, and shares detailed experiences, architectural evolution, operational practices, and future directions for large‑scale container platforms.