My homelab now has more dashboards than services worth monitoring. Prometheus scrapes the lot: node exporter on every box, cAdvisor for the containers, smartctl for the disks, a Pi-hole exporter, an exporter for the UPS, and an exporter I genuinely cannot remember installing. Grafana renders all of it across about a dozen panels of beautiful, mostly ignored graphs.
The honest truth is I look at one row. CPU, memory, disk free, and the "is anything actually down" panel. The rest exists because building a dashboard is more fun than the boring work of deciding what should alert me. A graph feels like progress. An alert rule that pages me at 3am feels like a commitment.
So this weekend's job is subtraction. I'm wiring up Alertmanager properly so that disk-nearly-full and host-down actually reach my phone, and I'm letting the pretty graphs become the thing you consult after the alert fires, not the thing you stare at hoping to catch a problem in real time. Nobody catches a failing drive by watching a graph. You catch it because something told you to look.
The dashboards can stay. They're nice. They're just not monitoring, they're decoration with a time axis.