Why we rehearse cordons before upgrades
Cordons feel simple until messaging, quotas, and StatefulSets disagree. Here is how we stage the conversation.
We open many cohorts with a quiet question: when was the last time your team practiced a cordon without touching production? The answers vary, but the pattern repeats—teams know the imperative command yet lack a shared vocabulary for what should happen to workloads that tolerate disruption differently.
In our Operations Under Load week we script cordons so participants hear the same kubelet phrases they will see at work, only against disposable Deployments and a handful of StatefulSets with relaxed storage classes. The goal is not speed; it is legible narration while nodes drain.
By afternoon, rotating incident leads practice reading node conditions aloud while a partner annotates a timeline. The debrief focuses on what language confused application owners and which metrics you would show a quality standards reviewer later. We leave theory at the door; the cluster is the text.