[SPARK-17732][SQL] ALTER TABLE DROP PARTITION should support comparators

[dongjoon-hyun]

What changes were proposed in this pull request?

This PR aims to support comparators, e.g. '<', '<=', '>', '>=', again in Apache Spark 2.0 for backward compatibility.

Spark 1.6.2

scala> sql("CREATE TABLE sales(id INT) PARTITIONED BY (country STRING, quarter STRING)")
res0: org.apache.spark.sql.DataFrame= [result: string]
scala> sql("ALTER TABLE sales DROP PARTITION (country < 'KR')")
res1: org.apache.spark.sql.DataFrame= [result: string]

Spark 2.0

scala> sql("CREATE TABLE sales(id INT) PARTITIONED BY (country STRING, quarter STRING)")
res0: org.apache.spark.sql.DataFrame= []
scala> sql("ALTER TABLE sales DROP PARTITION (country < 'KR')")
org.apache.spark.sql.catalyst.parser.ParseException:
mismatched input '<' expecting {')', ','}(line 1, pos 42)

After this PR, it's supported.

How was this patch tested?

Pass the Jenkins test with a newly added testcase.

[SparkQA]

Test build #66123 has finished for PR 15302 at commit c6c52fe.

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

[SparkQA]

Test build #66130 has finished for PR 15302 at commit b59d622.

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

[dongjoon-hyun]

Rebased to the master.

[SparkQA]

Test build #66219 has finished for PR 15302 at commit eca9c86.

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

[dongjoon-hyun]

The only failure is the following in ColumnTypeSuite. It's irrelevant and the testsuite passed locally.

[info] - MAP append/extract *** FAILED *** (2 milliseconds)
[info] java.lang.IllegalArgumentException:
[info] at java.nio.Buffer.position(Buffer.java:244)
[info] at org.apache.spark.sql.catalyst.expressions.UnsafeRow.writeFieldTo(UnsafeRow.java:650)

[dongjoon-hyun]

Retest this please.

[SparkQA]

Test build #66225 has finished for PR 15302 at commit eca9c86.

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

[dongjoon-hyun]

Hi, @hvanhovell .
Could you review this PR about 'ALTER TABLE DROP PARTITION'?

[dongjoon-hyun]

Hi, @hvanhovell .
Could you give some opinion about this PR when you have some time?

[hvanhovell]

@dongjoon-hyun I have taken a quick look. Shouldn't we just use Expressions for filtering partitions?

[dongjoon-hyun]

Thank you for review, @hvanhovell ! Do you mean SQL grammar or listPartition?

[dongjoon-hyun]

Recently, I've watched you improved those related function greatly.

[hvanhovell]

@dongjoon-hyun I think that AlterTableDropPartitionCommand just should take a set of catalyst Expressions instead of a PartitionRangeSpec.

I have added a few things the Catalog but I think that we shouldn't use those here. The `SessionCatalog is much more suited here.

[dongjoon-hyun]

I see. Then, how can evaluate the generic expression? Is it okay to use 'eval(null)'?

[dongjoon-hyun]

Thank you for the direction. I'll proceed to improve in that way.

[SparkQA]

Test build #66276 has finished for PR 15302 at commit 3c0585b.

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

[dongjoon-hyun]

The only failure looks irrelevant. Anyway, I'm revising the PR.

[info] ***1SUITEABORTED***
[error] Error:Total2604, Failed0, Errors1, Passed2603, Ignored48
[error] Error during tests:
[error] org.apache.spark.sql.jdbc.JDBCWriteSuite
[error] (sql/test:test) sbt.TestsFailedException:Tests unsuccessful
[error] Totaltime: 718 s, completed Oct3, 20162:23:41PM

[dongjoon-hyun]

Hi, @hvanhovell .

When using Expression, I faced two situations.

• checkAnalysis raises exceptions because the column is unresolved, e.g., country is unresolved.
• As a workaround, I tried to use string literal 'country', but then optimizer ConstantFolding replaces that as false because 'country' < 'KR' is false.

ALTERTABLE sales DROP PARTITION (country <'KR')

To avoid this situations, I can add some rule to checkAnalysis. But, it seems not a good idea. Could you give some advice for this?

[dongjoon-hyun]

With today's master, it's like the following. Should we use expression in AlterTableDropPartitionCommand?

org.apache.spark.sql.AnalysisException: cannot resolve '`country`'giveninputcolumns: []; line 1 pos 23;
'AlterTableDropPartitionCommand `sales`, [('country<KR)], false, false

[dongjoon-hyun]

Hi, @hvanhovell .
In DDL, do we have an example to use Expression like this?

[dongjoon-hyun]

Hi, @hvanhovell .
I made another attempt #15704 by using 'Expression' as you commented.

[hvanhovell]

@dongjoon-hyun I'll take a look tomorrow.

[dongjoon-hyun]

Thank you, @hvanhovell !

[dongjoon-hyun]

I'm closing this PR in favor of #15704 .