godfreyhe commented on a change in pull request #12966: URL: https://github.com/apache/flink/pull/12966#discussion_r467752120
########## File path: flink-table/flink-table-planner-blink/src/main/java/org/apache/flink/table/planner/plan/rules/logical/PushPartitionIntoTableSourceScanRule.java ########## @@ -0,0 +1,325 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.flink.table.planner.plan.rules.logical; + +import org.apache.flink.table.api.TableException; +import org.apache.flink.table.catalog.Catalog; +import org.apache.flink.table.catalog.CatalogPartitionSpec; +import org.apache.flink.table.catalog.CatalogTable; +import org.apache.flink.table.catalog.ObjectIdentifier; +import org.apache.flink.table.catalog.ObjectPath; +import org.apache.flink.table.catalog.exceptions.CatalogException; +import org.apache.flink.table.catalog.exceptions.PartitionNotExistException; +import org.apache.flink.table.catalog.exceptions.TableNotExistException; +import org.apache.flink.table.catalog.exceptions.TableNotPartitionedException; +import org.apache.flink.table.catalog.stats.CatalogColumnStatistics; +import org.apache.flink.table.catalog.stats.CatalogTableStatistics; +import org.apache.flink.table.connector.source.DynamicTableSource; +import org.apache.flink.table.connector.source.abilities.SupportsPartitionPushDown; +import org.apache.flink.table.expressions.Expression; +import org.apache.flink.table.expressions.ResolvedExpression; +import org.apache.flink.table.plan.stats.TableStats; +import org.apache.flink.table.planner.calcite.FlinkContext; +import org.apache.flink.table.planner.calcite.FlinkTypeFactory; +import org.apache.flink.table.planner.plan.schema.TableSourceTable; +import org.apache.flink.table.planner.plan.stats.FlinkStatistic; +import org.apache.flink.table.planner.plan.utils.FlinkRelOptUtil; +import org.apache.flink.table.planner.plan.utils.PartitionPruner; +import org.apache.flink.table.planner.plan.utils.RexNodeExtractor; +import org.apache.flink.table.planner.plan.utils.RexNodeToExpressionConverter; +import org.apache.flink.table.planner.utils.CatalogTableStatisticsConverter; +import org.apache.flink.table.types.logical.LogicalType; + +import org.apache.calcite.plan.RelOptRule; +import org.apache.calcite.plan.RelOptRuleCall; +import org.apache.calcite.rel.core.Filter; +import org.apache.calcite.rel.logical.LogicalTableScan; +import org.apache.calcite.rel.type.RelDataType; +import org.apache.calcite.rex.RexBuilder; +import org.apache.calcite.rex.RexInputRef; +import org.apache.calcite.rex.RexNode; +import org.apache.calcite.rex.RexShuttle; +import org.apache.calcite.rex.RexUtil; +import org.apache.calcite.tools.RelBuilder; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.List; +import java.util.Map; +import java.util.Optional; +import java.util.TimeZone; +import java.util.function.Function; +import java.util.stream.Collectors; + +import scala.Option; +import scala.Tuple2; +import scala.collection.JavaConversions; +import scala.collection.Seq; + +/** + * Planner rule that tries to push partition evaluated by filter condition into a {@link LogicalTableScan}. +*/ +public class PushPartitionIntoTableSourceScanRule extends RelOptRule { + public static final PushPartitionIntoTableSourceScanRule INSTANCE = new PushPartitionIntoTableSourceScanRule(); + + public PushPartitionIntoTableSourceScanRule(){ + super(operand(Filter.class, + operand(LogicalTableScan.class, none())), + "PushPartitionTableSourceScanRule"); + } + + @Override + public boolean matches(RelOptRuleCall call) { + Filter filter = call.rel(0); + if (filter.getCondition() == null) { + return false; + } + TableSourceTable tableSourceTable = call.rel(1).getTable().unwrap(TableSourceTable.class); + if (tableSourceTable == null){ + return false; + } + DynamicTableSource dynamicTableSource = tableSourceTable.tableSource(); + if (!