[jira] [Commented] (TEZ-1473) TEZ_RUNTIME_SHUFFLE_BUFFER is too large by default

[Tsuyoshi OZAWA (JIRA)]

    [ 
https://issues.apache.org/jira/browse/TEZ-1473?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14106448#comment-14106448
 ] 

Tsuyoshi OZAWA commented on TEZ-1473:
-------------------------------------

The workload is wordcount against 20 GB data generated by randomtext writer. 
Container killer of YARN killed some containers when I use a following 
configuration:

{quote}
<property>
  <name>tez.task.launch.cmd-opts</name>
  <value>-server -Xmx4096m -Djava.net.preferIPv4Stack=true -XX:+UseNUMA 
-XX:+UseParallelGC -Dhadoop.metrics.log.level=WARN</value>
</property>
<property>
  <name>tez.task.resource.memory.mb</name>
  <value>4096</value>
</property>
<property>
  <name>tez.task.resource.memory.mb</name>
  <value>4096</value>
</property>
<property>
  <name>tez.runtime.shuffle.buffersize</name>
  <value>8 * 1024</value>
</property>
{quote}

I can avoid it by making TEZ_RUNTIME_SHUFFLE_BUFFER small:

{quote}
<property>
  <name>tez.runtime.shuffle.buffersize</name>
  <value>3072</value>
</property>
{quote}

I can avoid it by making JVM heap size for tasks small as follows:

{quote}
<property>
  <name>tez.task.launch.cmd-opts</name>
  <value>-server -Xmx3800m -Djava.net.preferIPv4Stack=true -XX:+UseNUMA 
-XX:+UseParallelGC -Dhadoop.metrics.log.level=WARN</value>
</property>
<property>
  <name>tez.task.resource.memory.mb</name>
  <value>4096</value>
</property>
<property>
  <name>tez.runtime.shuffle.buffersize</name>
  <value>8 * 1024</value>
</property>
{quote}



> TEZ_RUNTIME_SHUFFLE_BUFFER is too large by default
> --------------------------------------------------
>
>                 Key: TEZ-1473
>                 URL: https://issues.apache.org/jira/browse/TEZ-1473
>             Project: Apache Tez
>          Issue Type: Improvement
>            Reporter: Tsuyoshi OZAWA
>            Assignee: Tsuyoshi OZAWA
>         Attachments: TEZ-1473.1.patch
>
>
> TEZ_RUNTIME_SHUFFLE_BUFFER is 8GB by default, while 
> TEZ_TASK_RESOURCE_MEMORY_MB_DEFAULT is 1GB. It leads OoM or Container Killer.



--
This message was sent by Atlassian JIRA
(v6.2#6252)