Thanks, that clarifies it. The checkpoint-based cancellation and the sync-vs-async locking model differences were exactly what I was trying to understand.