[SPARK-8607] SparkR -- jars not being added to application classpath correctly by cafreeman · Pull Request #7001 · apache/spark

cafreeman · 2015-06-24T22:47:16Z

Add getStaticClass method in SparkR's RBackendHandler

This is a fix for the problem referenced in SPARK-5185.

Add `getStaticClass` method in SparkR's `RBackendHandler`

cafreeman · 2015-06-24T22:50:00Z

JIRA issue open here: https://issues.apache.org/jira/browse/SPARK-8607

shivaram · 2015-06-24T22:50:23Z

Style nit: Space after // - Also could you expand the comment to say Looks up a class given a class name. This function first checks the current class loader and if a class is not found, it looks up the class in the context class loader.

shivaram · 2015-06-24T22:53:46Z

Jenkins, add to whitelist

shivaram · 2015-06-24T22:53:51Z

Jenkins, ok to test

SparkQA · 2015-06-24T23:00:22Z

Test build #35732 has finished for PR 7001 at commit 7c6bd0c.

This patch fails Scala style tests.
This patch merges cleanly.
This patch adds no public classes.

cafreeman · 2015-06-24T23:54:24Z

@shivaram Fixed style nits.

shivaram · 2015-06-25T00:09:22Z

@cafreeman Code change looks good -- I'm wondering if there is way to add a test for this in SparkR. We could add a dummy jar file and then try and see if we can call into that ? @davies Do we have a dummy jar file we use for similar tests in Python / Java etc.

SparkQA · 2015-06-25T01:38:31Z

Test build #35738 has finished for PR 7001 at commit 5a80844.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

shivaram · 2015-06-25T01:40:35Z

Jenkins, retest this please

SparkQA · 2015-06-25T03:51:15Z

Test build #35742 has finished for PR 7001 at commit 5a80844.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

cafreeman · 2015-06-25T16:03:54Z

@shivaram where would we store test data? I have a JAR we could probably use.

shivaram · 2015-06-25T16:10:03Z

We could put it in in R/pkg/inst/test_support or something like that. It would be similar to the python files at https://github.com/apache/spark/tree/master/python/test_support

cafreeman · 2015-06-25T17:16:34Z

@shivaram Just committed a first draft of a test, but there are two issues:

The current build appears to be broken. I'm getting issues about sparkPackages being an unused argument and "object 'sparkSubmitBinName' not found". Is this branch currently in a state of flux?
I'm using a relative path in the test to point to the JAR (which has been placed in test_support), and I'm not sure how that will play. I can't actually run it right now though because I can't launch a spark context due to the first issue.

shivaram · 2015-06-25T17:19:49Z

Sorry the first one was a bug we introduced yesterday. Let me send a hotfix for it right now

shivaram · 2015-06-25T17:28:25Z

@cafreeman The worrying thing is that the tests seem to pass right now on Jenkins and on my machine if I run ./run-tests.sh -- If you get a chance it'll be good to figure out why that is happening

shivaram · 2015-06-25T17:33:15Z

@cafreeman Created #7022 -- Could you take a look at it ?

shivaram · 2015-06-25T18:40:22Z

@cafreeman Regarding the test itself, a couple of things

Could we have a smaller JAR file ? This one looks like its around 5MB which is pretty big :) Instead of using sbt assembly to generate it, could you try to use sbt package ? That should make it smaller
Regarding the paths, you can use SPARK_HOME and refer things relative to that. See https://github.com/apache/spark/blob/master/python/pyspark/tests.py#L400 for an example in Python

SparkQA · 2015-06-25T18:47:00Z

Test build #35798 has finished for PR 7001 at commit 9a5c362.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

cafreeman · 2015-06-25T20:44:03Z

@shivaram Alright, I've uploaded a smaller JAR and switched to an absolute filepath using SPARK_HOME. So far this test runs successfully in both the sparkR shell and in regular R, but fails when triggered along with the entire test suite, but I'm not sure why. I added a sparkR.stop() so we could reboot the Spark Context and add the JAR, but it's still having trouble when it's run with run-tests.sh

SparkQA · 2015-06-25T22:58:47Z

Test build #35810 has finished for PR 7001 at commit 0bea809.

This patch fails SparkR unit tests.
This patch merges cleanly.
This patch adds no public classes.

shivaram · 2015-06-26T15:11:47Z

Thanks @cafreeman for the update. I'll try this out today and see why run-tests.sh is not picking it.

shivaram · 2015-06-26T17:04:05Z

So I took a look at this and I managed to figure out why the tests don't work but why it works from sparkR shell. Its due to the fact we have two different ways in which things are initialized

For the sparkR shell (and RStudio), the R process comes up first and then the JVM is launched when the spark context is created. In this case if we pass in sparkJars they are added to the spark-submit command and things work fine.
For the unit tests (and any script run with spark-submit) the JVM comes up first and forks the R process. In this case adding sparkJars to the context has no effect as the JVM is already running. In this case the user needs to run sparkR --jars <jarFile> to get things to work correctly.

