Fred Teunissen created FLINK-39247:
--------------------------------------
Summary: ClassCastException when executing INSERT INTO with nested
ROW containing NOT NULL MAP in Flink 2.2.0
Key: FLINK-39247
URL: https://issues.apache.org/jira/browse/FLINK-39247
Project: Flink
Issue Type: Bug
Components: Table SQL / Planner, Table SQL / Runtime
Affects Versions: 2.2.0
Reporter: Fred Teunissen
*Overview*
A breaking change has been introduced in Flink 2.2.0 where executing an INSERT
INTO statement with nested ROW types containing NOT NULL MAP fields fails with
a ClassCastException. The exact same SQL statements execute without issues in
Flink 1.20.3 and 2.1.1.
*Affected Scenario*
INSERT INTO queries that reference deeply nested ROW structures with NOT NULL
constraints on MAP fields and use MAP_ENTRIES() with UNNEST fail in Flink 2.2.0.
*Failing SQL Statement*
_Table Definitions_
{code:sql}
CREATE TABLE orders (
outerRow ROW<
innerRow ROW<
headers MAP<STRING NOT NULL, STRING NOT NULL> NOT NULL
>
> NOT NULL,
ingestionTime TIMESTAMP(3),
WATERMARK FOR ingestionTime AS ingestionTime
) WITH (
'connector' = 'datagen',
'rows-per-second' = '2',
'number-of-rows' = '10'
);
CREATE TABLE blackhole_table (key STRING) WITH ('connector' = 'blackhole');
{code}
*Query (Works in 1.20.3 and 2.1.1, fails in 2.2.0)*
{code:sql}
INSERT INTO blackhole_table
SELECT
unnestedEntry.key
FROM orders
CROSS JOIN UNNEST (MAP_ENTRIES(orders.outerRow.innerRow.headers)) as
unnestedEntry;
{code}
*Error Message*
{code:none}
[ERROR] Could not execute SQL statement. Reason:
java.lang.ClassCastException: class org.apache.calcite.rex.RexCall cannot be
cast to class org.apache.calcite.rex.RexFieldAccess
(org.apache.calcite.rex.RexCall and org.apache.calcite.rex.RexFieldAccess
are in unnamed module of loader
org.apache.flink.table.planner.loader.PlannerModule$PlannerComponentClassLoader
@42bc14c1)
{code}
*Expected Behavior*
The INSERT INTO statement should execute successfully, accessing the nested
field orders.outerRow.innerRow.headers and extracting keys from the MAP via
MAP_ENTRIES() and UNNEST. The query should produce output without errors.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
