Breaking your own etcd quorum
Reconstruction⚠ This is not a live capture: it is a condensed reconstruction, edited afterwards from the real session transcript. The prompts and the stop messages are the ones from the actual session; the timing is compressed and the long waits are cut. Hostnames sanitized, and some worker nodes are left out of the display — that is called out on screen.
Removing two etcd members from a three-member control plane. The first one leaves without a fuss. The second one, with exactly the same sequence of commands, cuts the cluster API dead — because at two members, there is no majority left to lose.
What makes the recording interesting is not the mistake itself: it is the thirty seconds that follow. The recovery command gets refused by the permission classifier, while the cluster is down. The guardrail does exactly its job — at the worst possible moment.