[IMPALA-1566] impala-server crashes with cast on a empty regexp_replace result

• Type: Bug
• Status: Resolved
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: Impala 2.0
• Fix Version/s: Impala 2.2
• Labels: cast crash regexp_replace
• Assignee: Victor Bittorf
• Reporter: gpolaert
• Created: 03/Dec/14 10:43
• Updated: 21/Mar/15 22:45
• Resolved: 11/Dec/14 19:41

Description

On version CDH-5.2, Impala 2.0,

I'm trying to replace french decimal values like "12,54" (string) to 12.54 (decimal(15,2)) from big CSV files loaded into a table.

All impala-server died during the insert or create as select statement.
I've isolated a bug when I cast some '' (empty string) into decimal after a call to regexp_replace UDF.

To reproduce the bug:

#### OK
Query: select cast('' as decimal(15,2))
+---------------------------+
| cast('' as decimal(15,2)) |
+---------------------------+
| 0.00                      |
+---------------------------+

#### CRASH
Query: select cast(regexp_replace('','a','b') as decimal(15,2))
Error communicating with impalad: TSocket read 0 bytes

#### WORKAROUND (replace one of the '' by your field)
#### like: select cast(if( FIELD = '', NULL, regexp_replace(FIELD, ',', '.')) as decimal(15,2));
Query: select cast(if('' = '', NULL, regexp_replace('', ',', '.')) as decimal(15,2))
+------------------------------------------------------------------------+
| cast(if('' = '', null, regexp_replace('', ',', '.')) as decimal(15,2)) |
+------------------------------------------------------------------------+
| NULL                                                                   |
+------------------------------------------------------------------------+