Github user shuai-xu commented on the issue: https://github.com/apache/flink/pull/5931 @GJL In blink, we solve this problem like this. When a container complete, we will first whether the container has registered yet, if it has registered before, RM will not request a new container, as the job master will ask for it when failover. If it has not registered, RM will request a new one.
---