[
https://issues.apache.org/jira/browse/HIVE-16484?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15980076#comment-15980076
]
Sahil Takiar commented on HIVE-16484:
-------------------------------------
Only related failure is {{TestSparkClient.testRemoteClient}}
The issue is what happens if {{SPARK_HOME}} is not set. The {{SparkClientImpl}}
has some code to handle the case where {{SPARK_HOME}} is not set; if it isn't
set, then it runs {{bin/java org.apache.spark.deploy.SparkSubmit}}.
This patch deleted the code in {{SparkClientImpl}} that calls {{bin/java
org.apache.spark.deploy.SparkSubmit}} since {{SparkLauncher}} is used for all
Spark job submissions. There was only one test that actually invoked that code
path ({{TestSparkClient.testRemoteClient}})
{{SparkLauncher}} requires {{SPARK_HOME}} to be set since it calls
{{bin/spark-submit}}, it doesn't attempt to call
{{org.apache.spark.deploy.SparkSubmit}} if {{SPARK_HOME}} is not present.
So we could (1) modify {{SparkLauncher}} to not require {{sparkHome}} to be
set, (2) modify this test so that {{SPARK_HOME}} is set, or (3) refactor the
code so that it can still directly invoke {{bin/java
org.apache.spark.deploy.SparkSubmit}} if {{SPARK_HOME}} isn't set.
I'm leaning towards approach 2. [~vanzin] the code to run {{bin/java
org.apache.spark.deploy.SparkSubmit}} if {{SPARK_HOME}} isn't set was added in
HIVE-8528 - is there a use case for launching Spark jobs without {{SPARK_HOME}}
being set, or was it just added for testing?
> Investigate SparkLauncher for HoS as alternative to bin/spark-submit
> --------------------------------------------------------------------
>
> Key: HIVE-16484
> URL: https://issues.apache.org/jira/browse/HIVE-16484
> Project: Hive
> Issue Type: Bug
> Components: Spark
> Reporter: Sahil Takiar
> Assignee: Sahil Takiar
> Attachments: HIVE-16484.1.patch, HIVE-16484.2.patch,
> HIVE-16484.3.patch
>
>
> The {{SparkClientImpl#startDriver}} currently looks for the {{SPARK_HOME}}
> directory and invokes the {{bin/spark-submit}} script, which spawns a
> separate process to run the Spark application.
> {{SparkLauncher}} was added in SPARK-4924 and is a programatic way to launch
> Spark applications.
> I see a few advantages:
> * No need to spawn a separate process to launch a HoS --> lower startup time
> * Simplifies the code in {{SparkClientImpl}} --> easier to debug
> * {{SparkLauncher#startApplication}} returns a {{SparkAppHandle}} which
> contains some useful utilities for querying the state of the Spark job
> ** It also allows the launcher to specify a list of job listeners
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
