[
https://issues.apache.org/jira/browse/YARN-6062?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15805518#comment-15805518
]
Rohith Sharma K S commented on YARN-6062:
-----------------------------------------
bq. In my understanding, RES include a process' heap memory. No?
Yes, JVM heap is included. But if heap size is growing, then JVM throws an OOM
and kill the process. But if leak in native memory, then process JVM wont get
OOM and it continues to run as long as system memory is utilized fully and OS
will kill the process.
As per analysis in YARN-6017, this issue also looks very similar from given
information(NM maximum heap memory configured is *2GB* but RES is grown till
*10GB*). In YARN-6017, it was very clear(from /proc/<pid>/smaps) that native
memory was leaking. But did not get what native calls were leaking :-(
However, lets wait for reporter to give all informations.
> nodemanager memory leak
> -----------------------
>
> Key: YARN-6062
> URL: https://issues.apache.org/jira/browse/YARN-6062
> Project: Hadoop YARN
> Issue Type: Bug
> Components: nodemanager
> Affects Versions: 2.7.1
> Reporter: gehaijiang
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 8986 data 20 0 21.3g 19g 7376 S 5.5 20.7 2458:09 java
> 38432 data 20 0 9.8g 7.9g 6300 S 95.5 8.4 35273:23 java
> 6653 data 20 0 4558m 3.4g 10m S 9.2 3.6 6640:37 java
> $ jps
> 6653 NodeManager
> Nodemanager memory has been up，Reach 10G。
> nodemanager yarn-env.sh configure (2G)
> YARN_NODEMANAGER_OPTS=" -Xms2048m -Xmn768m
> -Xloggc:${YARN_LOG_DIR}/nodemanager.gc.log -XX:+PrintGCDateStamps
> -XX:+PrintGCDetails -XX:+PrintGCTimeStamps"
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
