Upgrading Longhorn: 1.9.x to 1.10.x, the One Upgrade You Rehearse a Rollback For First

Overview

Every other bump in the maintenance pass is a version pin. Change one line, reconcile, verify, and roll back by reverting the line. Longhorn 1.10 is different. It drops the v1beta1 API version, so every Longhorn CRD has to store its objects as v1beta2 before the chart that removes the old version lands. A Helm upgrade attempted without that gate fails partway through, and a forced downgrade is the only way out. That difference is the reason this appendix exists. It is also why Longhorn's own upgrade page opens with one rule: back up volumes before upgrading, so a restore is the fallback if anything goes wrong[^1]. The upgrade itself is a one-line chart bump, as in the 1.7 to 1.9 upgrade before it. The work is everything that has to be true first. A backup target exists, a restore from it has been proven, and the cluster's stored versions say the old API is gone.