Apache Oozie Distcp Action with Kerberos enabled clusters

[#1]

my environment requires that I pass

-D ipc.client.fallback-to-simple-auth-allowed=true

to distcp command, in distcp 0.2 action specification for Oozie 4.2, I see java-opts option and I can't seem to make workflow run by passing this property. The only way I can imagine it work is if I put the property in core-site.xml which in production clusters is not feasible. My workflow for reference is

<action name="distcp_1">
        <distcp xmlns="uri:oozie:distcp-action:0.2">
            <job-tracker>${resourceManager}</job-tracker>
            <name-node>${nameNode}</name-node>
            <arg>hdfs://aervits-hdp70:8020/tmp/hellounsecure</arg>
            <arg>hdfs://hacluster:8020/user/centos/</arg>
            <java-opts>-Dipc.client.fallback-to-simple-auth-allowed=true</java-opts>
        </distcp>
        <ok to="end"/>
        <error to="kill"/>
    </action> 

[#2]

Can you add a space between -D and ipc.client... This is not a Java system property but should be the -D option to the Tool runner

[#3]

@Venkat Ranganathan can you be more specific, I tried with space character, I'm getting

Error: E0701 : E0701: XML schema error, cvc-complex-type.2.4.a: Invalid content was found starting with element 'java-opts'. One of '{"uri:oozie:distcp-action:0.2":arg}' is expected.

is this passed as an <arg> or <java-opts>?

<java-opts>-D ipc.client.fallback-to-simple-auth-allowed=true</java-opts>

[#4]

You should have the java-opts right after configuration element and before arg.

[#5]

@Venkat Ranganathan That didn't work, getting

Error: Could not find or load main class ipc.client.fallback-to-simple-auth-allowed=true

in the stderr log. My workflow as of now looks like so

<action name="distcp_1">
        <distcp xmlns="uri:oozie:distcp-action:0.2">
            <job-tracker>${resourceManager}</job-tracker>
            <name-node>${nameNode}</name-node>
            <configuration>
                <property>
                    <name>mapred.job.queue.name</name>
                    <value>${queueName}</value>
                </property>
            </configuration>
            <java-opts>-D ipc.client.fallback-to-simple-auth-allowed=true</java-opts>
            <arg>hdfs://aervits-hdp70:8020/tmp/hellounsecure</arg>
            <arg>hdfs://hacluster:8020/user/centos/</arg>
        </distcp>
        <ok to="end"/>
        <error to="kill"/>
    </action>

[#6]

Let me check. The Java opts might be getting added after the class

[#7]

You can set up your args like this and remove from java-opts

<arg>-D</arg>
<arg>ipc.client.fallback-to-simple-auth-allowed=true</arg>
<arg>hdfs://aervits-hdp70:8020/tmp/hellounsecure</arg>
<arg>hdfs://hacluster:8020/user/centos/</arg>

Thanks

[#8]

BTW, this has to come before distcp arguments (like -update etc)

[#9]

@Venkat Ranganathan tried many different options, as part of configuration block as well as a part of submission, i.e.

oozie job -D ipc.client.fallback-to-simple-auth-allowed=true -run

no luck, I read in the Oozie docs that I need property oozie.launcher.mapreduce.job.hdfs-servers and my jobs stopped getting submitted to YARN, hence I commented it out. I also added hadoop.proxy.oozie.hosts and guests to the 2nd cluster as per docs, no luck.

    <action name="distcp_1">
        <distcp xmlns="uri:oozie:distcp-action:0.2">
            <job-tracker>${resourceManager}</job-tracker>
            <name-node>${nameNode}</name-node>
<!--
            <configuration>
                <property>
                    <name>oozie.launcher.mapreduce.job.hdfs-servers</name>
                    <value>${nameNode},${nameNode2}</value>
                </property>
            </configuration>
-->
            <arg>-D</arg>
            <arg>ipc.client.fallback-to-simple-auth-allowed=true</arg>
            <arg>-overwrite</arg>
            <arg>${nameNode}/user/centos/primary</arg>
            <arg>${nameNode2}/tmp/</arg>
        </distcp>
        <ok to="end"/>
        <error to="kill"/>
    </action>

Finally, I stumbled on this note and I'm afraid I hit such case

IMPORTANT: The DistCp action may not work properly with all configurations (secure, insecure) in all versions of Hadoop.

I take instances where one cluster is secured and 2nd cluster is not is not suppored in Distcp action spec 0.2.

https://oozie.apache.org/docs/4.2.0/DG_DistCpActionExtension.html

[#10 · Accepted solution]

That property is used when one cluster is kerberized and the other isn't, is this your case? If both clusters are kerberized you need to set "oozie.launcher.mapreduce.job.hdfs-servers" property. Regarding your property can you try to pass it as an arg:

<arg>-Dipc.client.fallback-to-simple-auth-allowed=true</arg>

I don't think the space after "-D" matters, but you can try either way.

[#11]

@Predrag Minovic & @Venkat Ranganathan the only way this works with secure and insecure cluster is if core-site.xml has the property below on the secured cluster side.

    <property>
      <name>ipc.client.fallback-to-simple-auth-allowed</name>
      <value>true</value>
    </property>

uploading core-site.xml with this property in it in the lib folder of the workflow doesn't work. <arg> and <java-opts> for this property take no effect. job config would not update, the only way to make it work was to update on the hdfs side globally.

Thank you both for your recommendations, eventually it was Predrag who suggested to update HDFS with the property. This is not optimal but for lack of better option will do. It's best to have either two unsecure or two secure clusters.

[#12]

Can you try to put your copy of core-site.xml with that property set to true in the Oozie share lib folder for distcp?

[#13]

adding to sharelib of distcp core-site.xml with this property does not take effect and including it in lib directory of the workflow doesn't take effect either.

[#14]

That's good to know! Many restrictions with Oozie...