A scheduler for things that must step out of service to be replenished — a
battery that recharges, an aircraft pulled for maintenance, a vehicle-mounted Pi
that stops computing to harvest — and then return to work.
It treats the whole repeating schedule as one closed route: every stint of service
and every harvest is an edge in a graph, and a valid schedule is a walk that uses
each edge once and ends where it began. Hierholzer's algorithm (1873) builds that
walk in time proportional to the number of activities. Because every device that
leaves service is obliged to return, the structure guarantees nothing is stranded
— unless the energy budget makes it impossible.
Target energy postures are kept as vectors; the planner shapes the schedule so the
fleet's aggregate health settles as close as it can to a chosen posture, for the
least harvest cost. For OpenSky that means: given a panel wattage and scarce,
concurrency-limited harvest windows, how much of the fleet stays up — and what
that does to the business as it scales to a million devices.