LearnKube news

This tutorial shows how to design ASP.NET Core health checks for Kubernetes using separate liveness, readiness, and startup probes. More: https://ku.bz/Pxw14_JCD

Kubetail is a Kubernetes logging tool that streams workload logs into a browser or terminal, merges multi-container logs into one timeline, and works without sending logs to an external service. More: https://ku.bz/-c_dwmWJp

Kube Startup CPU Boost is a tool that increases CPU resource requests and limits during Kubernetes workload startup time and then returns them to their original values once the workload is up and running. More: https://ku.bz/TqDtnzFYK

Andrew Charlton, Staff Software Engineer at Timescale, shares specific operational improvements achieved by replacing StatefulSets with their custom Patroni-based operator called Popper. He explains how they successfully consolidated from five availability zones to two or three, improving node packing efficiency without disruption. Andrew details their innovative approach to minimizing downtime during instance resizing by using "scout pods" with low priority classes to verify placement before restarting database pods. He also describes how they implemented lazy node provisioning during Kubernetes upgrades and seamlessly migrated from EXT4 to XFS file systems for better PostgreSQL performance - operations that would have been impossible with standard StatefulSets. Watch the full episode: https://kube.fmhttps://ku.bz/fhZ_pNXM3

This article explains how to optimize AI agents for Kubernetes diagnostics by shifting from sequential Model Context Protocol tool calls to code execution mode, reducing token usage by up to 90%. More: https://ku.bz/hYKhvM28f

git-change-operator is a Kubernetes operator that enables automated Git operations from within clusters through GitCommit and PullRequest custom resources. More: https://ku.bz/Y1q8PFFvw

John McBride, VP of Infrastructure and AI Engineering at The Linux Foundation, shares his "hot take" on how Kubernetes might evolve to handle AI workloads over the next decade. He predicts that another player like Nvidia might enter the space with a purpose-built compute platform, addressing fundamental mismatches between Kubernetes' container paradigm and the requirements of large language models. John highlights specific technical challenges: "Downloading images onto a cluster and getting nodes to handle 10-15 gigabyte workloads is just not a good Kubernetes paradigm." He critiques Kubernetes' historically slow adaptation to GPU workloads, noting that "GPU drivers and GPU workloads on Kubernetes have been pretty painful for basically the whole inception of Kubernetes." Watch the full episode: https://kube.fmhttps://ku.bz/wP6bTlrFs

This week on Learn Kubernetes Weekly 190: 🌪️ Taming the Storm: Building Groww's Internal Chaos Engineering Platform 🧪 Taracode Testing a Go-Based CLI AI Agent in My Homelab 🧠 Building self-evolving AI systems: exploring the architecture 🔄 Migrating from slurm to Kubernetes 🏠 Lessons Learnt Self-hosting an AI Assistant Read it now: https://kube.today/issues/190 ⭐️ This newsletter is brought to you by LearnKube — master Kubernetes with hands-on training designed for engineers who want to learn the smart way https://ku.bz/hypSbyc-V

Mike Stefaniak, Head of Product, Kubernetes and Registries at Amazon Web Services (AWS), shares his vision for Kubernetes over the next decade. Mike explains how clusters should become an implementation detail that end users never need to concern themselves with. He describes EKS's strategic direction toward a model where customers simply bring their applications, and AWS handles all the underlying cluster orchestration and management automatically. Watch the full interview: https://ku.bz/PzjrglcZJ

This article teaches how to build self-evolving AI systems using Kubernetes, Temporal workflows, and automated deployment pipelines, enabling AI agents to detect errors, fix code, and redeploy services without manual intervention. More: https://ku.bz/CqL64wJNt

New on LearnKube: Server-side apply: what happens when you run kubectl apply Server-side apply changes how Kubernetes handles field ownership. Kubernetes objects are shared state. Manifests, controllers, release tools, autoscalers, webhooks, and operators can all shape the same object. With client-side apply, stale intent can overwrite live changes. With server-side apply, ownership moves into the API server. Kubernetes tracks which manager owns each field and surfaces conflicts when ownership is contested. Chiara put serious work into this guide, and you can read it in full here: https://learnkube.com/server-side-apply-kubernetes

Luxury Yacht is a cross-platform desktop app for managing Kubernetes clusters, available for Linux, macOS, and Windows, built with Go and Wails. More: https://ku.bz/fxHb6Vcsf

CPUs have Linux schedulers. GPUs don't. Andrew Hillier explains the fundamental difference: with CPUs, you can run up utilization by scheduling more jobs — if things get busy, you get throttled. With GPUs, there's no scheduler balancing loads. You can't easily increase demand; you have to reduce supply through partitioning. Memory is even more rigid. On CPUs, you can sacrifice buffering for density. On GPUs, your memory is your memory — use less or get OOM killed. Watch the full interview: https://ku.bz/wL-0d1X0y

This article tests Taracode, a Go-based CLI AI agent, against real K3s homelab tasks like Kubernetes troubleshooting, manifest generation, and GitOps workflows using a local Ollama LLM. More: https://ku.bz/4wr03JdLg

"The sidecar era is changing, and Kubernetes networking is being simplified." Abhishek Rao explains why Istio Ambient Mode, Gateway API, and eBPF are the three technologies he is watching most closely. He connects each one to practical platform decisions teams are making right now. The takeaway: pick tools that reduce friction while improving security and visibility. Watch the full interview: https://ku.bz/_q9XBgY2c

Kroc is an educational Kubernetes Operator built with Go and kubebuilder that watches arbitrary resources and reactively creates derived objects using Go templating. More: https://ku.bz/zfk42gx0M

Mac Chaffee explains why teams repeatedly reject established tools like Kubernetes only to rebuild them from scratch. He identifies this as a mentorship crisis in software engineering, where the industry lacks proper apprenticeship programs found in other fields. Mac argues that without mentors to guide understanding, engineers view complex tools as unnecessarily complicated and believe they can build simpler alternatives. While reimplementing systems can be valuable for learning (like rebuilding a database in a college course with proper guidance), attempting this in production environments without mentorship leads to expensive mistakes. Watch the full episode: https://ku.bz/9nFPmG85f

What Kubernetes tools should you keep an eye on? Mauro Morales from Spectro Cloud shares three picks: - Spegel for peer-to-peer image distribution across nodes (less bandwidth, lower costs) - Open Cluster Management for handling multi-cloud and hybrid setups with a proven framework - AI-assisted tooling like K8sGPT and kubectl AI for diagnosing cluster issues. The common thread: tools that reduce complexity without adding operational overhead. Watch the full interview: https://ku.bz/8cpgjFfjn

This case study explains how Groww built an internal chaos engineering platform on Kubernetes to run controlled failure drills like network faults, dependency outages, and traffic replay before real incidents hit production. More: https://ku.bz/vT3jsf-Yh

Most Kubernetes tools start as niche experiments — but a few are worth watching early. Abby Bangser highlights three emerging projects: OpenFeature for standardizing feature flags (building on what OpenTelemetry did for observability), and CRDify for building better Kubernetes APIs without deep CRD expertise. Watch the full interview: https://ku.bz/5Rqq275dl