Application application_1512618719369_147804 failed 2 times due to ApplicationMaster for attempt app

I encountered a very strange problem today. Previously, the hive task in etl kept reporting an error. It lasted all morning, and the cause was not found. The log of the wrong task was also found. I could find the cause, but when I opened the log, I felt a chill in my heart. There is no error, I don't know what's wrong. Finally, observe the node that reported the error, and finally focus on two machines, then see if it is caused by the hadoop program of these two machines? See that their programs are all there, but checking the log of nodemanager keeps reporting errors, and checking the cpu, the cpu occupied by the nodemanager process reaches more than 1000%, and it immediately shines, knowing that the cpu is occupied too much, causing the AP to be unable to contact, causing the task to fail. Finally, restart the nodemanager of these two machines and observe, the task is not reporting an error. keep it up.............