[jira] [Created] (AMBARI-11630) Falcon Server Seems To Have Hadoop 2.2 Paths After Upgrade

["Jonathan Hurley (JIRA)" <ji...@apache.org>]

Jonathan Hurley created AMBARI-11630:
----------------------------------------

             Summary: Falcon Server Seems To Have Hadoop 2.2 Paths After Upgrade
                 Key: AMBARI-11630
                 URL: https://issues.apache.org/jira/browse/AMBARI-11630
             Project: Ambari
          Issue Type: Bug
          Components: ambari-server
    Affects Versions: 2.1.0
            Reporter: Jonathan Hurley
            Assignee: Jonathan Hurley
            Priority: Critical
             Fix For: 2.1.0


After an upgrade from HDp 2.2 to HDP 2.3, Falcon seems to have a mixture of HDP 2.2 and 2.3 paths in its process path:

{noformat}
[root@jhurley-hdp22-ru-6 ~]# ps aux | grep falcon
falcon   16231  0.2  6.0 3623356 455580 ?      Sl   14:34   0:18 /usr/jdk64/jdk1.8.0_40/bin/java -Xmx1024m -noverify -Dfalcon.embeddedmq=True -Dfalcon.emeddedmq.port=61616 -Dfalcon.log.dir=/var/log/falcon -Dfalcon.embeddedmq.data=/hadoop/falcon/embeddedmq/data -Dfalcon.home=/usr/hdp/current/falcon-server -Dconfig.location=/usr/hdp/current/falcon-server/conf -Dfalcon.app.type=falcon -Dfalcon.catalog.service.enabled= -cp /usr/hdp/current/falcon-server/conf:/usr/hdp/2.2.7.0-2808/hadoop/conf:/usr/hdp/2.2.7.0-2808/hadoop/lib/*:/usr/hdp/2.2.7.0-2808/hadoop/.//*:/usr/hdp/2.2.7.0-2808/hadoop-hdfs/./:/usr/hdp/2.2.7.0-2808/hadoop-hdfs/lib/*:/usr/hdp/2.2.7.0-2808/hadoop-hdfs/.//*:/usr/hdp/2.2.7.0-2808/hadoop-yarn/lib/*:/usr/hdp/2.2.7.0-2808/hadoop-yarn/.//*:/usr/hdp/2.2.7.0-2808/hadoop-mapreduce/lib/*:/usr/hdp/2.2.7.0-2808/hadoop-mapreduce/.//*::/usr/hdp/current/hadoop-mapreduce-client/*:/usr/hdp/current/tez-client/*:/usr/hdp/current/tez-client/lib/*:/etc/tez/conf/:/usr/hdp/2.2.7.0-2808/tez/*:/usr/hdp/2.2.7.0-2808/tez/lib/*:/etc/tez/conf:/usr/hdp/current/falcon-server/server/webapp/falcon/WEB-INF/classes:/usr/hdp/current/falcon-server/server/webapp/falcon/WEB-INF/lib/*:/usr/hdp/current/falcon-server/libext/* org.apache.falcon.Main -app /usr/hdp/current/falcon-server/server/webapp/falcon -port 15000
root     28434  0.0  0.0 103252   860 pts/5    S+   16:44   0:00 grep falcon

[root@jhurley-hdp22-ru-6 ~]# ll /usr/hdp/current | grep falcon
lrwxrwxrwx. 1 root root 28 May 29 03:30 falcon-client -> /usr/hdp/2.2.7.0-2808/falcon
lrwxrwxrwx. 1 root root 28 May 29 14:34 falcon-server -> /usr/hdp/2.3.0.0-2162/falcon
{noformat}

A restart of Falcon post-upgrade corrects these problems, so it seems like it's an issue only during upgrade.

Also, {{falcon-client}} is not upgraded as part of the upgrade process. This may be covered by the {{hdp-select all}} work capture in another Jira, but falcon-server may still need the client updated ahead of that.

The root cause of this is the fact that Ambari is not setting {{HADOOP_HOME}} when invoking Falcon. Under most circumstances, this is OK since Falcon's {{falcon-config.sh}} script will then try to figure it out with {{which hadoop}}:

{code:title=falcon-config.sh HDP 2.2}
    HADOOPDIR=`which hadoop`
    if [ "$HADOOPDIR" != "" ]; then
      echo "Hadoop is installed, adding hadoop classpath to falcon classpath"
      FALCONCPPATH="${FALCONCPPATH}:`hadoop classpath`"
    elif [ "$HADOOP_HOME" != "" ]; then
      echo "Hadoop home is set, adding libraries from '${HADOOP_HOME}/bin/hadoop classpath' into falcon classpath"
      FALCONCPPATH="${FALCONCPPATH}:`${HADOOP_HOME}/bin/hadoop classpath`"
    else
      echo "Could not find installed hadoop and HADOOP_HOME is not set."
      echo "Using the default jars bundled in ${BASEDIR}/hadooplibs/"
      FALCONCPPATH="${FALCONCPPATH}:${BASEDIR}/hadooplibs/*"
    fi
{code}

The problem in HDP 2.2 is that it will always find the {{HADOOPDIR}} first and never pickup {{HADOOP_HOME}}. In an upgraded environment, it will always use the 2.2 libs. 

This appears to heave been fixed in HDP 2.3:

{code:title=falcon-config.sh HDP 2.3}
    FALCONCPPATH="$FALCON_CONF" 
    HADOOPDIR=`which hadoop`
    if [ "$HADOOP_HOME" != "" ]; then
      echo "Hadoop home is set, adding libraries from '${HADOOP_HOME}/bin/hadoop classpath' into falcon classpath"
      FALCONCPPATH="${FALCONCPPATH}:`${HADOOP_HOME}/bin/hadoop classpath`"
    elif [ "$HADOOPDIR" != "" ]; then
      echo "Hadoop is installed, adding hadoop classpath to falcon classpath"
      FALCONCPPATH="${FALCONCPPATH}:`hadoop classpath`"
    else
      echo "Could not find installed hadoop and HADOOP_HOME is not set."
      echo "Using the default jars bundled in ${BASEDIR}/hadooplibs/"
      FALCONCPPATH="${FALCONCPPATH}:${BASEDIR}/hadooplibs/*"
    fi
{code}

So we just need to ensure we pass in {{HADOOP_HOME}}.




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