[jira] [Created] (MYRIAD-125) Myriad /config API issue for "YARN_NODEMANAGER_OPTS" param when RM's hostname is passed.

["Sarjeet Singh (JIRA)" <ji...@apache.org>]

Sarjeet Singh created MYRIAD-125:
------------------------------------

             Summary: Myriad /config API issue for "YARN_NODEMANAGER_OPTS" param when RM's hostname is passed.
                 Key: MYRIAD-125
                 URL: https://issues.apache.org/jira/browse/MYRIAD-125
             Project: Myriad
          Issue Type: Bug
          Components: Scheduler
    Affects Versions: Myriad 0.1.0
            Reporter: Sarjeet Singh


The issue is with "YARN_NODEMANAGER_OPTS" param that has "-Dyarn.resourcemanager.hostname=<host-name>" keep appending on each offer created by myriad to yarnEnvironment. 

Found this piece of code that has the issue.

In TaskFactory.java:
<code>
String rmHostName = System.getProperty(YARN_RESOURCEMANAGER_HOSTNAME);
            if (rmHostName != null && !rmHostName.isEmpty()) {

                String nmOpts = nmTaskConfig.getYarnEnvironment().get(YARN_NODEMANAGER_OPTS_KEY);
                if (nmOpts == null) {
                    nmOpts = "";
                }
                nmOpts += " " + "-D" + YARN_RESOURCEMANAGER_HOSTNAME + "=" + rmHostName;
                nmTaskConfig.getYarnEnvironment().put(YARN_NODEMANAGER_OPTS_KEY, nmOpts);
                LOGGER.info(YARN_RESOURCEMANAGER_HOSTNAME + " is set to " + rmHostName +
                        " via YARN_RESOURCEMANAGER_OPTS. Passing it into YARN_NODEMANAGER_OPTS.");
            }

Observed this when I tried to start RM from marathon, and HTTP GET the /config API and checked "YARN_NODEMANAGER_OPTS" between 10-15 min interval .
</code>
 
Here is what I observed:

"yarnEnvironment": {
        "YARN_HOME": "/opt/mapr/hadoop/hadoop-2.7.0/", 
        "YARN_NODEMANAGER_OPTS": "-Dnodemanager.resource.io-spindles=4.0 -Dyarn.resourcemanager.hostname=rm.marathon.mesos"
    }, 

After 10-15 min of being idle:

"yarnEnvironment": {
        "YARN_HOME": "/opt/mapr/hadoop/hadoop-2.7.0/", 
        "YARN_NODEMANAGER_OPTS": "-Dnodemanager.resource.io-spindles=4.0 -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos -Dyarn.resourcemanager.hostname=rm.marathon.mesos"
    }, 

Let me know if need any addiotinal detail about the issue? Thanks.

FYI, this should be easy to reproducible by just starting Myriad and check /config API output between 10-15 min interval.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)