Attempt(map or reduce) remains NEW(state). And job is stuck in certain conditions.

Attempt(map or reduce) remains NEW(state). And job is stuck in certain conditions.

The following are the situations:

1. total task(map/reduce) count is same as the running limit of task(mapreduce.job.running.map.limit/mapreduce.job.running.reduce.limit).
2. And start job. -> And total tasks(map/reduce) are running. -> And failed attempt for some reasons.
3. Request allocation of new containers because the attempt failed.
4. Quickly receive allocation of new containers.
5. However, new container is released because failed attempts have not been cleared up.(allocated == total == running limit)
6. Subsequently, the failed attempts is terminated, but it is waiting forever.
7. Job is stuck.

We switched MR frameworks(2.7.1) and checked that it worked well.
Perhaps it is related to MAPREDUCE-6697

Can you help me?