SPARK-11295 Add packages to JUnit output for Python tests by gliptak · Pull Request #9263 · apache/spark

gliptak · 2015-10-24T16:47:47Z

SPARK-11295 Add packages to JUnit output for Python tests

This improves grouping/display of test case results.

JoshRosen · 2015-10-24T20:15:14Z

Jenkins, this is ok to test.

SparkQA · 2015-10-24T20:25:04Z

Test build #44299 has finished for PR 9263 at commit e5e03b0.

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

SparkQA · 2015-10-24T21:16:32Z

Test build #44300 has finished for PR 9263 at commit 52c3410.

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

gliptak · 2015-10-24T21:19:22Z

pyspark.tests shows up

https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/44300/testReport/

I will work the others.

SparkQA · 2015-10-24T21:52:15Z

Test build #44302 has finished for PR 9263 at commit 5a04363.

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

gliptak · 2015-10-24T22:08:57Z

The test errors with:

Finished test(python2.6): pyspark.mllib.util (9s)
Traceback (most recent call last):
File "/usr/lib64/python2.6/runpy.py", line 122, in _run_module_as_main
"__main__", fname, loader, pkg_name)
File "/usr/lib64/python2.6/runpy.py", line 34, in _run_code
exec code in run_globals
File "/home/jenkins/workspace/SparkPullRequestBuilder@2/python/pyspark/mllib/tests.py", line 1543, in <module>
from pyspark.mllib.tests import *
File "/home/jenkins/workspace/SparkPullRequestBuilder@2/python/pyspark/mllib/tests.py", line 79, in <module>
sc = SparkContext('local[4]', "MLlib tests")
File "/home/jenkins/workspace/SparkPullRequestBuilder@2/python/pyspark/context.py", line 111, in __init__
SparkContext._ensure_initialized(self, gateway=gateway)
File "/home/jenkins/workspace/SparkPullRequestBuilder@2/python/pyspark/context.py", line 257, in _ensure_initialized
callsite.function, callsite.file, callsite.linenum))
ValueError: Cannot run multiple SparkContexts at once; existing SparkContext(app=MLlib tests, master=local[4]) created by <module> at /usr/lib64/python2.6/runpy.py:34 Had test failures in pyspark.mllib.tests with python2.6; see logs.
[error] running /home/jenkins/workspace/SparkPullRequestBuilder@2/python/run-tests --modules=pyspark-core,pyspark-sql,pyspark-streaming,pyspark-mllib,pyspark-ml --parallelism=4 ; received return code 255

gliptak · 2015-10-25T01:49:01Z

http://stackoverflow.com/questions/28259756/how-to-create-multiple-sparkcontexts-in-a-console

#3121

SparkQA · 2015-10-25T02:11:40Z

Test build #44310 has finished for PR 9263 at commit b423482.

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

mengxr · 2015-10-27T04:39:28Z

Is it relevant to this PR?

@mengxr This was my (failing) attempt to correct the test errors (caused by --parallelism=4?). Maybe @JoshRosen could comment?

Yeah, you shouldn't set this. The parallelism in dev/run-tests will actually launch separate JVMs, so that's not the cause of this problem. In general, you should never set spark.driver.allowMultipleContexts (it was only added as an escape-hatch backwards-compatibility option for a feature that we never properly supported).
There must be some other problem in the tests, likely due to test cleanup or SparkContext teardown not being executed properly.

Reviewing the tests.py-s
https://github.com/apache/spark/blob/master/python/pyspark/streaming/tests.py
initiates SparkContext differently:
@classmethod def setUpClass(cls): class_name = cls.__name__ conf = SparkConf().set("spark.default.parallelism", 1) cls.sc = SparkContext(appName=class_name, conf=conf) cls.sc.setCheckpointDir("/tmp") @classmethod def tearDownClass(cls): cls.sc.stop() # Clean up in the JVM just in case there has been some issues in Python API try: jSparkContextOption = SparkContext._jvm.SparkContext.get() if jSparkContextOption.nonEmpty(): jSparkContextOption.get().stop() except: pass
Could this approach be retrofitted into https://github.com/apache/spark/blob/master/python/pyspark/mllib/tests.py to allow for concurrency?

SparkQA · 2015-10-28T00:52:24Z

Test build #44478 has finished for PR 9263 at commit 8dc37f8.

