[SPARK-29295] Duplicate result when dropping partition of an external table and then overwriting

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.4.4
• Fix Version/s: 3.0.0
• Component/s: SQL
• Assignee: L. C. Hsieh
• Reporter: feiwang
• Created: 30/Sep/19 01:19
• Updated: 12/Nov/19 10:32
• Resolved: 18/Oct/19 08:37

Description

When we drop a partition of a external table and then overwrite it, if we set CONVERT_METASTORE_PARQUET=true(default value), it will overwrite this partition.
But when we set CONVERT_METASTORE_PARQUET=false, it will give duplicate result.

Here is a reproduce code below(you can add it into SQLQuerySuite in hive module):

  test("spark gives duplicate result when dropping a partition of an external partitioned table" +
    " firstly and they overwrite it") {
    withTable("test") {
      withTempDir { f =>
        sql("create external table test(id int) partitioned by (name string) stored as " +
          s"parquet location '${f.getAbsolutePath}'")

        withSQLConf(HiveUtils.CONVERT_METASTORE_PARQUET.key -> false.toString) {
          sql("insert overwrite table test partition(name='n1') select 1")
          sql("ALTER TABLE test DROP PARTITION(name='n1')")
          sql("insert overwrite table test partition(name='n1') select 2")
          checkAnswer( sql("select id from test where name = 'n1' order by id"),
            Array(Row(1), Row(2)))
        }

        withSQLConf(HiveUtils.CONVERT_METASTORE_PARQUET.key -> true.toString) {
          sql("insert overwrite table test partition(name='n1') select 1")
          sql("ALTER TABLE test DROP PARTITION(name='n1')")
          sql("insert overwrite table test partition(name='n1') select 2")
          checkAnswer( sql("select id from test where name = 'n1' order by id"),
            Array(Row(2)))
        }
      }
    }
  }

Issue Links

• is related to: HIVE-18702 INSERT OVERWRITE TABLE doesn't clean the table directory before overwriting
• relates to: SPARK-25271 Creating parquet table with all the column null throws exception
• links to: GitHub Pull Request #25979