Github user StefanRRichter commented on the issue:
https://github.com/apache/flink/pull/5934
Hi, can you give some more detail about the actual problem you are trying
to fix here? To me it looks like duplicating the serializer only for the meta
data should not be required, because the serializer is just written and the
getter is only used in a restore, which is never async. You can make an
argument that this is just making the code more defensive, which is a good
thing. But I just want to raise awareness that duplicating a serializer is not
always super cheap, and this counts for the time spend in the synchronous part.
So there is a tradeoff and that is why I would like to discuss if this is
really a benefit?
---
