[
https://issues.apache.org/jira/browse/DRILL-928?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14097493#comment-14097493
]
Hanifi Gunes commented on DRILL-928:
------------------------------------
I cannot reprod this at master:687b9b0. It seems fixed.
{code:sql}
alter session set `planner.enable_hashjoin` = false;
select count(*)
from cp.`tpch/supplier.parquet` s, cp.`tpch/lineitem.parquet` l,
cp.`tpch/orders.parquet` o
where s.s_suppkey = l.l_suppkey
and o.o_orderkey = l.l_orderkey
and l.l_shipdate between date '1995-01-01' and date '1995-12-31' ;
{code}
{code:sql}
alter session set `planner.enable_hashjoin` = true;
select count(*)
from cp.`tpch/supplier.parquet` s, cp.`tpch/lineitem.parquet` l,
cp.`tpch/orders.parquet` o
where s.s_suppkey = l.l_suppkey
and o.o_orderkey = l.l_orderkey
and l.l_shipdate between date '1995-01-01' and date '1995-12-31' ;
{code}
Both queries return:
+------------+
| 8773 |
+------------+
> Regression: Forcing a merge join instead of a hash join results in a Failure
> while reading vector. Expected vector class of
> org.apache.drill.exec.vector.BigIntVector but was holding vector class
> org.apache.drill.exec.vector.VarBinaryVector.
> -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: DRILL-928
> URL: https://issues.apache.org/jira/browse/DRILL-928
> Project: Apache Drill
> Issue Type: Bug
> Components: Query Planning & Optimization
> Reporter: Ramana Inukonda Nagaraj
> Assignee: Hanifi Gunes
> Fix For: 0.5.0
>
>
> Test case:
> alter session set `planner.enable_hashjoin` = false;
> select count(*)
> from supplier s, lineitem l, orders o
> where s.s_suppkey = l.l_suppkey
> and o.o_orderkey = l.l_orderkey
> and l.l_shipdate between date '1995-01-01' and date '1995-12-31' ;
> If the alter session is removed the query works fine. Which leads me to
> believe its something in either the mergejoin or the sort before the
> mergejoin step.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
