Mark Tse created HDFS-7255:
------------------------------

             Summary: Customize Java Heap min/max settings for individual 
processes
                 Key: HDFS-7255
                 URL: https://issues.apache.org/jira/browse/HDFS-7255
             Project: Hadoop HDFS
          Issue Type: Improvement
          Components: datanode, journal-node, namenode
    Affects Versions: 2.5.1, 2.4.1
            Reporter: Mark Tse


The NameNode and JournalNode (and ZKFC) can all run on the same machine. 
However, they get their heap settings from HADOOP_HEAPSIZE/JAVA_HEAP_MAX. There 
are scenarios where the NameNode process should have different Java memory 
requirements than the JournalNode and ZKFC (e.g. if the machine has 10 GB of 
RAM, and I want the NameNode process to have 8 GB max). 

HADOOP_(.*)_OPTS variables exist for these processes and can be used to add the 
Xms and Xmx tags, but because of how the default for JAVA_HEAP_MAX is set, it 
will always add '-Xmx1000m' to the final call to start up the 
NameNode/JournalNode/ZKFC process, resulting in two different Java heap 
settings (e.g. -Xmx1000m and -Xmx8g is used when starting the NameNode).

Note: HADOOP_HEAPSIZE is deprecated according to [HADOOP-10950]



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to