[
https://issues.apache.org/jira/browse/HDFS-5456?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13813350#comment-13813350
]
Chris Nauroth commented on HDFS-5456:
-------------------------------------
I saw this happen in an HA setup using QJM. The backup namenode continually
loads new edits and ends up creating new startup progress steps under the
"Loading edits" phase.
> NameNode startup progress creates new steps if caller attempts to create a
> counter for a step that doesn't already exist.
> -------------------------------------------------------------------------------------------------------------------------
>
> Key: HDFS-5456
> URL: https://issues.apache.org/jira/browse/HDFS-5456
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Affects Versions: 3.0.0, 2.2.0
> Reporter: Chris Nauroth
> Assignee: Chris Nauroth
> Priority: Critical
>
> NameNode startup progress is supposed to be immutable after startup has
> completed. All methods are coded to ignore update attempts after startup has
> completed. However, {{StartupProgress#getCounter}} does not implement this
> correctly. If a caller attempts to get a counter for a new step that hasn't
> been seen before, then the method accidentally creates the step. This
> allocates additional space in the internal tracking data structures, so
> ultimately this is a memory leak.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
