KubeFM

Running GPU workloads on Kubernetes sounds straightforward until you need to isolate multiple tenants on the same server. The moment you virtualize GPUs for security, you lose access to NVIDIA kernel drivers — and almost every tool in the ecosystem assumes those drivers exist.Landon Clipp built a GPU-based Containers as a Service platform from scratch, solving each isolation layer — from kernel separation with Kata Containers + QEMU to NVLink fabric partitioning to network policies with Cilium/eBPF — and shares exactly what broke along the way.In this interview:Why standard NVIDIA tooling (GPU Operator) fails in multi-tenant setups, and how to use CDI with PCI topology scanning to make GPUs visible to Kubernetes without kernel driversHow to partition the NVLink fabric between tenants using a trusted service VM running Fabric Manager, and why the physical PCIe wiring differs between Supermicro HGX and NVIDIA DGX systemsWhy gVisor doesn't work for GPU workloads — NVIDIA's unstable ioctl ABI means Google has to update gVisor for every driver release, and they only support a handful of GPUsWhat caused 8-GPU VMs to take 30+ minutes to boot, and the specific fixes (IOMMUFD, cold plugging, kernel upgrades) that brought it down to minutesHow Cilium network policies enforce tenant isolation at the Kubernetes identity level instead of fragile IP-based rulesWhere Containers as a Service fits best: inference workloads where AI teams want to ship an OCI image without managing infrastructure or signing multi-million dollar cluster contracts.SponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/jjK_yJTDzInterested in sponsoring an episode? Learn more.

Build failures in Kubernetes CI/CD pipelines are a silent productivity killer. Developers spend 45+ minutes scrolling through cryptic logs, often just hitting rerun and hoping for the best.Ron Matsliah, DevOps engineer at Next Insurance, built an AI-powered assistant that cut build debugging time by 75% — not as a dashboard, but delivered directly in Slack where developers already work.In this episode:Why combining deterministic rules with AI produces better results than letting an LLM guess aloneHow correlating Kubernetes events with build logs catches spot instance terminations that produce misleading errorsWhy integrating into existing workflows and building feedback loops from day one drove adoptionThe prompt engineering lessons learned from testing with real production data instead of synthetic examplesThe takeaway: simple rules plus rich context consistently outperform complex AI queries on their own.SponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/PDdYfC00wInterested in sponsoring an episode? Learn more.

Managed Kubernetes on a major cloud provider can cost hundreds or even thousands of dollars a month — and much of that spending hides behind defaults, minimum resource ratios, and auxiliary services you didn't ask for.Fernando Duran, founder of SadServers, shares how his GKE Autopilot proof of concept ran close to $1,000/month on a fraction of the CPU of the actual workload and how he cut that to roughly $30/month by moving to Hetzner with Edka as a managed control plane.In this interview:Why Kubernetes hasn't delivered on its original promise of cost savings through bin packing — and what it actually provides insteadA real cost comparison: $1,000/month on GKE vs. $30/month on Hetzner with Edka for the same nominal capacityWhat you need to bring with you (observability, logging, dashboards) when leaving a fully managed cloud providerThe decision comes down to how tightly coupled you are to cloud-specific services and whether your team can spare the cycles to manage the gaps.SponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/6nSDbz9m4Interested in sponsoring an episode? Learn more.

Running multiple Kubernetes clusters on AWS with the cluster autoscaler? Every four months, you face the same grind: upgrading Kubernetes versions, recreating auto scaling groups, and hoping instance type changes stick.Adhi Sutandi, DevOps Engineer at Beekeeper by LumApps, shares how his team migrated from the cluster autoscaler to Karpenter across eight EKS clusters — and the hard lessons they learned along the way.In this episode:Why AWS auto scaling groups are immutable and how that creates upgrade bottlenecks at scaleHow the latest AMI tag accidentally turned less critical clusters into chaos engineering environments, dropping SLOs before anyone realized Karpenter was the causeWhy pre-stop sleep hooks solved pod restartability problems that Quarkus's built-in graceful shutdown couldn'tThe case for pod disruption budgets over Karpenter annotations when protecting critical workloads during node rotationsHow Karpenter's implicit 10% disruption budget caught the team off guard — and the explicit configuration that fixed itSponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/XyVfsSQPrInterested in sponsoring an episode? Learn more.