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

gliptak · 2015-10-28T12:47:17Z

This last run had a different failure than the previous run with the same code ...

gliptak · 2015-11-03T15:12:57Z

@JoshRosen Would you have some pointers on how to move this forward? Thanks

mengxr · 2015-11-06T16:58:30Z

test this please

mengxr · 2015-11-06T16:58:34Z

add to whitelist

SparkQA · 2015-11-06T17:28:12Z

Test build #45227 has finished for PR 9263 at commit 8dc37f8.

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

mengxr · 2015-11-06T21:13:34Z

@gliptak I think this is the cause: https://github.com/apache/spark/blob/master/python/pyspark/mllib/tests.py#L79. We didn't initialize SparkContext in setUp. So by import * it creates a new SparkContext. Could you move it to setUp and retry at your local?

gliptak · 2015-11-07T03:07:27Z

@mengxr Thank you for the pointer. This worked locally with

$ ./run-tests --python-executables=python --modules=pyspark-mllib

gliptak · 2015-11-08T17:19:05Z

@mengxr Could you trigger a build?

SparkQA · 2015-11-08T19:21:59Z

Test build #45309 has finished for PR 9263 at commit f253013.

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

SparkQA · 2015-11-09T01:17:39Z

Test build #45325 has finished for PR 9263 at commit 40b1a65.

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

gliptak · 2015-11-09T01:37:36Z

Failure for only one of the Pythons version?

FAIL: test_trainOn_predictOn (pyspark.mllib.tests.StreamingKMeansTest)
Test that prediction happens on the updated model.
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line 1198, in test_trainOn_predictOn
self._eventually(condition, catch_assertions=True)
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line 128, in _eventually
raise lastValue
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line 119, in _eventually
lastValue = condition()
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line 1195, in condition
self.assertEqual(predict_results, [[0, 1, 1], [1, 0, 1]])
AssertionError: Lists differ: [[0, 1, 1], [0, 0, 0]] != [[0, 1, 1], [1, 0, 1]]

The unit tests page shows no errors:

https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/45325/testReport/

gliptak · 2015-11-10T03:23:12Z

Are JUnit test results separated based on Java version running under?

gliptak · 2015-11-12T19:29:28Z

#9669

gliptak · 2015-11-14T19:57:39Z

Rebased to current master

SparkQA · 2015-11-14T20:26:11Z

Test build #45936 has finished for PR 9263 at commit f733447.

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

gliptak · 2015-11-14T21:49:25Z

Unit test timed out?

======================================================================
ERROR: test_kafka_direct_stream (pyspark.streaming.tests.KafkaStreamTests)
Test the Python direct Kafka stream API.
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/streaming/tests.py", line 753, in setUp
self._kafkaTestUtils.setup()
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/lib/py4j-0.9-src.zip/py4j/java_gateway.py", line 813, in __call__
answer, self.gateway_client, self.target_id, self.name)
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/lib/py4j-0.9-src.zip/py4j/protocol.py", line 308, in get_return_value
format(target_id, ".", name), value)
py4j.protocol.Py4JJavaError: An error occurred while calling o10749.setup.
: org.I0Itec.zkclient.exception.ZkTimeoutException: Unable to connect to zookeeper server within timeout: 6000
at org.I0Itec.zkclient.ZkClient.connect(ZkClient.java:880)
at org.I0Itec.zkclient.ZkClient.<init>(ZkClient.java:98)

zsxwing · 2015-12-11T00:21:48Z

retest this please

SparkQA · 2015-12-11T00:51:49Z

Test build #47554 has finished for PR 9263 at commit f733447.

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

gliptak · 2015-12-14T16:21:04Z

https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/47554/testReport/ shows 0 failures ...

gliptak · 2015-12-20T00:43:45Z

@zsxwing Are there some changes you would like to see to this pull request?

JoshRosen · 2016-01-13T20:27:34Z

Hey, sorry for forgetting about this. I'm going to trigger a retest now and will take a look at the test results.

@gliptak, could you please update the PR description to say something other than "WIP", since the PR description will become the commit message? Feel free to copy description from JIRA if you'd like. I'd also be nice to add a sentence or two summarizing the changes that you needed to make to get this to work.

JoshRosen · 2016-01-13T20:27:36Z

Jenkins, retest this please.

gliptak · 2016-01-13T22:15:50Z

@JoshRosen Please let me know if you would like to see other description changes.

