[SPARK-12048][SQL] Part 2 Prevent to close JDBC resources twice

[tedyu]

PR #10101 dealt with sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/jdbc/JDBCRDD.scala

This PR applies the same technique on JdbcRDD.scala

[SparkQA]

Test build #47787 has finished for PR 10320 at commit 522afb7.

• This patch passes all tests.
• This patch merges cleanly.
• This patch adds the following public classes (experimental):\n * trait JobSubmitter\n * class ComplexFutureAction[T](run : JobSubmitter => Future[T])\n * case class AttachCompletedRebuildUI(appId: String)\n * case class WrapOption(child: Expression, optType: DataType)\n

[tedyu]

@andrewor14 @zsxwing
Please take a look

[jerryshao]

From my understanding, we could call rs = null after it is closed, also for stmt and conn in the same way to make this function is reenterable.

[srowen]

That's another approach, though the other implementation opted for a close flag

[zsxwing]

I don't think we need this patch. I saw the following comments in javadoc

     * Calling the method <code>close</code> on a <code>Connection</code>
     * object that is already closed is a no-op.

     * Calling the method <code>close</code> on a <code>Statement</code>
     * object that is already closed has no effect.

     * Calling the method <code>close</code> on a <code>ResultSet</code>
     * object that is already closed is a no-op.

[srowen]

Javadoc of which? the JDK interfaces? I think the issue was that SQLite didn't actually seem to like being closed twice in practice.

[zsxwing]

Javadoc of which? the JDK interfaces? I think the issue was that SQLite didn't actually seem to like being closed twice in practice.

Forgot to read the JIRA at first. Didn't know that SQLite doesn't follow the Javadoc of the JDK interfaces. Sorry

[tedyu]

I compared JDBCRDD.scala with JdbcRDD.scala
From what I can tell according to the usage of java.sql.Connection and java.sql.ResultSet, the proposed change is needed

[tedyu]

@zsxwing @srowen
Anything I need to do for this PR ?

Thanks

[zsxwing]

Looks NextIterator already has a closed field.

[tedyu]

Looks like the case is covered by NextIterator already