Jesus Camacho Rodriguez created HIVE-14266:
----------------------------------------------
Summary: Partial stats in Join operator may lead to data size
estimate of 0
Key: HIVE-14266
URL: https://issues.apache.org/jira/browse/HIVE-14266
Project: Hive
Issue Type: Bug
Components: Statistics
Reporter: Nita Dembla
Assignee: Jesus Camacho Rodriguez
For some tables, we might not have the column stats available. However, if the
table is partitioned, we will have the stats for partition columns.
When we estimate the size of the data produced by a join operator, we end up
using only the columns that are available for the calculation e.g. partition
columns in this case.
However, even in these cases, we should add the data size for those columns for
which we do not have stats (_default size for the column type x estimated
number of rows_).
To reproduce, the following example can be used:
{noformat}
create table sample_partitioned (x int) partitioned by (y int);
insert into sample_partitioned partition(y=1) values (1),(2);
create temporary table sample as select * from sample_partitioned;
analyze table sample compute statistics for columns;
explain select sample_partitioned.x from sample_partitioned, sample where
sample.y = sample_partitioned.y;
{noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
