[SPARK-22329] Use NEVER_INFER for `spark.sql.hive.caseSensitiveInferenceMode` by default

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Won't Fix
• Affects Version/s: 2.2.0
• Component/s: SQL
• Assignee: Unassigned
• Reporter: Dongjoon Hyun
• Created: 22/Oct/17 17:57
• Updated: 02/Nov/17 15:41
• Resolved: 02/Nov/17 15:41

Description

In Spark 2.2.0, `spark.sql.hive.caseSensitiveInferenceMode` has a critical issue.

• SPARK-19611 uses `INFER_AND_SAVE` at 2.2.0 since Spark 2.1.0 breaks some Hive tables backed by case-sensitive data files.

  This situation will occur for any Hive table that wasn't created by Spark or that was created prior to Spark 2.1.0. If a user attempts to run a query over such a table containing a case-sensitive field name in the query projection or in the query filter, the query will return 0 results in every case.

• However, SPARK-22306 reports this also corrupts Hive Metastore schema by removing bucketing information (BUCKETING_COLS, SORT_COLS) and changing owner.

• Since Spark 2.3.0 supports Bucketing, BUCKETING_COLS and SORT_COLS look okay at least. However, we need to figure out the issue of changing owners. Also, we cannot backport bucketing patch into `branch-2.2`. We need more tests on before releasing 2.3.0.

Hive Metastore is a shared resource and Spark should not corrupt it by default. This issue proposes to recover that option back to `NEVER_INFO` like Spark 2.2.0 by default. Users can take a risk by enabling `INFER_AND_SAVE` by themselves.

Issue Links

• is superceded by: SPARK-22306 INFER_AND_SAVE overwrites important metadata in Parquet Metastore table
• relates to: SPARK-22306 INFER_AND_SAVE overwrites important metadata in Parquet Metastore table
• relates to: SPARK-19611 Spark 2.1.0 breaks some Hive tables backed by case-sensitive data files
• links to: [Github] Pull Request #19552 (dongjoon-hyun)