Github user srdo commented on the issue:
https://github.com/apache/storm/pull/2156
@WolfeeTJ Thanks for trying this out. I think you are right that it's
because partition reassignment is happening at a bad time. I think we should
move reassignment to be the first thing in nextTuple instead. Give me a little
bit to fix this and maybe add a test.
I've been thinking a bit about partition revocation and shuffling, and we
might want to add some warnings to the manual partition assignment API as well.
There are a few cases I can think of where shuffling partitions can cause bugs.
The Trident spout doesn't support partition shuffling because Trident doesn't
expect partitions to move from task to task, as far as I can tell. When we
implement at-most-once support for this spout there's also a requirement that
partitions don't move between tasks, since otherwise it is possible that tuples
are emitted more than once.
---
