Aman Sinha created DRILL-885:
--------------------------------

             Summary: Handle project pushdown for constant expressions
                 Key: DRILL-885
                 URL: https://issues.apache.org/jira/browse/DRILL-885
             Project: Apache Drill
          Issue Type: Bug
            Reporter: Aman Sinha


In the following query, notice in the Explain plan that the node  
Project($f0=[0])  is projecting a constant, so ideally we should not have to 
produce a whole bunch of columns from either side of the join unless those 
columns are needed for the join condition.  However, currently we do produce 
those unnecessary columns from the Scan below (see the Customer parquet scan on 
the left side of the HashJoin).   This hurts performance.

0: jdbc:drill:zk=local> explain plan for select count(*) from (select 
c.c_custkey, c.c_name, c.c_address, c.c_nationkey,  c.c_phone, c.c_acctbal, 
c.c_mktsegment, c.c_comment, n.n_nationkey, n.n_name, n.n_nationkey, 
n.n_comment from cp.`tpch/customer.parquet` c JOIN cp.`tpch/nation.parquet` n 
ON (c.c_nationkey = n.n_nationkey));
+------------+------------+
|    text    |    json    |
+------------+------------+
| 00-00    Screen
00-01      StreamAgg(group=[{}], EXPR$0=[SUM($0)])
00-02        UnionExchange
01-01          StreamAgg(group=[{}], EXPR$0=[COUNT()])
01-02            Project($f0=[0])
01-03              HashJoin(condition=[=($1, $10)], joinType=[inner])
01-05                HashToRandomExchange(dist0=[[$1]])
02-01                  Scan(groupscan=[ParquetGroupScan 
[entries=[ReadEntryWithPath [path=/tpch/customer.parquet]], 
selectionRoot=/tpch/customer.parquet, columns=[SchemaPath [`c_nationkey`], 
SchemaPath [`c_custkey`], SchemaPath [`c_name`], SchemaPath [`c_address`], 
SchemaPath [`c_phone`], SchemaPath [`c_acctbal`], SchemaPath [`c_mktsegment`], 
SchemaPath [`c_comment`]]]])
01-04                Project(*0=[$0], n_nationkey=[$1], n_name=[$2], 
n_comment=[$3])
01-06                  HashToRandomExchange(dist0=[[$1]])
03-01                    Scan(groupscan=[ParquetGroupScan 
[entries=[ReadEntryWithPath [path=/tpch/nation.parquet]], 
selectionRoot=/tpch/nation.parquet, columns=[SchemaPath [`n_nationkey`], 
SchemaPath [`n_name`], SchemaPath [`n_comment`]]]])

Here's the Drill Logical plan for the same query:
| DrillScreenRel
  DrillAggregateRel(group=[{}], EXPR$0=[COUNT()])
    DrillProjectRel($f0=[0])
      DrillJoinRel(condition=[=($1, $10)], joinType=[inner])
        DrillScanRel(table=[[cp, tpch/customer.parquet]])
        DrillScanRel(table=[[cp, tpch/nation.parquet]])



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to