You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 13 Next »

Document the state by adding a label to the FLIP page with one of "discussion", "accepted", "released", "rejected".

Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).

Motivation

Currently, Flink Table/SQL does not expose fine-grained control for users to enable or disable filter pushdown. However, filter pushdown has some side effects, such as additional computational pressure on external systems. Moreover, Improper queries can lead to issues such as full table scans, which in turn can impact the stability of external systems.

Different with table.optimizer.source.predicate-pushdown-enabled

The existing configuration "table.optimizer.source.predicate-pushdown-enabled" was introduced by legacy FilterableTableSource interface which allows the optimizer to disable all predicate pushdown, but it cannot provide fine-grained control for each individual source.

Suppose we have an SQL query with two sources: Kafka and a database. The database is sensitive to pressure, and we want to configure it to not perform filter pushdown to the database source. However, we still want to perform filter pushdown to the Kafka source to decrease network IO.

Considering compatibility with older versions, the newly added option needs to respect the optimizer's "table.optimizer.source.predicate-pushdown-enabled" option.  But maybe we can deprecate "table.optimizer.source.predicate-pushdown-enabled" option and drop it in Flink 2.0 if it is not necessary anymore.

Goals

Support fine-grained configuration to disable filter push down for each Table/SQL sources to let user decide whether to perform filter pushdown.

Public Interfaces

The user specifies option for sources through a new connector option:

OptionTypeDefault value
source.filter-push-down.enabledBooleantrue


For other  source abilities, we can also consider adding similar configuration options in the future to give users the choice. This is not urgent and can be continuously introduced.

OptionTypeDefault value
source.aggregate-push-down.enabledBooleantrue
...Booleantrue

Proposed Changes

For sources that need to support disabling filter pushdown, it is necessary to check this configuration in the applyFilters method to determine whether to perform predicate pushdown.

Compatibility, Deprecation, and Migration Plan

In the implementation of the applyFilters method, source needs to follow the following pattern to respond to the configuration option.

@Override
public Result applyFilters(List<ResolvedExpression> filters) {
    if (config.get(ENABLE_FILTER_PUSH_DOWN)) {
        return applyFiltersInternal(filters);
    } else {
        return Result.of(Collections.emptyList(), filters);
    }
}

Test Plan

The changes will be covered by UTs.

Rejected Alternatives

Add default method enableFilterPushDown in SupportsFilterPushDown interface. In FilterPushDownSpec, determine whether filter pushdown is allowed. If it is not allowed, the applyFilters logic will not be executed.

@PublicEvolving
public interface SupportsFilterPushDown {

    /**
     * Provides a list of filters in conjunctive form. A source can pick filters and return the
     * accepted and remaining filters.
     *
     * <p>See the documentation of {@link SupportsFilterPushDown} for more information.
     */
    Result applyFilters(List<ResolvedExpression> filters);


    /**
     * Returns whether filter push down is enabled.
     */
    default boolean enableFilterPushDown() {
        return true;
    }
}

Reject Reason

The current interface can already satisfy the requirements. The connector can reject all the filters by returning the input filters as `Result#remainingFilters`.


  • No labels