All Downloads are FREE. Search and download functionalities are using the official Maven repository.

com.amazonaws.athena.connectors.dynamodb.util.DDBPredicateUtils Maven / Gradle / Ivy

Go to download

There is a newer version: 2024.46.1
Show newest version
/*-
 * #%L
 * athena-dynamodb
 * %%
 * Copyright (C) 2019 Amazon Web Services
 * %%
 * Licensed 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.
 * #L%
 */
package com.amazonaws.athena.connectors.dynamodb.util;

import com.amazonaws.athena.connector.lambda.domain.predicate.EquatableValueSet;
import com.amazonaws.athena.connector.lambda.domain.predicate.Marker;
import com.amazonaws.athena.connector.lambda.domain.predicate.Range;
import com.amazonaws.athena.connector.lambda.domain.predicate.SortedRangeSet;
import com.amazonaws.athena.connector.lambda.domain.predicate.ValueSet;
import com.amazonaws.athena.connectors.dynamodb.model.DynamoDBIndex;
import com.amazonaws.athena.connectors.dynamodb.model.DynamoDBTable;
import com.google.common.base.Joiner;
import com.google.common.collect.ImmutableList;
import software.amazon.awssdk.services.dynamodb.model.AttributeValue;
import software.amazon.awssdk.services.dynamodb.model.ProjectionType;

import java.util.ArrayList;
import java.util.HashSet;
import java.util.List;
import java.util.Map;
import java.util.Set;
import java.util.stream.Collectors;
import java.util.stream.Stream;

import static com.google.common.base.Preconditions.checkState;
import static com.google.common.collect.Iterables.getOnlyElement;

/**
 * Provides utility methods relating to predicate handling.
 */
public class DDBPredicateUtils
{
    private DDBPredicateUtils() {}

    private static final Joiner AND_JOINER = Joiner.on(" AND ");
    private static final Joiner COMMA_JOINER = Joiner.on(",");
    private static final Joiner OR_JOINER = Joiner.on(" OR ");

    /**
     * Attempts to pick an optimal index (if any) from the given predicates. Returns the original table index if
     * one was not found.
     *
     * @param table the original table
     * @param predicates the predicates
     * @return the optimal index if found, otherwise the original table index
     */
    public static DynamoDBIndex getBestIndexForPredicates(DynamoDBTable table, List requestedCols, Map predicates)
    {
        Set columnNames = predicates.keySet();
        ImmutableList.Builder hashKeyMatchesBuilder = ImmutableList.builder();

        // create the original table index
        DynamoDBIndex tableIndex = new DynamoDBIndex(table.getName(), table.getHashKey(), table.getRangeKey(), ProjectionType.ALL, ImmutableList.of());

        // if the original table has a hash key matching a predicate, start with that
        if (columnNames.contains(tableIndex.getHashKey())) {
            // here, treat table as a special index
            hashKeyMatchesBuilder.add(tableIndex);
        }

        // requested columns must be projected in index
        List candidateIndices = table.getIndexes().stream()
              .filter(index -> indexContainsAllRequiredColumns(requestedCols, index, table))
              .collect(Collectors.toList());

        // get indices with hash keys that match a predicate
        candidateIndices.stream()
                .filter(index -> columnNames.contains(index.getHashKey()) && !getHashKeyAttributeValues(predicates.get(index.getHashKey())).isEmpty())
                .forEach(hashKeyMatchesBuilder::add);
        List hashKeyMatches = hashKeyMatchesBuilder.build();

        // if the original table has a range key matching a predicate, start with that
        ImmutableList.Builder rangeKeyMatchesBuilder = ImmutableList.builder();
        if (tableIndex.getRangeKey().isPresent() && columnNames.contains(tableIndex.getRangeKey().get())) {
            rangeKeyMatchesBuilder.add(tableIndex);
        }

        // get indices with range keys that match a predicate
        candidateIndices.stream()
                .filter(index -> index.getRangeKey().isPresent() && columnNames.contains(index.getRangeKey().get()))
                .forEach(rangeKeyMatchesBuilder::add);
        List rangeKeyMatches = rangeKeyMatchesBuilder.build();

