Github user devriesb commented on the issue:
https://github.com/apache/nifi/pull/2416
So, I was under the impression this was still a WIP. I am a HUGE -1 on
this change. As @markap14 stated above, this is a critical section of code.
And while the previous version has serious flaws, they are at least somewhat
know based on a long period of use. In other words, we know there are
problems, but they only bite us every so often.
This major rewrite of a critical piece being essentially forced on users,
likely without their knowledge unless they are paying close attention, seems
less than ideal. This new implementation will need SIGNIFICANT testing before
it can be trusted to the same degree as the previous, even with its issues.
I would have greatly preferred, and will still advocate for, making this a
new implementation vs. a change to the WriteAheadFlowFileRepository (e.g.
SequentialWriteAheadFlowFileRepository). Again, there are other issues, but
avoiding repo corruption in the previous WriteAheadFlowFileRepository would
have been a reasonably simple fix, not requiring this rewrite. While the
rewrite may have other benefits, making it a new implementation (even if you
were to make it the default...) would give users the opportunity to evaluate
and decide for themselves when they are ready to move to the new repo, without
forcing them to postpone an upgrade to 1.6.0 which has other worthwhile changes.
I know critical sections are change all the time, and that users won't
always be aware of the changes. However, the criticality of this section
combined with it's history means I think we should tread a little more lightly.
---
