Table Pushdown Optimizer Rule
openLooKeng supports Table Pushdown optimizer rule to improve latency for eligible queries.
The user must execute the following command to generate stats for all the tables which are part of the SQL query so that the join sources are reordered based on the size of the tables-
User Hint Comment-
The user may also add a special comment in the following format to leverage the benefit from the Table Pushdown rule by specifying the table names and their distinct(primary key) column names.
/* #distinct@ table1 = col1, col2, ... #*/
This hint is optional as openLookeng can identify all distinct columns if the relevant statistics are available for the table.
The following query (modified from Query 17 of TPC-H benchmark) is one such eligible query and pushes the outer table into the subquery and improves overall query latency.
SELECT Sum(lineitem.extendedprice) / 7.0 AS avg_yearly FROM lineitem, part, ( SELECT 0.2 * Avg(lineitem.quantity) AS s_avg, lineitem.partkey AS s_partkey FROM lineitem GROUP BY lineitem.partkey ) WHERE part.partkey = lineitem.partkey AND part.brand = 'Brand#43' AND part.container = 'LG PACK' AND part.partkey = s_partkey AND lineitem.quantity < s_avg /* #distinct@ part = partkey #*/;
In the above query, table
part is the relevant outer query table with
partkey as the unique column, being joined with table
lineitem. An equivalent rewritten query, with
part pushed into the subquery is given as follows.
Equivalent Rewritten Query-
SELECT Sum(lineitem.extendedprice) / 7.0 AS avg_yearly FROM lineitem, ( SELECT 0.2 * Avg(lineitem.quantity) AS s_avg, lineitem.partkey AS s_partkey FROM lineitem, part WHERE part.brand = 'Brand#43' AND part.container = 'LG PACK' AND part.partkey = lineitem.partkey GROUP BY lineitem.partkey ) WHERE s_partkey = lineitem.partkey AND lineitem.quantity < s_avg;
The feature is disabled by default. The user can enable it by executing the following command to set the session parameter-
SET SESSION push_table_through_subquery = true;