        // return first index where both hash and range key can be specified with predicates
        for (DynamoDBIndex index : hashKeyMatches) {
            if (rangeKeyMatches.contains(index)) {
                return index;
            }
        }
        // else return the first index with a hash key predicate, or the original table if there are none
        return hashKeyMatches.isEmpty() ? tableIndex : hashKeyMatches.get(0);
    }

    /**
     * Generates a list of distinct values from the given {@link ValueSet} or an empty list if not possible.
     *
     * @param valueSet the value set to generate from
     * @return the list of distinct values
     */
    public static List getHashKeyAttributeValues(ValueSet valueSet)
    {
        if (valueSet.isSingleValue()) {
            return ImmutableList.of(valueSet.getSingleValue());
        }
        else if (valueSet instanceof SortedRangeSet) {
            List ranges = valueSet.getRanges().getOrderedRanges();
            ImmutableList.Builder attributeValues = ImmutableList.builder();
            for (Range range : ranges) {
                if (range.isSingleValue()) {
                    attributeValues.add(range.getSingleValue());
                }
                else {
                    // DDB Query can't handle non-equality conditions for the hash key
                    return ImmutableList.of();
                }
            }
            return attributeValues.build();
        }
        else if (valueSet instanceof EquatableValueSet) {
            EquatableValueSet equatableValueSet = (EquatableValueSet) valueSet;
            if (equatableValueSet.isWhiteList()) {
                ImmutableList.Builder values = ImmutableList.builder();
                for (int pos = 0; pos < equatableValueSet.getValueBlock().getRowCount(); pos++) {
                    values.add(equatableValueSet.getValue(pos));
                }
                return values.build();
            }
        }

        return ImmutableList.of();
    }

    /**
     * Generates a simple alias for a column to satisfy filter expressions. Uses a regex to convert illegal characters
     * (any character or combination of characters that are NOT included in [a-zA-Z_0-9]) to underscore.
     * Example: "column-$1`~!@#$%^&*()-=+[]{}\\|;:'\",.<>/?f3" -> "#column_1_f3"
     *
     * @see 
     *     https://docs.aws.amazon.com/amazondynamodb/latest/developerguide/Expressions.ExpressionAttributeNames.html
     * @param columnName the input column name
     * @return the aliased column name
     */
    public static String aliasColumn(String columnName)
    {
        return "#" + columnName.replaceAll("\\W+", "_");
    }

    /*
    Adds a value to the value accumulator.
     */
    private static void bindValue(String columnName, Object value, List accumulator, DDBRecordMetadata recordMetadata)
    {
        accumulator.add(DDBTypeUtils.toAttributeValue(DDBTypeUtils.convertArrowTypeIfNecessary(columnName, value, recordMetadata)));
    }

    /*
    Adds a value to the value accumulator and also returns an expression with given operands.
     */
    private static String toPredicate(String columnName, String operator, Object value, List accumulator,
                                      String valueName, DDBRecordMetadata recordMetadata)
    {
        bindValue(columnName, value, accumulator, recordMetadata);
        return aliasColumn(columnName) + " " + operator + " " + valueName;
    }

    private static void validateColumnRange(Range range)
    {
        if (!range.getLow().isLowerUnbounded()) {
            switch (range.getLow().getBound()) {
                case ABOVE:
                    break;
                case EXACTLY:
                    break;
                case BELOW:
                    throw new IllegalArgumentException("Low marker should never use BELOW bound");
                default:
                    throw new AssertionError("Unhandled lower bound: " + range.getLow().getBound());
            }
        }
        if (!range.getHigh().isUpperUnbounded()) {
            switch (range.getHigh().getBound()) {
                case ABOVE:
                    throw new IllegalArgumentException("High marker should never use ABOVE bound");
                case EXACTLY:
                    break;
                case BELOW:
                    break;
                default:
                    throw new AssertionError("Unhandled upper bound: " + range.getHigh().getBound());
            }
        }
    }

