YARN Resource Manager "Reserved Memory" and "Reserved Cores" Are Higher than Allocated Resources Causing Users to be Unable to Schedule Yarn Jobs After BDA 4.9 Upgrade (Doc ID 2313773.1)

Last updated on NOVEMBER 08, 2022

Applies to:

Symptoms

After upgrading a 6 node Mammoth cluster from 4.8/CDH 5.10.1 to 4.9/CDH 5.11.1 where each node has 88TB and 256 GB RAM with 88V cores, the Hadoop Yarn Resource Manager shows higher memory and vcores reserved than those that are actually allocated.

For example, Hadoop shows Cluster Metrics with values like:

Due to this problem, users are unable to run any new jobs.

Stopping the services and trying to clear the issue does not resolve it.

Cause

My Oracle Support provides customers with access to over a million knowledge articles and a vibrant support community of peers and Oracle experts.