For the unit tests we can't test the first case as it is a batch script. But we can test the second case by actually running the sparkR script from the unit test. This would be something similar to https://github.com/apache/spark/blob/master/python/pyspark/tests.py#L1641

`test_includeJAR.R` now executes an external sparkR script which, in turn, initializes a new spark context with the test JAR, runs the test functions, and returns the output to the original test. As a result, we can now run this test inside the SparkR test suite and get around the fact that we can't add the "sparkJars" argument to `sparkR.init()` once the JVM has already started.

cafreeman · 2015-06-26T19:00:08Z

@shivaram Took your advice and changed things so that the test runs another script using R's system2. The tests work now :)

SparkQA · 2015-06-26T20:37:27Z

Test build #35876 has finished for PR 7001 at commit 8f81194.

This patch fails Spark unit tests.
This patch merges cleanly.
This patch adds no public classes.

shivaram · 2015-06-26T20:40:27Z

Hmm - weirdly the Scala spark-submit test failed. Lets give this another go and see if it was just a flaky test.

Jenkins, retest this please

shivaram · 2015-06-26T21:23:58Z

Jenkins, retest this please

SparkQA · 2015-06-26T23:38:15Z

Test build #35884 has finished for PR 7001 at commit 8f81194.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

shivaram · 2015-06-26T23:56:19Z

LGTM. That test case looks great. Thanks @cafreeman

cafreeman · 2015-06-27T00:02:04Z

No problem!

…correctly Add `getStaticClass` method in SparkR's `RBackendHandler` This is a fix for the problem referenced in [SPARK-5185](https://issues.apache.org/jira/browse/SPARK-5185). cc shivaram Author: cafreeman <cfreeman@alteryx.com> Closes#7001 from cafreeman/branch-1.4 and squashes the following commits: 8f81194 [cafreeman] Add missing license 31aedcf [cafreeman] Refactor test to call an external R script 2c22073 [cafreeman] Merge branch 'branch-1.4' of github.com:apache/spark into branch-1.4 0bea809 [cafreeman] Fixed relative path issue and added smaller JAR ee25e60 [cafreeman] Merge branch 'branch-1.4' of github.com:apache/spark into branch-1.4 9a5c362 [cafreeman] test for including JAR when launching sparkContext 9101223 [cafreeman] Merge branch 'branch-1.4' of github.com:apache/spark into branch-1.4 5a80844 [cafreeman] Fix style nits 7c6bd0c [cafreeman] [SPARK-8607] SparkR

[SPARK-8607] SparkR
7c6bd0c
Add `getStaticClass` method in SparkR's `RBackendHandler`

shivaram reviewed Jun 24, 2015
View reviewed changes

Fix style nits
5a80844

cafreeman added 2 commits June 25, 2015 11:51

Merge branch 'branch-1.4' of github.com:apache/spark into branch-1.4
9101223

test for including JAR when launching sparkContext
9a5c362

cafreeman added 2 commits June 25, 2015 13:48

Merge branch 'branch-1.4' of github.com:apache/spark into branch-1.4
ee25e60

Fixed relative path issue and added smaller JAR
0bea809

cafreeman added 2 commits June 26, 2015 12:58

Merge branch 'branch-1.4' of github.com:apache/spark into branch-1.4
2c22073

Add missing license
8f81194

asfgit closed this in 9d11817Jun 27, 2015

Uh oh!

Conversation

cafreeman commented Jun 24, 2015

Uh oh!

cafreeman commented Jun 24, 2015

Uh oh!

shivaramJun 24, 2015

Choose a reason for hiding this comment

Uh oh!

shivaram commented Jun 24, 2015

Uh oh!

shivaram commented Jun 24, 2015

Uh oh!

SparkQA commented Jun 24, 2015

Uh oh!

cafreeman commented Jun 24, 2015

Uh oh!

shivaram commented Jun 25, 2015

Uh oh!

SparkQA commented Jun 25, 2015

Uh oh!

shivaram commented Jun 25, 2015

Uh oh!

SparkQA commented Jun 25, 2015

Uh oh!

cafreeman commented Jun 25, 2015

Uh oh!

shivaram commented Jun 25, 2015

Uh oh!

cafreeman commented Jun 25, 2015

Uh oh!

shivaram commented Jun 25, 2015

Uh oh!

shivaram commented Jun 25, 2015

Uh oh!

shivaram commented Jun 25, 2015

Uh oh!

shivaram commented Jun 25, 2015

Uh oh!

SparkQA commented Jun 25, 2015

Uh oh!

cafreeman commented Jun 25, 2015

Uh oh!

SparkQA commented Jun 25, 2015

Uh oh!

shivaram commented Jun 26, 2015

Uh oh!

shivaram commented Jun 26, 2015

Uh oh!

cafreeman commented Jun 26, 2015

Uh oh!

SparkQA commented Jun 26, 2015

Uh oh!

shivaram commented Jun 26, 2015

Uh oh!

shivaram commented Jun 26, 2015

Uh oh!

SparkQA commented Jun 26, 2015

Uh oh!

shivaram commented Jun 26, 2015

Uh oh!

cafreeman commented Jun 27, 2015

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants