Github user devriesb commented on the issue: https://github.com/apache/nifi/pull/2416 @markap14 yes... under this somewhat unusual circumstance, my proposed solution would sacrifice data for consistency. However, if alwaysSync is set false, there isn't a guarantee of no loss anyway. So, we'd really be no worse off than we are, except that the plug getting kicked out of the wall wouldn't kill your repo. Also, I fully agree that an implementation that DOES guarantee no loss of CREATEs would be for the best. My proposal above doesn't address that in any way. It lives inside the "all or nothing" guarantee of the current (/previous?) WriteAheadFFR.
---