Vitaly Brodetskyi created AMBARI-24718:
------------------------------------------
Summary: STS fails after start, after stack upgrade from 3.0.1 to
3.0.3
Key: AMBARI-24718
URL: https://issues.apache.org/jira/browse/AMBARI-24718
Project: Ambari
Issue Type: Bug
Reporter: Vitaly Brodetskyi
Assignee: Vitaly Brodetskyi
Fix For: 2.7.3
See this exception in SHS log:
{code}
========================================
Warning: Master yarn-client is deprecated since 2.0. Please use master "yarn"
with specified deploy mode instead.
Exception in thread "main" java.lang.IllegalArgumentException: requirement
failed: Keytab file: none does not exist
at scala.Predef$.require(Predef.scala:224)
at
org.apache.spark.deploy.SparkSubmit$.doPrepareSubmitEnvironment(SparkSubmit.scala:390)
at
org.apache.spark.deploy.SparkSubmit$.prepareSubmitEnvironment(SparkSubmit.scala:250)
at org.apache.spark.deploy.SparkSubmit$.submit(SparkSubmit.scala:171)
at org.apache.spark.deploy.SparkSubmit$.main(SparkSubmit.scala:137)
at org.apache.spark.deploy.SparkSubmit.main(SparkSubmit.scala)
{code}
After i've removed spark.yarn.keytab/principal properties, it start work fine.
By the way, this cluster is NOT kerberized. It's strange why SHS is trying to
use these properties. In the same time properties
spark.history.kerberos.keytab/principal also available but there is no issues.
I expect question why spark.yarn.keytab/principal were added during stack
upgrade if cluster is not kerberized, here is answer:
{code}
<transfer operation="copy" from-type="spark2-defaults"
from-key="spark.history.kerberos.keytab" to-key="spark.yarn.keytab"
default-value="" if-type="spark2-thrift-sparkconf" if-key="spark.yarn.keytab"
if-key-state="absent"/>
<transfer operation="copy" from-type="spark2-defaults"
from-key="spark.history.kerberos.principal" to-key="spark.yarn.principal"
default-value="" if-type="spark2-thrift-sparkconf"
if-key="spark.yarn.principal" if-key-state="absent"/>
{code}
I thought if "spark.history.kerberos.keyta/principal" is available in non
kerberized cluster then "spark.yarn.keytab/principal" could be added too. Also
we have same logic for many other components in ambari. So the question should
it be fixed on ambari side, i mean add spark.yarn.keytab/principal only if
kerberos enabled or some condition should be modified/added on SPARK side, not
to use it if kerberos disabled or value empty/none?
Cluster with repro: http://104.196.75.237:8080 (GCE)
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
