karthick-rn commented on a change in pull request #270: Support Accumulo
installs on Microsoft Azure
URL: https://github.com/apache/fluo-muchos/pull/270#discussion_r313175634
##########
File path: ansible/roles/hadoop/tasks/main.yml
##########
@@ -34,17 +34,52 @@
with_items:
- workers
when: hadoop_major_version == '3'
-- name: "copy spark yarn shuffle jar to hadoop lib"
- command: cp {{ spark_home }}/yarn/spark-{{ spark_version }}-yarn-shuffle.jar
{{ hadoop_home }}/share/hadoop/yarn/lib/ creates={{ hadoop_home
}}/share/hadoop/yarn/lib/spark-{{ spark_version }}-yarn-shuffle.jar
- when: "'spark' in groups"
+- name: "configure hadoop to send metrics"
+ template: src={{ item }} dest={{ hadoop_home }}/etc/hadoop/{{ item }}
+ with_items:
+ - hadoop-metrics2.properties
- name: "setup hadoop short circuit socket dir"
file: path=/var/lib/hadoop-hdfs state=directory owner={{ cluster_user }}
group={{ cluster_group }} mode=0755
become: yes
- name: "Configure hadoop log dir"
replace:
path: "{{ hadoop_home }}/etc/hadoop/hadoop-env.sh"
regexp: '.*export\s+HADOOP_LOG_DIR.*'
- replace: "export HADOOP_LOG_DIR={{ worker_data_dirs[0] }}/logs/hadoop"
+ replace: "export HADOOP_LOG_DIR={{ worker_data_dirs[1] }}/logs/hadoop"
+- name: "Configure max log size to 5g for Azure log analytics integration"
+ replace:
+ path: "{{ hadoop_home }}/etc/hadoop/log4j.properties"
+ regexp: 'hadoop.log.maxfilesize=256MB'
+ replace: "hadoop.log.maxfilesize=5g"
+ when: cluster_type == 'azure'
+- name: "Configure hadoop pid dir"
+ replace:
+ path: "{{ hadoop_home }}/etc/hadoop/hadoop-env.sh"
+ regexp: '.*export\s+HADOOP_PID_DIR.*'
+ replace: "export HADOOP_PID_DIR={{ worker_data_dirs[1] }}/hadoop"
+- name: "Configure namenode heap"
+ replace:
+ path: "{{ hadoop_home }}/etc/hadoop/hadoop-env.sh"
+ regexp: '.*export\s+HADOOP_NAMENODE_OPTS.*'
+ replace: "export HADOOP_NAMENODE_OPTS=\"-server -XX:ParallelGCThreads=8
-XX:+UseConcMarkSweepGC -XX:CMSInitiatingOccupancyFraction=70
-XX:+UseCMSInitiatingOccupancyOnly {{ namenode_heap }}
-Dhadoop.security.logger=${HADOOP_SECURITY_LOGGER:-INFO,RFAS}
-Dhdfs.audit.logger=${HDFS_AUDIT_LOGGER:-INFO,NullAppender} -verbose:gc
-XX:+PrintGCDetails -XX:+PrintGCTimeStamps -XX:+PrintGCDateStamps -Xloggc:{{
worker_data_dirs[1] }}/logs/hadoop/gc.log-`date +'%Y%m%d%H%M'` -XX:ErrorFile={{
worker_data_dirs[1] }}/logs/hadoop/hs_err_pid%p.log
-XX:+HeapDumpOnOutOfMemoryError -XX:HeapDumpPath={{ worker_data_dirs[1]
}}/logs/hadoop\""
Review comment:
@keith-turner Thanks for notifying. I had a look and found
[JEP-291](https://openjdk.java.net/jeps/291) which talks about CMS deprecated
in JDK 9. This does affect Accumulo as well because both Hadoop & Accumulo had
been using CMS from the original muchos build & we haven't introduced this for
Azure changes. Let's discuss if we want to switch to G1GC?
`-verbose:gc -XX:+PrintGCDetails -XX:+PrintGCTimeStamps
-XX:+PrintGCDateStamps -Xloggc:{{ worker_data_dirs[1]
}}/logs/hadoop/gc.log-`date +'%Y%m%d%H%M'` -XX:ErrorFile={{ worker_data_dirs[1]
}}/logs/hadoop/hs_err_pid%p.log -XX:+HeapDumpOnOutOfMemoryError
-XX:HeapDumpPath={{ worker_data_dirs[1] }}/logs/hadoop`
In regards to the above parameters, they have been added for diagnostics and
troubleshooting purpose. I think they are pretty much standard & independent of
CMS or G1GC. Correct me if I'm wrong?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
