Github user devriesb commented on the issue:
https://github.com/apache/nifi/pull/2416
Again, I would submit that it isn't violating the data loss guarantees of
the original implementation. Yes, some data could potentially be lost, but
none that it guaranteed to keep, and in doing so it prevents corruption... in a
handful of easily understandable lines, that don't fundamentally alter the
implementation.
I want to be clear... I'm in no way opposed to improvements. However, for
operational data flows there is going to be some trepidation about large
changes to such a key component... especially one that we've had previous
issues with. The smaller / more incremental the changes, and the more time
allowed for evaluation under load, the more comfortable I would imagine people
would be.
---
