[SPARK-20661][SparkR][Test] SparkR tableNames() test fails

[falaki]

What changes were proposed in this pull request?

Cleaning existing temp tables before running tableNames tests

How was this patch tested?

SparkR Unit tests

[yhuai]

@felixcheung fyi. I think the main problem of this test is that it will be broken if tests executed before this one leak any table. I think this change makes sense. I will merge it once it passes jenkins.

[shivaram]

LGTM. Thanks @falaki -- BTW is this a problem only on master or should we also backport this ?

[falaki]

@shivaram it seems to have started today with this build: https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test/job/spark-master-test-sbt-hadoop-2.7/2844/consoleFull

The build just before that does not have the failure: https://amplab.cs.berkeley.edu/jenkins/view/Spark%20QA%20Test/job/spark-master-test-sbt-hadoop-2.7/2843/consoleFull

[SparkQA]

Test build #76592 has finished for PR 17903 at commit 7020372.

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

[yhuai]

Seems 2.2 build is fine. But, I'd like to get this merged in branch-2.2 since this test will fail if any previous tests leak tables.

[yhuai]

Thanks @falaki. Merging to master and branch-2.2.

[felixcheung]

Hmm, do we know why temp tables are getting leak since #17649 But not before?

[yhuai]

I do not think #17649 caused the problem. I saw failures without that internally.

[felixcheung]

ah ok, i thought we said it was this build introducing this failure #17892 (comment) and #17903 (comment)

thanks, it's good to know

edit: looking into this, I'm not sure how this failure occurs - Perhaps it is picking up tables created in Scala test. I will follow up for more deterministic way to check this.