    /**
     * Generates a filter expression for a single column given a {@link ValueSet} predicate for that column.
     *
     * @param originalColumnName the column name
     * @param predicate the associated predicate
     * @param accumulator the value accumulator to add values to
     * @param valueNameProducer the value name producer to generate value aliases with
     * @param recordMetadata object containing any necessary metadata from the glue table
     * @param columnIsSortKey whether or not the originalColumnName column is a sort key
     * @return the generated filter expression
     */
    public static String generateSingleColumnFilter(String originalColumnName, ValueSet predicate, List accumulator,
            IncrementingValueNameProducer valueNameProducer, DDBRecordMetadata recordMetadata, boolean columnIsSortKey)
    {
        String columnName = aliasColumn(originalColumnName);

        if (predicate.isNone()) {
            return "(attribute_not_exists(" + columnName + ") OR " + toPredicate(originalColumnName, "=", null, accumulator, valueNameProducer.getNext(), recordMetadata) + ")";
        }

        if (predicate.isAll()) {
            return "(attribute_exists(" + columnName + ") AND " + toPredicate(originalColumnName, "<>", null, accumulator, valueNameProducer.getNext(), recordMetadata) + ")";
        }

        List disjuncts = new ArrayList<>();
        List singleValues = new ArrayList<>();
        boolean isWhitelist = true;
        if (predicate instanceof SortedRangeSet) {
            for (Range range : predicate.getRanges().getOrderedRanges()) {
                checkState(!range.isAll()); // Already checked
                if (range.isSingleValue()) {
                    singleValues.add(range.getLow().getValue());
                    continue;
                }
                validateColumnRange(range);
                List rangeConjuncts = new ArrayList<>();
                // DDB Only supports one condition per sort key, so here we attempt to combine the situations wherever we
                // have both inclusive upper and lower bounds (BETWEEN).
                if (range.getLow().getBound().equals(Marker.Bound.EXACTLY) && range.getHigh().getBound().equals(Marker.Bound.EXACTLY)) {
                    String startBetweenPredicate = toPredicate(originalColumnName, "BETWEEN", range.getLow().getValue(), accumulator, valueNameProducer.getNext(), recordMetadata);
                    String endBetweenPredicate = valueNameProducer.getNext();
                    bindValue(originalColumnName, range.getHigh().getValue(), accumulator, recordMetadata);
                    rangeConjuncts.add(startBetweenPredicate);
                    rangeConjuncts.add(endBetweenPredicate);
                }
                else {
                    // Otherwise in this situation, since we still want to
                    // push down, we will just push down one of the bounds
                    // here if it is a sort key.
                    // We will prioritize upper bounds because they should
                    // theoretically result in fewer matches vs lower bounds.
                    boolean upperBoundConditionAdded = false;
                    if (!range.getHigh().isUpperUnbounded()) {
                        switch (range.getHigh().getBound()) {
                            case EXACTLY:
                                rangeConjuncts.add(toPredicate(originalColumnName, "<=", range.getHigh().getValue(), accumulator, valueNameProducer.getNext(), recordMetadata));
                                upperBoundConditionAdded = true;
                                break;
                            case BELOW:
                                rangeConjuncts.add(toPredicate(originalColumnName, "<", range.getHigh().getValue(), accumulator, valueNameProducer.getNext(), recordMetadata));
                                upperBoundConditionAdded = true;
                                break;
                        }
                    }
                    // We can always add the lower bound if the column is not
                    // a sort key.
                    // But if it is a sort key, then we can only add it if we
                    // have not already added the upper bound.
                    boolean canAddLowerBound = (!columnIsSortKey || !upperBoundConditionAdded);
                    if (canAddLowerBound && !range.getLow().isLowerUnbounded()) {
                        switch (range.getLow().getBound()) {
                            case ABOVE:
                                rangeConjuncts.add(toPredicate(originalColumnName, ">", range.getLow().getValue(), accumulator, valueNameProducer.getNext(), recordMetadata));
                                break;
                            case EXACTLY:
                                rangeConjuncts.add(toPredicate(originalColumnName, ">=", range.getLow().getValue(), accumulator, valueNameProducer.getNext(), recordMetadata));
                                break;
                        }
                    }
                }
                // If rangeConjuncts is null, then the range was ALL, which should already have been checked for
                checkState(!rangeConjuncts.isEmpty());
                disjuncts.add("(" + AND_JOINER.join(rangeConjuncts) + ")");
            }
        }
        else {
            EquatableValueSet equatablePredicate = (EquatableValueSet) predicate;
            isWhitelist = equatablePredicate.isWhiteList();
            long valueCount = equatablePredicate.getValueBlock().getRowCount();
            for (int i = 0; i < valueCount; i++) {
                singleValues.add(equatablePredicate.getValue(i));
            }
        }

