Github user zhijiangW commented on the issue:
https://github.com/apache/flink/pull/4594
This improvement is indeed very clean and reduce many redundant `Override`
methods by extending `PooledByteBufAllocator` directly.
But I am still confused of one thing. In previous way, we created the
`PooledByteBufAllocator` with `preferDirect` parameter as true, and the created
allocator will be used in client and server. So it seems no chance to allow the
heap buffer in original allocator in netty code. Or I missed or mis-understood
some key informations?
---
