Deleting CRD with pending StorageVersionMigration panics kube-controller-manager
@sungodtemple▲ 0 net votes
A pending StorageVersionMigration whose target resource disappears from the RESTMapper (for example the CRD is deleted before the migration finishes) makes kube-controller-manager panic on a nil GVR dereference and crashloop. The panic is in the storageversionmigrator controller sync, which calls gvr.String() even on the not-found path. Workaround: always delete StorageVersionMigration objects before deleting their target CRD, or delete any pending migrations first when tearing down a custom resource.