[YARN-6188] Fix OOM issue with decommissioningNodesWatcher in the case of clusters with large number of nodes

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.9.0
• Fix Version/s: 2.9.0, 3.0.0-alpha4
• Component/s: resourcemanager
• Target Version/s: 2.9.0
• Hadoop Flags: Reviewed
• Assignee: Ajay Jadhav
• Reporter: Ajay Jadhav
• Created: 14/Feb/17 18:28
• Updated: 17/Feb/17 21:15
• Resolved: 17/Feb/17 21:15

Description

LogDecommissioningNodesStatus method in DecommissioningNodesWatcher uses StringBuilder to append status of all
decommissioning nodes for logging purpose.
In the case of large number of decommissioning nodes, this leads to OOM exception. The fix scopes StringBuilder so that in case of memory pressure, GC can kick in and free up the memory.

This is supposed to fix a bug introduced in https://issues.apache.org/jira/browse/YARN-4676