        // Add back all of the possible single values either as an equality or an IN predicate
        if (singleValues.size() == 1) {
            disjuncts.add(toPredicate(originalColumnName, isWhitelist ? "=" : "<>", getOnlyElement(singleValues), accumulator, valueNameProducer.getNext(), recordMetadata));
        }
        else if (singleValues.size() > 1) {
            for (Object value : singleValues) {
                bindValue(originalColumnName, value, accumulator, recordMetadata);
            }
            String values = COMMA_JOINER.join(Stream.generate(valueNameProducer::getNext).limit(singleValues.size()).collect(Collectors.toList()));
            disjuncts.add((isWhitelist ? "" : "NOT ") + columnName + " IN (" + values + ")");
        }

        // at this point we should have some disjuncts
        checkState(!disjuncts.isEmpty());

        // add nullability disjuncts
        if (predicate.isNullAllowed()) {
            disjuncts.add("attribute_not_exists(" + columnName + ") OR " +
                    toPredicate(originalColumnName, "=", null, accumulator, valueNameProducer.getNext(), recordMetadata));
        }

        // DDB doesn't like redundant parentheses
        if (disjuncts.size() == 1) {
            return disjuncts.get(0);
        }

        return "(" + OR_JOINER.join(disjuncts) + ")";
    }

    /**
     * Generates a combined filter expression for the given predicates.
     *
     * columnsToIgnore will contain any column that is of custom type (such as timestamp with tz)
     * as these types are not natively supported by ddb or glue.
     * we will need to filter them separately in the ddb query/scan result.
     *
     * @param columnsToIgnore the columns to not generate filters for
     * @param predicates the map of columns to predicates
     * @param accumulator the value accumulator to add values to
     * @param valueNameProducer the value name producer to generate value aliases with
     * @param recordMetadata object containing any necessary metadata from the glue table
     * @return the combined filter expression
     */
    public static String generateFilterExpression(Set columnsToIgnore, Map predicates, List accumulator,
            IncrementingValueNameProducer valueNameProducer, DDBRecordMetadata recordMetadata)
    {
        ImmutableList.Builder builder = ImmutableList.builder();
        for (Map.Entry predicate : predicates.entrySet()) {
            String columnName = predicate.getKey();
            if (!columnsToIgnore.contains(columnName)) {
                builder.add(generateSingleColumnFilter(columnName, predicate.getValue(), accumulator, valueNameProducer, recordMetadata, false));
            }
        }
        ImmutableList filters = builder.build();
        if (!filters.isEmpty()) {
            return AND_JOINER.join(filters);
        }
        return null;
    }

    public static boolean indexContainsAllRequiredColumns(List requestedCols, DynamoDBIndex index, DynamoDBTable table)
    {
        Set indexColumns = new HashSet<>();

        if (index == null || table == null) {
            return false;
        }

        // The partition key and sort key of the table are always projected into the index
        indexColumns.add(table.getHashKey());
        if (table.getRangeKey().isPresent()) {
            indexColumns.add(table.getRangeKey().get());
        }

        // including partition key and sort key of index
        indexColumns.add(index.getHashKey());
        if (index.getRangeKey().isPresent()) {
            indexColumns.add(index.getRangeKey().get());
        }

        // including more index columns based on projection type
        switch (index.getProjectionType()) {
            case ALL:
                return true;
            case INCLUDE:
                indexColumns.addAll(index.getProjectionAttributeNames());
                break;
            case KEYS_ONLY:
            default:
                break;
        }

        return indexColumns.containsAll(requestedCols);
    }
}
    

    

    

            

    

            



    
    






    
© 2015 - 2024 Weber Informatics LLC | Privacy Policy