[ https://issues.apache.org/jira/browse/MAPREDUCE-3408?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13151835#comment-13151835 ]
Bruno Mahé commented on MAPREDUCE-3408: --------------------------------------- IMHO option 2/ is the safest and most conservative approach. I will prepare a patch tomorrow. Please let me know if you think of a better way. > yarn-daemon.sh unconditionnaly sets yarn.root.logger > ---------------------------------------------------- > > Key: MAPREDUCE-3408 > URL: https://issues.apache.org/jira/browse/MAPREDUCE-3408 > Project: Hadoop Map/Reduce > Issue Type: Bug > Components: mrv2, nodemanager, resourcemanager > Affects Versions: 0.23.0, 0.23.1 > Reporter: Bruno Mahé > Assignee: Bruno Mahé > Labels: bigtop > > yarn-daemon.sh unconditionnaly sets yarn.root.logger which then prevent any > override from happening. > From ./hadoop-mapreduce-project/hadoop-yarn/bin/yarn-daemon.sh: > > export YARN_ROOT_LOGGER="INFO,DRFA" > > export YARN_JHS_LOGGER="INFO,JSA" > and then yarn-daemon.sh will call "$YARN_HOME"/bin/yarn which does the > following: > > YARN_OPTS="$YARN_OPTS > > -Dhadoop.root.logger=${YARN_ROOT_LOGGER:-INFO,console}" > > YARN_OPTS="$YARN_OPTS -Dyarn.root.logger=${YARN_ROOT_LOGGER:-INFO,console}" > This has at least 2 issues: > * I cannot override hadoop.root.logger when using the yarn-daemon.sh script > * I cannot have different values for hadoop.root.logger and yarn.root.logger > I currently see two different ways to proceed forward: > 1/ Make the script yarn-daemon.sh only sets a default value for > YARN_ROOT_LOGGER if this variable is not defined > 2/ Remove the quoted code from yarn-daemon.sh since yarn already does > something similar > 3/ Entirely remove that chunk and let people define their logging however > they want through some properties files (see log4j.properties in the conf > directories for instance) > I would also use the variable HADOOP_ROOT_LOGGER for hadoop.root.logger if > either option 1/ or 2/ would be taken. > I don't really have any preference toward any of these solutions. What would > you recommend? What is the Apache Hadoop way for this matter? > Note: This is probably happening as well for the other daemons, and I will > take a look at it once this issue is resolved. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira