[ https://issues.apache.org/jira/browse/IGNITE-3185?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15334440#comment-15334440 ]
Ivan Veselovsky edited comment on IGNITE-3185 at 6/16/16 7:02 PM: ------------------------------------------------------------------ As I understand, in case of almost all non-apache distributions the hdfs & mapred homes are not sub-folders of HADOOP_HOME. HADOOP_HOME by its original sense represents hadoop client base folder (${HADOOP_HOME}/bin/hadoop executable file assumed to exist). So, common , hdfs and mapred base folders typically have some common root , but this root is not a value of HADOOP_HOME. E.g. out docs v1.0 say ( http://apacheignite.gridgain.org/v1.0/docs/installing-on-hortonworks-hdp ) : {code} HDP=/usr/hdp/current export HADOOP_HOME=$HDP/hadoop-client/ export HADOOP_COMMON_HOME=$HDP/hadoop-client/ export HADOOP_HDFS_HOME=$HDP/hadoop-hdfs-client/ export HADOOP_MAPRED_HOME=$HDP/hadoop-mapreduce-client/ {code} Note that more late version of that page defines HADOOP_HOME differently ( https://apacheignite-fs.readme.io/docs/installing-on-hortonworks-hdp ): here HADOOP_HOME is not a client home folder. I would suggest to implement solution described in IGNITE-483 , item "2)": specify a directory location profile for each supported distribution. The user will just need to chose the right profile name to get correct layout. was (Author: iveselovskiy): As I understand, in case of almost all non-apache distributions the hdfs & mapred homes are not sub-folders of HADOOP_HOME. HADOOP_HOME by its original sense represents hadoop client base folder (${HADOOP_HOME}/bin/hadoop executable file assumed to exist). So, common , hdfs and mapred base folders typically have some common root , but this root is not a value of HADOOP_HOME. E.g. out docs v1.0 say ( http://apacheignite.gridgain.org/v1.0/docs/installing-on-hortonworks-hdp ) : {code} HDP=/usr/hdp/current export HADOOP_HOME=$HDP/hadoop-client/ export HADOOP_COMMON_HOME=$HDP/hadoop-client/ export HADOOP_HDFS_HOME=$HDP/hadoop-hdfs-client/ export HADOOP_MAPRED_HOME=$HDP/hadoop-mapreduce-client/ {code} Note that more late version of that page defines HADOOP_HOME differently ( https://apacheignite-fs.readme.io/docs/installing-on-hortonworks-hdp ): here HADOOP_HOME is not a client home folder. I would suggest another solution: specify a directory location profile for each supported distribution. The user will just need to chose the right profile name to get correct layout. > Hadoop: Improve Hadoop JAR search logic. > ---------------------------------------- > > Key: IGNITE-3185 > URL: https://issues.apache.org/jira/browse/IGNITE-3185 > Project: Ignite > Issue Type: Bug > Components: hadoop > Affects Versions: 1.5.0.final > Reporter: Vladimir Ozerov > Assignee: Ivan Veselovsky > Fix For: 1.7 > > > *Problem* > Currently our {{HadoopClassLoader}} is trying to find common/hdfs/mapred JARs > automatically if {{HADOOP_HOME}} is set. However, it checks only for standard > Apache Hadoop distribution paths (/share/hadoop/*). For other distributions > (CDH, HDP), user has to bother with additional env variables manually. > *Solution* > Try to search for the following folders as well: > $HADOOP_HOME/hadoop > $HADOOP_HOME/hadoop-hdfs > $HADOOP_HOME/hadoop-mapreduce > If all these folders are found, then this is either HDP or CDH deployment and > we can create the classpath without additional env vars. -- This message was sent by Atlassian JIRA (v6.3.4#6332)