Baike Xia created IMPALA-11424:
----------------------------------
Summary: Support pushdown non-equi join predicate from OUTER/INNER
JOIN to SCANNODE
Key: IMPALA-11424
URL: https://issues.apache.org/jira/browse/IMPALA-11424
Project: IMPALA
Issue Type: Improvement
Components: Frontend
Reporter: Baike Xia
Fix For: Impala 4.1.0
In order to reduce the amount of data read and transmitted, the non-equivalent
condition of Join can be pushed to SCAN_NODE.
For example:
{code:java}
// code placeholder
select count(1) from ( select ss.ss_ticket_numberFROM store_sales ss LEFT
OUTER JOIN store_returns sr ON (sr.sr_item_sk = ss.ss_item_sk AND
sr.sr_ticket_number >= ss.ss_ticket_number) where ss.ss_sold_date_sk = 2450816)
t where t.ss_ticket_number = 79577; {code}
Current plan:
{code:java}
// code placeholder
PLAN-ROOT SINK
|
07:AGGREGATE [FINALIZE]
| output: count:merge(*)
| row-size=8B cardinality=1
|
06:EXCHANGE [UNPARTITIONED]
|
03:AGGREGATE
| output: count(*)
| row-size=8B cardinality=1
|
02:HASH JOIN [RIGHT OUTER JOIN, PARTITIONED]
| hash predicates: sr.sr_item_sk = ss.ss_item_sk
| other join predicates: sr.sr_ticket_number >= ss.ss_ticket_number
| runtime filters: RF000 <- ss.ss_item_sk
| row-size=32B cardinality=16
|
|--05:EXCHANGE [HASH(ss.ss_item_sk)]
| |
| 00:SCAN HDFS [tpcds_parquet.store_sales ss]
| partition predicates: ss.ss_sold_date_sk = 2450816
| partitions=1/1824 files=1 size=70.77KB
| predicates: ss.ss_ticket_number = 79577
| row-size=16B cardinality=1
|
04:EXCHANGE [HASH(sr.sr_item_sk)]
|
01:SCAN HDFS [tpcds_parquet.store_returns sr]
partitions=1/1 files=1 size=15.42MB
runtime filters: RF000 -> sr.sr_item_sk
row-size=16B cardinality=287.51K{code}
After Pushdown:
{code:java}
// code placeholder
PLAN-ROOT SINK
|
07:AGGREGATE [FINALIZE]
| output: count:merge(*)
| row-size=8B cardinality=1
|
06:EXCHANGE [UNPARTITIONED]
|
03:AGGREGATE
| output: count(*)
| row-size=8B cardinality=1
|
02:HASH JOIN [RIGHT OUTER JOIN, PARTITIONED]
| hash predicates: sr.sr_item_sk = ss.ss_item_sk
| other join predicates: sr.sr_ticket_number >= ss.ss_ticket_number
| runtime filters: RF000 <- ss.ss_item_sk
| row-size=32B cardinality=16
|
|--05:EXCHANGE [HASH(ss.ss_item_sk)]
| |
| 00:SCAN HDFS [tpcds_parquet.store_sales ss]
| partition predicates: ss.ss_sold_date_sk = 2450816
| partitions=1/1824 files=1 size=70.77KB
| predicates: ss.ss_ticket_number = 79577
| row-size=16B cardinality=1
|
04:EXCHANGE [HASH(sr.sr_item_sk)]
|
01:SCAN HDFS [tpcds_parquet.store_returns sr]
partitions=1/1 files=1 size=15.42MB
predicates: sr.sr_ticket_number >= 79577
runtime filters: RF000 -> sr.sr_item_sk
row-size=16B cardinality=28.75K {code}
For pushdown of Join non-equi conjuncts, the current qualifications:
1. Only support LEFT_OUTER_JOIN, RIGHT_OUTER_JOIN, INNER_JOIN;
2. Only valid for non-equi predicates containing literalExpr,
for example: slot >= Literal, slot in Literal list;
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]