Back in 1996 I went back to school to learn the latest in-demand skills. Best decision of my career. Internet access wasn’t a given yet, so after commuting home from class I’d get on my own computer and practice what I’d just learned. Bought a new machine, installed Windows Server, got to work. Today people call that a homelab. They’re popular for a reason: they work. Ask anyone in IT about their best lessons, and a lot of them will tell you about something they broke at home and had to fix themselves. ...
TIL: ReadWriteOnce doesn’t mean one pod. It means one node. I was working through PersistentVolumes and PVCs today, expecting RWO to lock a volume to a single pod. It doesn’t. RWO restricts the volume to a single node, and Kubernetes has no problem letting multiple pods on that same node mount it. In a small cluster or test environment this stays invisible, since pods for the same workload often land on the same node anyway. The gotcha only shows up when the scheduler moves a pod to a different node (drain, scale event, node pressure) and the volume can’t follow because block storage attaches to exactly one node at a time. That pod sits in ContainerCreating waiting on a mount that isn’t coming. ...