RM job was stuck when running with oozie

I'm running a mapreduce wordcount job task on oozie. 2 jobs were submitted to the yarn, and then the monitoring tasks running upto 99% were stuck. Wordcount job has been 0%.

When I kill off the monitor job, wordcount job runs smoothly.

I use a cluster of 3 virtual machines, configuration is as follows:

Profile per VM: cores=2 memory=2048MB reserved=0GB usableMem=0GB disks=1
Num Container=3
Container Ram=640MB
Used Ram=1GB
Unused Ram=0GB
yarn.scheduler.minimum-allocation-mb=640
yarn.scheduler.maximum-allocation-mb=1920
yarn.nodemanager.resource.memory-mb=1920
mapreduce.map.memory.mb=640
mapreduce.map.java.opts=-Xmx512m
mapreduce.reduce.memory.mb=1280
mapreduce.reduce.java.opts=-Xmx1024m
yarn.app.mapreduce.am.resource.mb=640
yarn.app.mapreduce.am.command-opts=-Xmx512m
mapreduce.task.io.sort.mb=256
Is there any way to solve this, to make two jobs run smoothly and finish at the same time ?