During a reconfiguration, we may cancel builds if they are no longer
necessary, however we do not do the same for node requests. Currently
we let them run to completion and then return them unused. It would
be more efficient (and behaviorally consistent) to cancel the request
as soon as we determine that it won't be used.
Also, change some warning log messages to info (as they don't indicate
a situation which may benefit from corrective action).