Resource Manager Heap usage when cluster is idle

[#1]

I notice when a cluster is idle (no jobs running) that the Resource Manager heap usage fluctuates between 6% up to 32%. This is when no jobs are running on the cluster. Is it normal for heap usage to get this high? I start the cluster and all services via Ambari, heap is at 6%, goes up to 32%, drops, and then repeats.

[#2 · Accepted solution]

RM store some application states to render the UI which is controlled by yarn.resourcemanager.max-completed-applications, the default value for this is 10000, so at any time RM need somewhere ~1G memory to store these applications in memory. you can try to lower the value for yarn.resourcemanager.max-completed-applications to see the drop in RM heap utilization.