Today, there are different options for how to handle the current VM when doing a rolling update (for example, destroy never and destroy last). Instead of supporting these polices, we should always process the current node last.
On a rolling update, the configured flavor's Drain is invoked and then the configured instance plugin's Destroy is invoked. For the current VM we only want to invoke Drain because we do not want to actually destroy the VM that is acting as the Infrakit leader.
Today, there are different options for how to handle the current VM when doing a rolling update (for example, destroy never and destroy last). Instead of supporting these polices, we should always process the current node last.
On a rolling update, the configured flavor's
Drainis invoked and then the configured instance plugin'sDestroyis invoked. For the current VM we only want to invokeDrainbecause we do not want to actually destroy the VM that is acting as the Infrakit leader.