Tencent Cloud EKS: The Elegant Balance Between Serverless and Serverful

The author shares insights on the evolution from Serverful to Serverless computing, presenting Tencent Cloud's Elastic Container Service (EKS) as a progressive Serverless solution that balances operational efficiency with flexibility.

The discussion begins by exploring the fundamental nature of cloud computing, drawing an analogy to the second industrial revolution where each company needed to build its own power infrastructure. The author contrasts AWS's IaaS approach (treating cloud as utilities like electricity) with Google's PaaS approach (providing factories for production).

By 2014, Google launched GKE while AWS introduced Lambda, representing two different paths: Cloud Native (open, standard, multi-cloud) versus Serverless (极致、专有、个性化). The author notes that while Kubernetes gained broader impact due to its open-source nature and CNCF support, Serverless and Cloud Native share the same ultimate goal of maximizing business value.

The technical comparison reveals a trade-off: as abstraction levels increase from Serverful VMs to Serverless functions, operational burden decreases and elasticity improves, but flexibility and migration cost become challenges. Kubernetes offers strong generality and low migration costs but requires significant node maintenance, while FaaS provides zero-ops but with high migration complexity.

EKS addresses these challenges through a nodeless architecture that replaces traditional cloud host nodes with virtual nodes. This virtual node presents itself to Kubernetes as a machine with infinite CPU and memory, enabling Kubernetes to schedule containers to Tencent Cloud's massive resource pool. The solution maintains full Kubernetes API compatibility while eliminating node operational burden.

Key use cases include: 1) Serverless transformation of existing K8s clusters with one-click deployment, 2) Hybrid cloud scenarios integrating IDC K8s clusters, and 3) Progressive big data containerization allowing dynamic workload distribution between traditional clusters and EKS nodes.

Future plans include transitioning from container instance-based billing to actual CPU/memory usage-based billing, and aligning elastic scaling metrics with business requirements.