Presto: Join triggers "Nulls fraction should be within [0, 1]" error on Hive/Parquet tables

[talnicolas]

Hi, I am trying to run a simple join between two tables like so:

SELECT * FROM table_one A JOIN table_two B ON A.USER_KEY=B.USER_KEY ;

and this always result in:

Query [...] failed: Nulls fraction should be within [0, 1] or NaN, got: -1.606566858743409E-9

I am able to run a select * separately successfully on both tables (both relying on Parquet files). Couldn't find anything in existing issues or google, does anybody know where this could come from?

Thanks!

[rschlussel]

cc: @arhimondr @highker

@talnicolas: Thanks for reporting this error! What version of Presto are you running? Can you show us the output of SHOW STATS for each of the two tables you're joining?

[talnicolas]

Presto version: 0.203-T.0.3

For the stats:

presto:some_db> show stats for (select * from table_one);
         column_name         | data_size | distinct_values_count | nulls_fraction |       row_count       | low_value | high_value 
-----------------------------+-----------+-----------------------+----------------+-----------------------+-----------+------------
 col_a                       | NULL      | NULL                  | NULL           | NULL                  | NULL      | NULL       
 user_key                    | NULL      | NULL                  | NULL           | NULL                  | NULL      | NULL       
 some_other_key              | NULL      |                3543.0 |            0.0 | NULL                  | 20090119  | 20181001   
 NULL                        | NULL      | NULL                  | NULL           | 1.1015643838588234E10 | NULL      | NULL

presto:some_db> show stats for (select * from table_two);
             column_name              |       data_size       | distinct_values_count |    nulls_fraction     |  row_count   | low_value | high_value 
--------------------------------------+-----------------------+-----------------------+-----------------------+--------------+-----------+------------
 col_a                                | NULL                  |                   1.0 | -1.606566858743409E-9 | NULL         | NULL      | NULL       
 col_b                                |  1.4012488751183999E9 |                 137.0 | -1.606566858743409E-9 | NULL         | NULL      | NULL       
 col_c                                | 2.0053009232965504E10 |          6.13331008E8 | -1.606566858743409E-9 | NULL         | NULL      | NULL       
 user_key                             | NULL                  |          6.11070144E8 | -1.606566858743409E-9 | NULL         | NULL      | NULL       
 NULL                                 | NULL                  | NULL                  | NULL                  | 6.22445306E8 | NULL      | NULL

Hope this helps.

Thanks.

[arhimondr]

There were some bugs in the cost estimation code. Try updating the Presto version and see if the issue reoccurs.