Migrating from ECS to Kubernetes sounds straightforward — until you hit spot capacity failures, firewall rules silently dropping traffic, and memory metrics that lie to your autoscaler.Radosław Miernik, Head of Engineering at aleno, walks through a real production migration: what broke, what they missed, and the fixes that made it work.In this interview:Running Flux and Argo CD together — Flux for the infra team, Argo CD's UI for developers who don't want to touch YAMLHow the wrong memory metric caused OOM errors, and why switching to jemalloc cut memory usage by 20%Splitting WebSocket and API containers into separate deployments with independent autoscalingFour months of migration, over 100 configuration changes in the first month, and a concrete breakdown of what platform work looks like when you can't afford downtime.SponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/x6wFMhVsxInterested in sponsoring an episode? Learn more.

Kubernetes nodes on EKS can take over a minute to become ready, and pods often wait even longer — but most teams never look into why.Jan Ludvik, Senior Staff Reliability Engineer at Outreach, shares how he cut node startup from 65 to 45 seconds and reduced P90 pod startup by 30 seconds across ~1,000 nodes — by tackling overlooked defaults and EBS bottlenecks.In this episode:Why Kubelet's serial image pull default quietly blocks pod startup, and how parallel pulls fix itHow EBS lazy loading can silently negate image caching in AMIs — and the critical path workaroundA Lambda-based automation that temporarily boosts EBS throughput during startup, then reverts to save costThe kubelet metrics and logs that expose pod and node startup latenc,y most teams never monitorEvery second saved translates to faster scaling, lower AWS bills, and better end-user experience.SponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/B7TzKXyxfInterested in sponsoring an episode? Learn more.

You self-host services at home, but upgrades break things, rollbacks require SSH-ing in to kill containers manually, and there's no safety net if your hardware fails.Thibault Martin, Director of Program Development at the Matrix Foundation, walked this exact path — from Docker Compose to Podman with Ansible to Kubernetes on a single server — and explains why each transition happened and what it solved.In this interview:Why Ansible's declarative promise fell short with the Podman collection, forcing sequential imperative steps instead of desired-state definitionsHow community Helm charts replace the need to write and maintain every manifest yourselfWhy GitOps isn't just a deployment workflow — it's a disaster recovery strategy when your infrastructure lives in your living roomHow k3s removes the barrier to entry by bundling opinionated defaults so you can skip choosing CNI plugins and storage providersKubernetes doesn't have to be enterprise-scale — with the right distribution and community tooling, it can be a practical, low-overhead choice for anyone who cares about their data.SponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/Xk5S7VqXzInterested in sponsoring an episode? Learn more.

Your database backup strategy shouldn't be the thing that takes your production systems down.Ziv Yatzik manages 600+ Postgres clusters in a closed network environment with no public cloud. After existing backup solutions proved unreliable — causing downtime when disks filled up — his team built a new architecture using pgBackRest, Argo CD, and Kubernetes CronJobs.In this episode:Why storing WAL files on shared NAS storage prevents backup failures from cascading into database outagesHow GitOps with Argo CD lets them manage backups for hundreds of clusters by adding a single YAML fileThe Ansible + Kubernetes hybrid approach that keeps VM-based Patroni clusters in sync with Kubernetes-orchestrated backupsA practical blueprint for making database backups boring, reliable, and safe.SponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/Rg_sQYSmwInterested in sponsoring an episode? Learn more.

Varnit Goyal, a Senior Software Engineer who experiments with cloud-native tooling and home labs. He describes building a full three-node Kubernetes cluster for $2.16/month using Rackspace Spot Instances. He covers bidding and instance-choice strategies, multi-region workers and automated replacement, replacing paid load balancers with Tailscale Funnel, and running real workloads like Kafka and Postgres on a budget.

Dilshan Wijesooriya, Senior Cloud Engineer, discusses a real incident where migrating EKS nodes to AL2023 caused the cluster autoscaler to lose AWS permissions silently.You will learn:Why AL2023 blocks pod access to instance metadata by default, breaking components that relied on node IAM roles (like cluster autoscaler, external-DNS, and AWS Load Balancer Controller)How to implement IRSA correctly by configuring IAM roles, Kubernetes service accounts, and OIDC trust relationships, and why both AWS IAM and Kubernetes RBAC must be configured independentlyThe recommended migration strategy: move critical system components to IRSA before changing AMIs, test aggressively in non-production, and decouple identity changes from OS upgradesHow to audit which pods currently rely on node roles and clean up legacy IAM permissions to reduce attack surface after migrationSponsorThis episode is sponsored by LearnKube — get started on your Kubernetes journey through comprehensive online, in-person or remote training.More infoFind all the links and info for this episode here: https://ku.bz/T_YPfTfDbInterested in sponsoring an episode? Learn more.