We can avoid invocations being dropped because of ingress to partition processor connection issues if we auto-attach an idempotency key to all non-idempotent invocations. The trouble is that we would like to apply a finite scatter width to unscoped invocations because otherwise every invocation might land in its own vqueue. The challenge is to figure out whether it is possible to assign idempotency keys as well as imposing a finite scatter width. For that to work, one needs to answer how idempotent invocation are resolved wrt to already completed ones. #4321 is related.
We can avoid invocations being dropped because of ingress to partition processor connection issues if we auto-attach an idempotency key to all non-idempotent invocations. The trouble is that we would like to apply a finite scatter width to unscoped invocations because otherwise every invocation might land in its own vqueue. The challenge is to figure out whether it is possible to assign idempotency keys as well as imposing a finite scatter width. For that to work, one needs to answer how idempotent invocation are resolved wrt to already completed ones. #4321 is related.