(dynamicTableSource instanceof SupportsPartitionPushDown)) { + return false; + } + CatalogTable catalogTable = tableSourceTable.catalogTable(); + if (!catalogTable.isPartitioned() || catalogTable.getPartitionKeys().isEmpty()) { + return false; + } + return Arrays.stream(tableSourceTable.extraDigests()).noneMatch(digest -> digest.startsWith("partitions=[")); + } + + @Override + public void onMatch(RelOptRuleCall call) { + Filter filter = call.rel(0); + LogicalTableScan scan = call.rel(1); + FlinkContext context = call.getPlanner().getContext().unwrap(FlinkContext.class); + TableSourceTable tableSourceTable = scan.getTable().unwrap(TableSourceTable.class); + + // build pruner + RelDataType inputFieldTypes = filter.getInput().getRowType(); + List<String> inputFieldNames = inputFieldTypes.getFieldNames(); + List<String> partitionFieldNames = tableSourceTable.catalogTable().getPartitionKeys(); + RelBuilder relBuilder = call.builder(); + RexBuilder rexBuilder = relBuilder.getRexBuilder(); + Tuple2<Seq<RexNode>, Seq<RexNode>> allPredicates = RexNodeExtractor.extractPartitionPredicateList( + filter.getCondition(), + FlinkRelOptUtil.getMaxCnfNodeCount(scan), + inputFieldNames.toArray(new String[0]), + rexBuilder, + partitionFieldNames.toArray(new String[0]) + ); + RexNode partitionPredicate = RexUtil.composeConjunction(rexBuilder, JavaConversions.seqAsJavaList(allPredicates._1)); + + if (partitionPredicate.isAlwaysTrue()){ + return; + } + + List<LogicalType> partitionFieldTypes = partitionFieldNames.stream().map(name -> { + int index = inputFieldNames.indexOf(name); + if (index < 0) { + throw new TableException(String.format("Partitioned key '%s' isn't found in input columns. " + + "Validator should have checked that.", name)); + } return inputFieldTypes.getFieldList().get(index).getType(); }) + .map(FlinkTypeFactory::toLogicalType).collect(Collectors.toList()); + + RexNode finalPartitionPredicate = adjustPartitionPredicate(inputFieldNames, partitionFieldNames, partitionPredicate); + Function<List<Map<String, String>>, List<Map<String, String>>> defaultPruner = partitions -> PartitionPruner.prunePartitions( + context.getTableConfig(), + partitionFieldNames.toArray(new String[0]), + partitionFieldTypes.toArray(new LogicalType[0]), + partitions, + finalPartitionPredicate); + + // get partitions from table/catalog and prune + Optional<Catalog> catalogOptional = context.getCatalogManager().getCatalog(tableSourceTable.tableIdentifier().getCatalogName()); + List<Map<String, String>> remainingPartitions = null; + Optional<List<Map<String, String>>> optionalPartitions; + // fields to read partitions from catalog and build new statistic + DynamicTableSource dynamicTableSource = tableSourceTable.tableSource().copy(); + ObjectIdentifier identifier = tableSourceTable.tableIdentifier(); + ObjectPath tablePath = identifier.toObjectPath(); + try { + optionalPartitions = ((SupportsPartitionPushDown) dynamicTableSource).listPartitions(); + if (optionalPartitions.isPresent() && !optionalPartitions.get().isEmpty()) { + remainingPartitions = defaultPruner.apply(optionalPartitions.get()); + } + } catch (UnsupportedOperationException e) { + // check catalog whether is available + // we will read partitions from catalog if table doesn't support listPartitions. + if (!catalogOptional.isPresent()){ + throw new TableException( + String.format("Table %s must from a catalog, but %s is not a catalog", + identifier.asSummaryString(), identifier.getCatalogName()), e); + } + try { + optionalPartitions = readPartitionFromCatalogAndPrune( + context, + catalogOptional.get(), + identifier, + inputFieldNames, + allPredicates._1(), + defaultPruner + ); + if (optionalPartitions.isPresent() && !optionalPartitions.get().isEmpty()) { + remainingPartitions = optionalPartitions.get(); + } + } catch (TableNotExistException tableNotExistException) { + throw new TableException(String.format("Table %s is not found in