SparkQA · 2016-01-13T22:49:25Z

Test build #49329 has finished for PR 9263 at commit f733447.

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

JoshRosen · 2016-01-13T22:57:49Z

I can see why sc needs to be created on a per-test-case basis, but is it possible to leave ser as-is and keep it here?

I ask because it looks like this change conflicted with a recently-modified test, causing the test to break:
======================================================================ERROR [7.323s]: test_als_ratings_id_long_error (pyspark.mllib.tests.ALSTests) ----------------------------------------------------------------------Traceback (mostrecentcalllast): File"/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line1571, intest_als_ratings_id_long_errorself.assertRaises(Py4JJavaError, self.sc._jvm.SerDe.loads, bytearray(ser.dumps(r))) NameError: globalname'ser'isnotdefined======================================================================ERROR [0.779s]: test_als_ratings_serialize (pyspark.mllib.tests.ALSTests) ----------------------------------------------------------------------Traceback (mostrecentcalllast): File"/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line1562, intest_als_ratings_serializejr=self.sc._jvm.SerDe.loads(bytearray(ser.dumps(r))) NameError: globalname'ser'isnotdefined

I recall seeing some concurrency issues with ser too (it has been a while). I'm pushing up a rebase/update.

What do you mean by concurrency errors? AFAIK we run tests serially in these files.

Maybe I was running with parallel flag locally. It has been a while ...

@JoshRosen From the log the tests are running parallel=4.
/home/jenkins/workspace/SparkPullRequestBuilder@2/python/run-tests --modules=pyspark-core,pyspark-sql,pyspark-streaming,pyspark-mllib,pyspark-ml --parallelism=4
I will roll back the ser changes in a few.

JoshRosen · 2016-01-13T23:03:23Z

By the way, aside from the ser issue this looks good to me.

For other reviewers: this change makes the test output much nicer in Jenkins (and thus nicer on https://spark-tests.appspot.com):

One interesting thing: Jenkins seems to fail when trying to show package-level test information for the PySpark tests: https://amplab.cs.berkeley.edu/jenkins/job/SparkPullRequestBuilder/49329/testReport/pyspark.tests/. I wonder whether this could be related to the presence of multiple XML reports from concurrent PySpark test runs with different Python versions.

gliptak · 2016-01-13T23:40:42Z

Yes, the multiple reports might play into it. I have no visibility into Jenkins to assess further.

SparkQA · 2016-01-14T00:07:15Z

Test build #49349 has finished for PR 9263 at commit f23fb62.

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

SparkQA · 2016-01-14T01:15:16Z

Test build #49357 has finished for PR 9263 at commit 99d63c5.

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

gliptak · 2016-01-14T02:35:51Z

FAIL: test_trainOn_predictOn (pyspark.mllib.tests.StreamingKMeansTest)
Test that prediction happens on the updated model.
----------------------------------------------------------------------
Traceback (most recent call last):
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line 1211, in test_trainOn_predictOn
self._eventually(condition, catch_assertions=True)
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line 129, in _eventually
raise lastValue
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line 120, in _eventually
lastValue = condition()
File "/home/jenkins/workspace/SparkPullRequestBuilder/python/pyspark/mllib/tests.py", line 1208, in condition
self.assertEqual(predict_results, [[0, 1, 1], [1, 0, 1]])
AssertionError: Lists differ: [[0, 1, 1], [0, 0, 0]] != [[0, 1, 1], [1, 0, 1]]
First differing element 1:
[0, 0, 0]
[1, 0, 1]
- [[0, 1, 1], [0, 0, 0]]
? ^^^^
+ [[0, 1, 1], [1, 0, 1]]
? +++ ^

gliptak · 2016-01-14T23:41:31Z

@JoshRosen Would you like to see some other changes?

JoshRosen · 2016-01-14T23:57:24Z

I'd like to see what happens if you roll back that ser change.

mengxr · 2016-01-15T01:20:08Z

@gliptak @JoshRosen I think we shouldn't block this feature because of some failed MLlib unit tests. Feel free to disable the tests in this PR and create JIRAs under components "MLlib" and "PySpark" to track them. We could fix them in a follow-up PR.

JoshRosen · 2016-01-15T18:06:05Z

I still maintain that we should roll back the ser change if we don't know why it's necessary.

