Github user fmthoma commented on the issue:
https://github.com/apache/flink/pull/6021
@tzulitai I agree on adding additional docs, where do you suggest I should
put them? In the Javadoc on `setQueueLimit()`?
My current suggestion is to look at the size of your individual records,
and choose the queue limit so that about 10MB per shard are aggregated. 1MB
would be too small (since the KPL aggregates the user records to 1MB batches).
But I'll run some more performance tests, in particular also with the
`wait()`/`notify()` change you suggested above.
---
