HDFS/YARN start by Ambari v.s command line & rack arware

[#1]

Just wondering whether Abmari does anything additional, when start HDFS/YARN, comparing to from command line directly.

I noted below case while testing rack awareness which prompts this questions. Thanks, Allen.

• I have 4 node cluster installed by Ambari By default, it seems Ambari has already has rack awareness enabled.
  • core-site.xml: <name>net.topology.script.file.name</name>core-site.xml: <value>/etc/hadoop/conf/topology_script.py</value>
  • topology_script.py reads config data from topology_mappings.data

  topology_script.py:DATA_FILE_NAME = os.path.dirname(os.path.abspath(__file__)) + "/topology_mappings.data"

• For testing purpose, I modified the rack info in topology_mappings.data

• a: when restart the HDFS and YARN manually, the change takes effect

• b. when restart by Ambari, the change is NOT take effect. And further, the topology_mappings.data content is over-written with 'default' value and all my change has gone. (Is Ambari collecting the rack info automatically somewhere ? )

[#2 · Accepted solution]

This question can be ignored. After further reading, i understood this is expected. Ambari keeps the configuration somewhere else (most likely in its own repository and then extract and refresh the .xml during start up. Sound makes sense for the usage of repository).