Shengkai Fang created FLINK-22118:
-------------------------------------
Summary: Always apply projection push down in blink planner
Key: FLINK-22118
URL: https://issues.apache.org/jira/browse/FLINK-22118
Project: Flink
Issue Type: Improvement
Components: Table SQL / Planner
Affects Versions: 1.13.0
Reporter: Shengkai Fang
Please add the case in `TableSourceTest`.
{code:java}
s"""
|CREATE TABLE NestedItemTable (
| `id` INT,
| `result` ROW<
| `data_arr` ROW<`value` BIGINT> ARRAY,
| `data_map` MAP<STRING, ROW<`value` BIGINT>>>,
| ) WITH (
| 'connector' = 'values',
| 'nested-projection-supported' = 'true',
| 'bounded' = 'true'
| )
|""".stripMargin
util.tableEnv.executeSql(ddl4)
util.verifyExecPlan(
s"""
|SELECT
| `result`.`data_arr`[`id`].`value`,
| `result`.`data_map`['item'].`value`
|FROM NestedItemTable
|""".stripMargin
)
{code}
we can get optimized plan
{code:java}
Calc(select=[ITEM(result.data_arr, id).value AS EXPR$0, ITEM(result.data_map,
_UTF-16LE'item').value AS EXPR$1])
+- TableSourceScan(table=[[default_catalog, default_database,
NestedItemTable]], fields=[id, result])
{code}
but expected is
{code:java}
Calc(select=[ITEM(result_data_arr, id).value AS EXPR$0, ITEM(result_data_map,
_UTF-16LE'item').value AS EXPR$1])
+- TableSourceScan(table=[[default_catalog, default_database, NestedItemTable,
project=[result_data_arr, result_data_map, id]]], fields=[result_data_arr,
result_data_map, id])
{code}
It seems the planner doesn't apply the rule to push projection into scan. The
reason why we have different results is the optimized plan has more fields than
before.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
