https://ivomarino.com/post/Recent content in Posts on ivomarino.comHugo -- gohugo.ioen-USAll rights reserved - 2026Sun, 18 Oct 2026 00:00:00 +0000https://ivomarino.com/post/ai-workflow-build-log-1/Sun, 18 Oct 2026 00:00:00 +0000https://ivomarino.com/post/ai-workflow-build-log-1/A new series where I build AI workflows end to end and write down how. Who it is for, the problem it solves, its goals, and five checks that sit around the model.https://ivomarino.com/post/gateway-failover-ssh-tunnels/Thu, 23 Apr 2026 00:00:00 +0000https://ivomarino.com/post/gateway-failover-ssh-tunnels/Building reliable automatic gateway failover for critical infrastructure recovery access. How naive health checks broke everything, and why a 30-second stable recovery is better than a 5-second flapping failover.https://ivomarino.com/post/keda-queue-autoscaling/Tue, 31 Mar 2026 00:00:00 +0000https://ivomarino.com/post/keda-queue-autoscaling/Kubernetes HPA with CPU metrics fails for queue-based workloads. Learn why queue depth is the right signal, how KEDA makes event-driven autoscaling trivial, and the configuration that reduced processing latency from 45 minutes to 5-15 minutes with zero manual scaling.https://ivomarino.com/post/production-security-updates/Sat, 28 Mar 2026 00:00:00 +0000https://ivomarino.com/post/production-security-updates/Security patches can’t wait, but unplanned reboots break production. The update runbook for Debian, FreeBSD, and OpenBSD — with real scripts, hold patterns, and a zero-downtime cluster update sequence.https://ivomarino.com/post/mlops-guide/Fri, 27 Mar 2026 00:00:00 +0000https://ivomarino.com/post/mlops-guide/Production machine learning requires more than training good models. This guide covers building a complete MLOps system with Kubernetes, Prometheus monitoring, and automated accuracy checks to catch model drift and prevent silent failures in production.https://ivomarino.com/post/zfs-nfs-kubernetes-storage/Thu, 26 Mar 2026 00:00:00 +0000https://ivomarino.com/post/zfs-nfs-kubernetes-storage/Kubernetes storage is either ephemeral or expensive. We run a FreeBSD VM with ZFS + NFS in the same VPC as our K8s cluster. Hourly snapshots protect your data. Fast recovery. No vendor lock-in. Here’s how we do it.https://ivomarino.com/post/openbsd-carp-k8s-firewall/Wed, 25 Mar 2026 00:00:00 +0000https://ivomarino.com/post/openbsd-carp-k8s-firewall/We run two OpenBSD firewalls in CARP HA mode in front of a private Kubernetes cluster. Sub-second failover, full state sync, and pf rules that load-balance across worker nodes.https://ivomarino.com/post/mysql-scaling-beyond-k8s/Mon, 23 Mar 2026 00:00:00 +0000https://ivomarino.com/post/mysql-scaling-beyond-k8s/We ran MySQL in Kubernetes for years until the data scaled from GB to TB. Kubernetes wasn’t the problem—backups were. Kasten (the Kubernetes backup tool) was never designed for databases at our scale. Learn why we moved MySQL off Kubernetes and the backup solution we built to solve this infrastructure challenge.https://ivomarino.com/post/relaunch/Thu, 01 Jan 2026 00:00:00 +0000https://ivomarino.com/post/relaunch/Relaunching with a focus on pragmatic infrastructure: backup strategies, Kubernetes autoscaling, and lessons from 10+ years of production operations.