[IMPALA-2741] Normalise view SQL in same way as Hive

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Duplicate
• Affects Version/s: Impala 2.2, Impala 2.3.0
• Labels: catalog-server impala ramp-up
• Environment: cdh5.4.5+0
• Assignee: Unassigned
• Reporter: Philippe Devault
• Created: 04/Dec/15 14:11
• Updated: 04/Jan/17 23:58
• Resolved: 04/Dec/15 19:42

Description

Hive normalises view SQL in several ways for the VIEW_EXPANDED_TEXT metastore column. The column list, e.g. from select * is expanded and field names are normalised to include the database name and be enclosed in backticks. This does not appear to be documented and is essentially implementation-defined. Some DDL operations in Hive make assumptions about VIEW_EXPANDED_TEXT content and will crash if Impala generates something slightly different. Note that actually using the view in a query appears to work fine.

We should consider modifying CREATE VIEW in Impala so that it generates VIEW_EXPANDED_TEXT that is normalised in a similar way to Hive to avoid some of these compatibility problems.

The original bug report follows:

In Impala
create table myTable (x int , y int);
create view myView AS SELECT * FROM myTable;

in "Hive"
SHOW TABLE EXTENDED FROM default LIKE 'myview'; --> CRASH

To fix the issue we create the view in hive.

We use aquadatastudio and the "Hive" tree browser crash because he try to run this query.
SHOW TABLE EXTENDED FROM <database> LIKE `*`

Not very good option for us to ask all users to create all views only in "Hive" to avoid this crash.

Issue Links

• duplicates: IMPALA-1710 Impala doesn't populate column list in "view expanded text"