catalog.", identifier.asSummaryString()), e); + } catch (TableNotPartitionedException tableNotPartitionedException) { + remainingPartitions = null; + } + } + if (remainingPartitions != null) { + ((SupportsPartitionPushDown) dynamicTableSource).applyPartitions(remainingPartitions); + } + + // build new statistic + TableStats newTableStat = null; + Optional<TableStats> partitionStats; + if (remainingPartitions != null && catalogOptional.isPresent()) { + for (Map<String, String> partition: remainingPartitions) { + partitionStats = getPartitionStats(catalogOptional.get(), tablePath, partition); + if (!partitionStats.isPresent()) { + // clear all information before + newTableStat = null; + break; + } else { + newTableStat = newTableStat == null ? partitionStats.get() : newTableStat.merge(partitionStats.get()); + } + } + } + FlinkStatistic newStatistic = FlinkStatistic.builder() + .statistic(tableSourceTable.getStatistic()) + .tableStats(newTableStat) + .build(); + + String extraDigest = remainingPartitions == null ? "partitions=[]" : + ("partitions=[" + + String.join(", ", remainingPartitions + .stream() + .map(Object::toString) + .toArray(String[]::new)) + + "]"); + TableSourceTable newTableSourceTable = tableSourceTable.copy(dynamicTableSource, newStatistic, new String[]{extraDigest}); + + LogicalTableScan newScan = LogicalTableScan.create(scan.getCluster(), newTableSourceTable, scan.getHints()); + + RexNode nonPartitionPredicate = RexUtil.composeConjunction(rexBuilder, JavaConversions.seqAsJavaList(allPredicates._2())); + if (nonPartitionPredicate.isAlwaysTrue()) { + call.transformTo(newScan); + } else { + Filter newFilter = filter.copy(filter.getTraitSet(), newScan, nonPartitionPredicate); + call.transformTo(newFilter); + } + } + + /** + * adjust the partition field reference index to evaluate the partition values. + * e.g. the original input fields is: a, b, c, p, and p is partition field. the partition values + * are: List(Map("p"->"1"), Map("p" -> "2"), Map("p" -> "3")). If the original partition + * predicate is $3 > 1. after adjusting, the new predicate is ($0 > 1). + * and use ($0 > 1) to evaluate partition values (row(1), row(2), row(3)). + */ + private RexNode adjustPartitionPredicate(List<String> inputFieldNames, List<String> partitionFieldNames, RexNode partitionPredicate) { + return partitionPredicate.accept(new RexShuttle(){ + @Override + public RexNode visitInputRef(RexInputRef inputRef) { + int index = inputRef.getIndex(); + String fieldName = inputFieldNames.get(index); + int newIndex = partitionFieldNames.indexOf(fieldName); + if (newIndex < 0) { + throw new TableException(String.format("Field name '%s' isn't found in partitioned columns." + + " Validator should have checked that.", fieldName)); + } + if (newIndex == index){ + return inputRef; + } else { + return new RexInputRef(newIndex, inputRef.getType()); + } + } + }); + } + + private Optional<List<Map<String, String>>> readPartitionFromCatalogAndPrune( + FlinkContext context, + Catalog catalog, + ObjectIdentifier tableIdentifier, + List<String> allFieldNames, + Seq<RexNode> partitionPredicate, + Function<List<Map<String, String>>, List<Map<String, String>>> pruner) throws TableNotExistException, TableNotPartitionedException{ + RexNodeToExpressionConverter converter = new RexNodeToExpressionConverter( + allFieldNames.toArray(new String[0]), + context.getFunctionCatalog(), + context.getCatalogManager(), + TimeZone.getTimeZone(context.getTableConfig().getLocalTimeZone())); + ArrayList<Expression> partitionFilters = new ArrayList<>(); + Option<ResolvedExpression> subExpr; + for (RexNode node: JavaConversions.seqAsJavaList(partitionPredicate)) { + subExpr = node.accept(converter); + if (!subExpr.isEmpty()) { Review comment: if `subExpr` is empty, its corresponding sub-filter is dropped, then the result is incorrect. There is no test coverage... ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
