[
https://issues.apache.org/jira/browse/SPARK-11580?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yadong Qi updated SPARK-11580:
------------------------------
Description:
I do the SQL as below:
{code}
cache table src as select * from src distribute by key;
select key, count(value) from src group by key;
{code}
and the Physical Plan is
{code}
TungstenAggregate(key=[key#0],
functions=[(count(value#1),mode=Final,isDistinct=false)],
output=[key#0,_c1#28L])
TungstenAggregate(key=[key#0],
functions=[(count(value#1),mode=Partial,isDistinct=false)],
output=[key#0,currentCount#41L])
InMemoryColumnarTableScan [key#0,value#1], (InMemoryRelation [key#0,value#1],
true, 10000, StorageLevel(true, true, false, true, 1), (TungstenExchange
hashpartitioning(key#0)), Some(src))
{code}
I think if there is no *Exchange*, just do final aggregation is better, like:
{code}
TungstenAggregate(key=[key#0],
functions=[(count(value#1),mode=Final,isDistinct=false)],
output=[key#0,_c1#28L])
InMemoryColumnarTableScan [key#0,value#1], (InMemoryRelation [key#0,value#1],
true, 10000, StorageLevel(true, true, false, true, 1), (TungstenExchange
hashpartitioning(key#0)), Some(src))
{code}
was:
I do the SQL as below:
{code}
cache table src as select * from src distribute by key;
select key, count(value) from src group by key;
{code}
and the Physical Plan is
{code}
TungstenAggregate(key=[key#0],
functions=[(count(value#1),mode=Final,isDistinct=false)],
output=[key#0,_c1#28L])
TungstenAggregate(key=[key#0],
functions=[(count(value#1),mode=Partial,isDistinct=false)],
output=[key#0,currentCount#41L])
InMemoryColumnarTableScan [key#0,value#1], (InMemoryRelation [key#0,value#1],
true, 10000, StorageLevel(true, true, false, true, 1), (TungstenExchange
hashpartitioning(key#0)), Some(src))
{code}
I think if there is no Exchange, just do final aggregation is better, like:
{code}
TungstenAggregate(key=[key#0],
functions=[(count(value#1),mode=Final,isDistinct=false)],
output=[key#0,_c1#28L])
InMemoryColumnarTableScan [key#0,value#1], (InMemoryRelation [key#0,value#1],
true, 10000, StorageLevel(true, true, false, true, 1), (TungstenExchange
hashpartitioning(key#0)), Some(src))
{code}
> Just do final aggregation when there is no Exchange
> ---------------------------------------------------
>
> Key: SPARK-11580
> URL: https://issues.apache.org/jira/browse/SPARK-11580
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 1.5.1
> Reporter: Yadong Qi
>
> I do the SQL as below:
> {code}
> cache table src as select * from src distribute by key;
> select key, count(value) from src group by key;
> {code}
> and the Physical Plan is
> {code}
> TungstenAggregate(key=[key#0],
> functions=[(count(value#1),mode=Final,isDistinct=false)],
> output=[key#0,_c1#28L])
> TungstenAggregate(key=[key#0],
> functions=[(count(value#1),mode=Partial,isDistinct=false)],
> output=[key#0,currentCount#41L])
> InMemoryColumnarTableScan [key#0,value#1], (InMemoryRelation
> [key#0,value#1], true, 10000, StorageLevel(true, true, false, true, 1),
> (TungstenExchange hashpartitioning(key#0)), Some(src))
> {code}
> I think if there is no *Exchange*, just do final aggregation is better, like:
> {code}
> TungstenAggregate(key=[key#0],
> functions=[(count(value#1),mode=Final,isDistinct=false)],
> output=[key#0,_c1#28L])
> InMemoryColumnarTableScan [key#0,value#1], (InMemoryRelation
> [key#0,value#1], true, 10000, StorageLevel(true, true, false, true, 1),
> (TungstenExchange hashpartitioning(key#0)), Some(src))
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
