resourcemanager can not work while there has no any errors

recently , while the hadoop cluster are working .Sometime , there has no any errors in the RM or NM , but the RM do not assigin resource while remaining much resource in the cluster .The following representation are
1、the command line such as "yarn node -list " could execute .The result are all well.
2、all nodes with running containers are all the MRapp container , no yarnchild process is running .
3、all applications that had been submited to the cluster early are all in the progress of 5% ,simultanously they will hang forever.
the key is that we can not find any error or exception , but the applications hung.
The temporary solution is to switch the active resourcemanager,but where is the problem???Could you give some advices?