Github user markap14 commented on the issue:
https://github.com/apache/nifi/pull/2416
@devriesb I believe the issue with this proposed solution is that if we
have 256 partitions (the default), for instance, then if one partition is not
flushed to disk during a power failure, we would end up losing all data written
to the other 255 partitions. Even if we were to use an fsync() for each CREATE
event, that data would be lost with the proposed solution. So this could lead
to quite a bit of data loss even with fsync enabled.
---
