Background:

When a <downtime> element in the poller configuration has delete="true", the poller signals Provisiond (via an event) to delete any services that have been down long enough to qualify for deletion. The event handler in Provisiond cascades the deletion to delete the service's parent interface and, if the service's parent interface is the last IP interface remaining on the node, to delete the node itself.

If the node has one or more unmanaged IP interfaces, though, the presence of those interfaces precludes the deletion of the node.

It should be possible to exclude unmanaged interfaces when considering whether to delete the node as the result of a downtime action.

It's easy enough to add a flag (optional, defaulting to current behavior) to the poller configuration. Given that Provisiond should not have access to the poller configuration, though, I think the right way to pass the flag's value to Provisiond is via an event parameter. Faint whiff of data envy in the way it shakes out, but I don't see a way around it.