Abstract
This document establishes public, dated prior art over a deployment-orchestration mechanism that eliminates the ReadWriteOnce (RWO) "Multi-Attach" deadlock during a Recreate-strategy rollout of a Kubernetes workload that mounts a single-attach persistent volume. At deploy time, the mechanism discovers the volume-holder node from the running pod's spec.nodeName, then cordons every other node in the workload's scheduling pool for the rollout window only, so the replacement pod can schedule onto exactly one node — the holder — making Multi-Attach structurally impossible rather than merely unlikely. Bookkeeping is ownership-aware: only nodes this run cordoned are recorded in a ledger; nodes already cordoned by others are skipped and left untouched; and an exit trap reverses only self-cordoned nodes, firing on success, real failure, rollback, timeout, or operator interrupt. The fence is deliberately held through failure rollback because the rollback pod mounts the same volume, and is safely released after set-image even on a timeout because a pod, once scheduled, is bound to its node and cannot migrate. This disclosure provides an enabling description, four figures, a data model, a worked example, a prior-art delta, one independent claim with sixteen dependent claims, and a clean-room, dependency-free, offline-runnable Node.js reference implementation. It is published so that the described mechanism and its obvious variants remain free for public practice and cannot be exclusively appropriated by a later patent.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Assuncao, gustavo matthew, "Transient Cordon-Fence Pinning of Recreate Rollouts to the ReadWriteOnce-Volume-Holder Node", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/10952