Github user NicoK commented on the issue:
https://github.com/apache/flink/pull/4529
I haven't had time to test this yet - I doubt that this is simply solved by
using our memory segments but I need to understand the problem better.
Maybe this also goes away with [flow
control](https://issues.apache.org/jira/browse/FLINK-7282) being added since
then we have more control over what the sender sends and can make sure that we
have enough buffers to process its data.---
