At the end of a k8s worker upgrade on a fast connection, occasionally the node is replying with an unexpected version when checking the node config. That causes the script to stop processing there and doesn't uncordon the node during the upgrade.
We only ever saw it when ran the script relatively nearby to the datacenter with a higher speed connection. Rather than the typical running of the script, namely from across the Atlantic Ocean or on a mobile connection. The error could have been caused by something other than a fast connection race condition, but that's a working theory. This needs to be resolved before the next tools k8s upgrade.