[
https://issues.apache.org/jira/browse/PIG-4489?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14499286#comment-14499286
]
Mohit Sabharwal commented on PIG-4489:
--------------------------------------
Thanks, [~kellyzly], the patch is to enable the 212 tests, which are not
running for Spark engine.
For example, before this patch, unit tests like
{{TestLoadStoreFunctionLifeCycle}} were not running for Spark engine because
this tests creates {{PigServer}} using {code} PigServer pigServer = new
PigServer(Util.getLocalTestMode()); {code} and {{Util.getLocalTestMode()}}
currently returns "local" for Spark unit tests -- which means "MapReduce local".
I think we'd like to run *unit tests* for Spark engine in two modes:
(1) {{ant test-spark-local}} would mean "local" (i.e. SPARK_MASTER="local")
(2) {{ant test-spark}} would mean "local cluster" (i.e.
SPARK_MASTER="local-cluster [ 2 ,2, 1024 ]")
We can add support for "local cluster" in a separate jira. This is a mode
supported in Spark that allows to unit test against a Spark cluster running on
a local node, which multiple JVMs. With "local", unit tests are running against
embedded Spark all in the same JVM.
I don't think we'd ever want to set SPARK_MASTER="spark://xxx:7077" when
running unit tests.
Currently, it is confusing, because {{ant test-spark}} means
SPARK_MASTER="local"
We also need to add a file called {{test/spark-local-tests}} like we have
{{test/spark-tests}} ( just like {{tets/tez-tests}} and
{{test/tez-local-tests}}.) Any idea how we decide which tests go to which file ?
> Enable local mode tests for Spark engine
> ----------------------------------------
>
> Key: PIG-4489
> URL: https://issues.apache.org/jira/browse/PIG-4489
> Project: Pig
> Issue Type: Sub-task
> Components: spark
> Reporter: Mohit Sabharwal
> Assignee: Mohit Sabharwal
> Fix For: spark-branch
>
> Attachments: PIG-4489.1.patch, PIG-4489.patch
>
>
> Util.getLocalTestMode() currently only returns "tez_local" or "local".
> I see that ~212 testcases do this check, and we are not running these tests
> against Spark at this point.
> Currently all Spark tests run in local mode ("local" as a the Spark Cluster
> URL passed to JavaSparkContext), so we should enable these tests as well.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
