If the “reporting” layer is essentially micro-batching over bounded backlogs, run it with availableNow (or a scheduled batch job) so each run is naturally bounded and exits cleanly on its own, no manual cancel. This greatly reduces chances of partial micro-batches during redeploys.