Zoltán Borók-Nagy created IMPALA-15430:
------------------------------------------
Summary: Enable ENABLE_OUTER_JOIN_TO_INNER_TRANSFORMATION by
default
Key: IMPALA-15430
URL: https://issues.apache.org/jira/browse/IMPALA-15430
Project: IMPALA
Issue Type: Improvement
Components: Frontend
Reporter: Zoltán Borók-Nagy
IMPALA-5022 (2020) added outer join simplification. When a WHERE predicate
rejects NULLs
coming from the nullable side of an outer join, the join is turned into an
inner join, or
into a narrower outer join. The option has been off by default since then
(common/thrift/Query.thrift:540):
{code}
114: optional bool enable_outer_join_to_inner_transformation = false;
{code}
Without it, queries such as TPC-DS q93 keep the large fact table on the
preserved side of
a LEFT OUTER JOIN, so no runtime filter can reach its scan:
{code:sql}
... FROM store_sales LEFT OUTER JOIN store_returns
ON (sr_item_sk = ss_item_sk AND sr_ticket_number = ss_ticket_number),
reason
WHERE sr_reason_sk = r_reason_sk AND r_reason_desc = 'reason 28'
{code}
In the checked-in plan (PlannerTest/tpcds/tpcds-q93.test), the store_sales scan
reads all
2.88M rows and has no runtime filters. {{sr_reason_sk = r_reason_sk}} rejects
NULLs, so
the rewrite turns the join into an inner join. store_returns, already filtered
by the
reason join, can then send filters to store_sales.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
