Kube Architect

This article explains why common GPU utilization numbers can hide idle compute and how allocation, active time, and useful model work reveal different kinds of GPU waste. More: https://ku.bz/6b4-HyXgv

This case study shows how Albert Heijn built a centralized LGTM-stack observability platform across 1800 engineers, replacing ELK, Nagios, Dynatrace and Azure Monitor setups. More: https://ku.bz/7PG1-ZNH3

This week on Learn Kubernetes Weekly 204: 🧠 Workload-Aware Scheduling in Kubernetes 1.37 ⚙️ Automaxprocs Now Costs Your Go Binary a CPU 🔀 Your gRPC Service Only Talks to One Pod. Here’s Why. 🔐 Your Kubernetes OIDC Issuer Is Just Two Static Files 🧩 Why We Open-Sourced ZSvirt Read it now: https://kube.today/issues/204 ⭐️ This newsletter is brought to you by NeuBird — move beyond dashboards and runbooks with an AI agent that detects production issues, finds their root cause, and helps your team resolve them https://ku.bz/nh5ZFW_b9

GitOps resolves real deployment problems, but it also creates new operational work. Shani Adadi Kazaz explains that a Git repo as the source of truth improves rollbacks and consistency, but teams still need to operate tools like Argo CD or Flux across many clusters. The key point is that GitOps reduces deployment chaos, not platform responsibility. Watch the full interview: https://ku.bz/kThBjp3Lj

WaaS creates browser-accessible Linux and Windows desktops as Kubernetes resources, with GitOps workflows, secure remote access, quotas, and OIDC and RBAC controls. More: https://ku.bz/TgntmDfyC

This article explains how Kubernetes 1.37 gives controllers shared APIs and a Go library for workload-aware scheduling, including gang scheduling and grouped workloads. More: https://ku.bz/t7qQCk1Hz

Why can a Go service be OOM-killed while its heap looks healthy? The heap is only part of the container's memory usage. Goroutine stacks, native allocations, and runtime overhead also count toward the container limit. In this new article from Gulcan, you will learn: - How CPU quotas affect Go's parallelism and why extra threads can increase throttling. - How to measure total container memory, including goroutine stacks and native allocations. - Why a tighter memory budget can increase garbage collection work and reduce throughput. Read: https://learnkube.com/go-kubernetes-requests-limits This article is also included in our book on Kubernetes rightsizing: https://learnkube.com/kubernetes-rightsizing

Join NeuBird and industry leaders at FLOCK ’26 on October 14 in San Francisco. Built for SREs, platform engineers, and reliability leaders exploring autonomous ops. Expect keynotes from former NASA astronaut Mike Massimino and MIT Media Lab’s Ramesh Raskar, hands-on breakouts, and curated seating. 📆 October 14 · SF Register here: https://na2.hubs.ly/H08984m0

Understudy is a Kubernetes operator that starts a temporary replacement for a single-replica workload before a planned node disruption, helping reduce downtime without running two pods all day. More: https://ku.bz/kG1TYjcCT

This case study shows how an AKS-based GitOps platform gave on-premises clusters workload identity by publishing static OIDC discovery and JWKS files, avoiding API server changes after provisioning. More: https://ku.bz/TpvjylBlF

This case study shows how WSC Sports rebuilt their entire deployment model from helm upgrade pipelines to full GitOps using ArgoCD ApplicationSets, covering: - single-chart monorepo design, - shadow deployments, - AppProjects, - CI code freeze gates. More: https://ku.bz/k0MjkJlfX

PigeonEye is a fast native Kubernetes GUI that loads large clusters through the Table API and gives you multi-cluster search, logs, safe edits, debugging tools, and automatic CRD support. More: https://ku.bz/B0087NZrd

MariaDB Operator manages MariaDB on Kubernetes with high-availability clusters, backups, point-in-time recovery, TLS, rolling updates, and declarative SQL resources. More: https://ku.bz/4LmZWWbm-

InfraLens uses eBPF to map live TCP and UDP connections across Kubernetes and Linux servers, identify services, and show real-time traffic without changing application code. More: https://ku.bz/1ZBpgFSz9

Mac Chaffee shares his approach to building internal platforms that expose rather than hide Kubernetes complexity. He argues against creating additional abstraction layers over the Kubernetes API, explaining how this often leads to reinventing the Kubernetes API in less feature-rich ways. Using Docker Compose as an example, Mac demonstrates how simplified abstractions still contain the same core concepts (health checks, security contexts) but in condensed, limited formats. Instead of hiding complexity, he advocates for exposing the Kubernetes API directly to developers while building comprehensive guardrails around it. Watch the full episode: https://ku.bz/9nFPmG85f

Postgres Operator manages external PostgreSQL databases and users via Kubernetes custom resources, automatically creates credentials, and supports AWS RDS, Azure, and GCP Cloud SQL. More: https://ku.bz/CqZGlb7zl

Attune watches real pod usage and adjusts CPU and memory requests without restarting pods, with canary rollouts, safety checks, and automatic rollback. More: https://ku.bz/bqN9cJ24f

Sveltos manages apps and add-ons across multiple Kubernetes clusters from one management cluster using Helm, YAML, Kustomize, Carvel, or Jsonnet, with deployment ordering and drift repair. More: https://ku.bz/s0f7l2QBf

Fabián Sellés Rosa, Tech Lead in the Runtime team @ Adevinta, explains the technical challenges that emerge when platform tenants want to integrate with external managed Kafka services. The core DNS challenge becomes clear: Kafka clients need to connect to external broker infrastructure without knowing the specific connection details. This requires DNS rewriting capabilities, allowing applications to seamlessly access external Kafka services while the platform handles the complexity of routing traffic through private links to the provider's infrastructure. Watch the full episode: https://ku.bz/NsBZ-FwcJ

This week on Learn Kubernetes Weekly 203: 🔥 Building Modelplane on Crossplane 🚪 Kubernetes Gateway API: Why Ingress Is Being Replaced and Which Gateway Controller to Pick 🛡️ From Fragile VMs to Bulletproof GitOps: Modernizing a DevOps Platform on AWS EKS 💾 How a 500 MB Buffer Killed Our Archival Job, and Why Streaming Fixed It 🎮 How GPU MIG + Kueue Can Transform Multi-Tenant AI Workloads on Kubernetes Read it now: https://kube.today/issues/203 ⭐️ This newsletter is brought to you by LearnKube — understand how Kubernetes works, and what to do when it breaks. Live training with 60% hands-on labs.https://ku.bz/hypSbyc-V