Github user fmthoma commented on the issue:
https://github.com/apache/flink/pull/6021
@tzulitai @bowenli86 I've made some more changes while investigating
awslabs/amazon-kinesis-producer#183:
* I've followed your suggestion and used `wait()` instead of
`Thread.sleep()`, see `TimeoutLatch`. This allows much smaller queue sizes.
* The timeout of 500ms is too high, i've lowered it to 100ms.
* I've added two metrics: A `Gauge` for the `outstandingRecordsCount`, and
a `Counter` for the backpressure cycles (i.e. the number of times the check
`outstandingRecordsCount <= queueLimit` fails).
I updated the documentation, the recommended queue limit is now much lower.
---
