Github user zhijiangW commented on the issue:
https://github.com/apache/flink/pull/4509
@NicoK , I have fixed the other code issues.
For UT, I only added one test for verifying there is no race condition
between requesting and recycling floating buffers. I am not sure whether to
control the process by `CountDownLatch` is enough. Or I should submit two
different threads to execute the process repeated. If this way is ok, I will
add more tests for race condition along with
`NetworkBufferPool#createBufferPool()` modifications you suggested.
For fair distribution of buffers test, I will add one test, maybe in this
PR or in #4735 later.
---
