Problem in general
- We have a write consensus of n-f, and consensus for catch-up of f+1.
- So, if more than n-f-(f+1)=n-2f-1 nodes agreed to order a txn but failed/crashed before writing it to the Ledger, then we have a stalled/broken Ledger.
A possible solution is to persist COMMITs, so that we can recover after start-up all txns that were agreed to be ordered, but not ordered yet for some reasons.
INDY-467 for more details.