Kubernetes

Kubernetes (often abbreviated K8s) is an open-source system for automating the deployment, scaling and management of containerised applications. Originally designed by Google and released publicly in 2014, it has become the de facto standard for running software across clusters of machines — a role that now extends to serving machine-learning models and running AI infrastructure.^[1][3]

History

Kubernetes was announced by Google on 6 June 2014, conceived by engineers Joe Beda, Brendan Burns and Craig McLuckie after a 2013 pitch to open-source an internal container management system. The design drew directly on Google's experience with Borg and Omega, the cluster managers that run Search, Gmail and YouTube, and the name is Greek for "helmsman" or "pilot".^[2][4]

Version 1.0 was released on 21 July 2015, and Google offered Kubernetes as the seed technology of the Cloud Native Computing Foundation (CNCF) under the Linux Foundation, retaining the trademark while development passed to a global community of contributors.^[3] By 2017 the project had outpaced rivals such as Docker Swarm and Apache Mesos to become the industry standard. A CNCF journey report cited by IBM describes it as the second-largest open-source project in the world after Linux and the primary container orchestration tool for 71% of Fortune 100 companies.^[5]

Key Concepts

Kubernetes works declaratively: operators describe the desired state of an application in configuration files (usually YAML), and a set of independent control loops continuously reconciles the cluster toward that state rather than executing a fixed sequence of steps.^[1] Core abstractions include:

• Pods — the smallest deployable unit, wrapping one or more containers that share storage and network.
• Nodes — the machines (virtual or physical) that run pods, managed by a control plane that schedules work and stores cluster state.
• Self-healing — failed containers are restarted, and nodes or pods that fail health checks are replaced or rescheduled automatically.
• Scaling — replicas can be added or removed horizontally in response to load, either manually or through the Horizontal Pod Autoscaler.
• Services — stable network endpoints that route traffic to changing sets of pods.

Because it is portable across clouds and on-premises hardware, most organisations consume Kubernetes through managed services such as Google Kubernetes Engine (GKE), Amazon Elastic Kubernetes Service (EKS) and Azure Kubernetes Service (AKS).^[4]

Applications

Kubernetes underpins cloud-native software delivery: microservices platforms, API gateways, streaming pipelines and web applications. Its relevance to AI is direct. GPU and accelerator resources can be exposed to the cluster through device plugins, allowing schedulers to place training or inference jobs on nodes with the right hardware; projects such as Kubeflow add machine-learning workflows on top; and model-serving platforms use Kubernetes to scale inference endpoints up and down as traffic changes.^[6] In practice, most commercial AI products — chatbots, image generators, recommendation services — ultimately run as containers on a Kubernetes cluster somewhere.^[1]

>See Also

• GPU Cluster
• MLOps
• Inference
• Distributed Training
• AI Data Centres in Malaysia
• Docker