SparkQA · 2016-01-16T17:15:54Z

Test build #49531 has finished for PR 9263 at commit c524ab0.

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

mengxr · 2016-01-19T22:09:35Z

LGTM. Merged into master. Thanks for making this work! Btw, @gliptak please do not squash your commits into one when you push new changes. It is easier to see what you changed if you keep the old commits unmodified. Thanks!

gliptak · 2016-01-19T22:12:13Z

@mengxr I will keep that in mind. Do you usually squash commits before committing to master? Thanks

JoshRosen · 2016-01-19T22:19:25Z

@gliptak, our merge script automatically squashes at commit time, so there's no need for you to do it yourself.

gliptak · 2016-01-19T22:59:28Z

@JoshRosen I see. Thanks

mengxr · 2016-01-20T00:52:33Z

@gliptak I reverted the change because 0ddba6d was merged before this one and it didn't use self.sc, which causes Jenkins failures. Could you reopen this PR and update your patch?

Btw, this line needs a patch: 0ddba6d#diff-ce16909e38fc8bee429dc638b2b2dde2R426.

SPARK-11295 Add packages to JUnit output for Python tests This improves grouping/display of test case results. Author: Gábor Lipták <gliptak@gmail.com> Closesapache#9263 from gliptak/SPARK-11295.

mengxr · 2016-01-20T18:28:12Z

I made a new PR with fix: #10850

This is #9263 from gliptak (improving grouping/display of test case results) with a small fix of bisecting k-means unit test. Author: Gábor Lipták <gliptak@gmail.com> Author: Xiangrui Meng <meng@databricks.com> Closes#10850 from mengxr/SPARK-11295.

gliptak · 2016-01-20T19:41:58Z

@mengxr Thank you (I didn't get to this last night)

gliptak mentioned this pull request Oct 24, 2015
SPARK-7021 Add JUnit output for Python unit tests #8323
Closed

mengxr reviewed Oct 27, 2015
View reviewed changes

JoshRosen reviewed Jan 13, 2016
View reviewed changes

SPARK-11295Add packages to JUnit output for Python tests
c524ab0

asfgit closed this in c6f971bJan 19, 2016

mengxr mentioned this pull request Jan 20, 2016
[SPARK-11295] [PYSPARK] Add packages to JUnit output for Python tests #10850
Closed

Uh oh!

Conversation

gliptak commented Oct 24, 2015

Uh oh!

JoshRosen commented Oct 24, 2015

Uh oh!

SparkQA commented Oct 24, 2015

Uh oh!

SparkQA commented Oct 24, 2015

Uh oh!

gliptak commented Oct 24, 2015

Uh oh!

SparkQA commented Oct 24, 2015

Uh oh!

gliptak commented Oct 24, 2015

Uh oh!

gliptak commented Oct 25, 2015

Uh oh!

SparkQA commented Oct 25, 2015

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Oct 28, 2015

Uh oh!

gliptak commented Oct 28, 2015

Uh oh!

gliptak commented Nov 3, 2015

Uh oh!

mengxr commented Nov 6, 2015

Uh oh!

mengxr commented Nov 6, 2015

Uh oh!

SparkQA commented Nov 6, 2015

Uh oh!

mengxr commented Nov 6, 2015

Uh oh!

gliptak commented Nov 7, 2015

Uh oh!

gliptak commented Nov 8, 2015

Uh oh!

SparkQA commented Nov 8, 2015

Uh oh!

SparkQA commented Nov 9, 2015

Uh oh!

gliptak commented Nov 9, 2015

Uh oh!

gliptak commented Nov 10, 2015

Uh oh!

gliptak commented Nov 12, 2015

Uh oh!

gliptak commented Nov 14, 2015

Uh oh!

SparkQA commented Nov 14, 2015

Uh oh!

gliptak commented Nov 14, 2015

Uh oh!

zsxwing commented Dec 11, 2015

Uh oh!

SparkQA commented Dec 11, 2015

Uh oh!

gliptak commented Dec 14, 2015

Uh oh!

gliptak commented Dec 20, 2015

Uh oh!

JoshRosen commented Jan 13, 2016

Uh oh!

JoshRosen commented Jan 13, 2016

Uh oh!

gliptak commented Jan 13, 2016

Uh oh!

SparkQA commented Jan 13, 2016

Uh oh!

Choose a reason for hiding this comment

Uh oh!