[
https://issues.apache.org/jira/browse/FLINK-41261?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Rion Williams updated FLINK-41261:
----------------------------------
Description:
FLINK-41260 rejects ARRAY_AGG(x ORDER BY y) and JSON_ARRAY_AGG(x ORDER BY y)
during validation, because the runtime ignored the ordering. Postgres, BigQuery
and Snowflake support the clause.
Supporting it needs:
* the call to resolve to Flink's ARRAY_AGG with the ORDER BY kept as the
AggregateCall collation, instead of keeping Calcite's
SqlLibraryOperators.ARRAY_AGG (whose ORDERED_FUNCTION syntax SymbolUtil cannot
serialize);
* the collation in the compiled plan, since AggregateCallJsonSerializer does
not write it today, with a restore test;
* an accumulator that keeps the sort keys and orders the elements on getValue,
with retraction support for streaming. Today the accumulator holds only the
values (the reason given in FLINK-21949).
** All of the above is also applicable to the JSON_ARRAY_AGG function.
was:
FLINK-41260 rejects ARRAY_AGG(x ORDER BY y) during validation, because the
runtime ignored the ordering. Postgres, BigQuery and Snowflake support the
clause.
Supporting it needs:
* the call to resolve to Flink's ARRAY_AGG with the ORDER BY kept as the
AggregateCall collation, instead of keeping Calcite's
SqlLibraryOperators.ARRAY_AGG (whose ORDERED_FUNCTION syntax SymbolUtil cannot
serialize);
* the collation in the compiled plan, since AggregateCallJsonSerializer does
not write it today, with a restore test;
* an accumulator that keeps the sort keys and orders the elements on getValue,
with retraction support for streaming. Today the accumulator holds only the
values (the reason given in FLINK-21949).
> Support ORDER BY in ARRAY_AGG/JSON_ARRAY_AGG
> --------------------------------------------
>
> Key: FLINK-41261
> URL: https://issues.apache.org/jira/browse/FLINK-41261
> Project: Flink
> Issue Type: New Feature
> Components: Table SQL / Planner
> Reporter: Rion Williams
> Priority: Major
>
> FLINK-41260 rejects ARRAY_AGG(x ORDER BY y) and JSON_ARRAY_AGG(x ORDER BY y)
> during validation, because the runtime ignored the ordering. Postgres,
> BigQuery and Snowflake support the clause.
> Supporting it needs:
> * the call to resolve to Flink's ARRAY_AGG with the ORDER BY kept as the
> AggregateCall collation, instead of keeping Calcite's
> SqlLibraryOperators.ARRAY_AGG (whose ORDERED_FUNCTION syntax SymbolUtil
> cannot serialize);
> * the collation in the compiled plan, since AggregateCallJsonSerializer does
> not write it today, with a restore test;
> * an accumulator that keeps the sort keys and orders the elements on
> getValue, with retraction support for streaming. Today the accumulator holds
> only the values (the reason given in FLINK-21949).
> ** All of the above is also applicable to the JSON_ARRAY_AGG function.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
