[IMPALA-1862] Invalid bool value not reported as a scanner error

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: Impala 2.0
• Fix Version/s: Impala 2.8.0
• Component/s: Backend
• Labels: codegen
• Target Version: Impala 2.8.0
• Assignee: Sailesh Mukil
• Reporter: Matthew Jacobs
• Created: 09/Mar/15 22:51
• Updated: 08/Dec/16 18:39
• Resolved: 08/Dec/16 18:39

Description

In some cases, reading a bad boolean value does not report a scanner error though the value is NULL, as expected for an invalid value. This is only broken with codegen enabled.

Here is a query including the tinyint_col so that the row data is shown

[localhost:21000] > select bool_col, tinyint_col from functional.alltypeserror where id = 30;
Query: select bool_col, tinyint_col from functional.alltypeserror where id = 30
+----------+-------------+
| bool_col | tinyint_col |
+----------+-------------+
| NULL     | NULL        |
+----------+-------------+

ERRORS ENCOUNTERED DURING EXECUTION: 
Backend 1:Error converting column: 2 TO TINYINT (Data is: err30)
file: hdfs://localhost:20500/test-warehouse/alltypeserror/year=2009/month=3/0903.txt
record: 30,t\rue,err30,err30,err30,err300,err30..000000,err300.900000,01/01/10,10,0000-01-01 00:00:00

Returned 1 row(s) in 0.67s

now if only the bool_col is selected there is no error, but there is no valid boolean value in the row data.

[localhost:21000] > select bool_col from functional.alltypeserror where id = 30;
Query: select bool_col from functional.alltypeserror where id = 30
+----------+
| bool_col |
+----------+
| NULL     |
+----------+
Returned 1 row(s) in 0.63s

In both cases, there should be an error reported:

'Error converting column: 1 TO BOOL (Data is: t\rue)'

Issue Links

• is related to: IMPALA-4594 WriteSlot and CodegenWriteSlot handle escaped NULL slots differently