Github user shuai-xu commented on the issue:
https://github.com/apache/flink/pull/5931
@StephanEwen, Good idea, I prefer the first one. As for the second one, the
pending request may have been fulfilled when task executor is killed. so job
master can not cancel the pending request. And when job master failover the job
at the same time with resource manager request a new container, it may ask one
more container than needed.---
