diff --git a/processing/src/main/java/org/apache/druid/query/filter/DruidPredicateMatch.java b/processing/src/main/java/org/apache/druid/query/filter/DruidPredicateMatch.java index 7979481ec8ae..cc34ed0899c6 100644 --- a/processing/src/main/java/org/apache/druid/query/filter/DruidPredicateMatch.java +++ b/processing/src/main/java/org/apache/druid/query/filter/DruidPredicateMatch.java @@ -20,7 +20,10 @@ package org.apache.druid.query.filter; /** - * Three-value logic result for matching values with predicates produced by {@link DruidPredicateFactory} + * Three-value logic result for matching values with predicates produced by {@link DruidPredicateFactory}. + *

+ * Also serves as a general 3VL atom for filter-tree composition where a sub-expression can be provably true, + * provably false, or undecidable. * * @see DruidObjectPredicate * @see DruidLongPredicate @@ -58,4 +61,35 @@ public static DruidPredicateMatch of(boolean val) } return FALSE; } + + public static DruidPredicateMatch and(DruidPredicateMatch a, DruidPredicateMatch b) + { + if (a == FALSE || b == FALSE) { + return FALSE; + } + if (a == TRUE && b == TRUE) { + return TRUE; + } + return UNKNOWN; + } + + public static DruidPredicateMatch or(DruidPredicateMatch a, DruidPredicateMatch b) + { + if (a == TRUE || b == TRUE) { + return TRUE; + } + if (a == FALSE && b == FALSE) { + return FALSE; + } + return UNKNOWN; + } + + public static DruidPredicateMatch not(DruidPredicateMatch a) + { + return switch (a) { + case TRUE -> FALSE; + case FALSE -> TRUE; + default -> UNKNOWN; + }; + } } diff --git a/processing/src/main/java/org/apache/druid/segment/ConcatenatingCursor.java b/processing/src/main/java/org/apache/druid/segment/ConcatenatingCursor.java new file mode 100644 index 000000000000..2cfef8eec930 --- /dev/null +++ b/processing/src/main/java/org/apache/druid/segment/ConcatenatingCursor.java @@ -0,0 +1,152 @@ +/* + * 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.druid.segment; + +import com.google.common.base.Supplier; +import org.apache.druid.error.DruidException; +import org.apache.druid.segment.projections.ClusteringColumnSelectorFactory; + +import javax.annotation.Nullable; +import java.util.List; + +/** + * {@link Cursor} that walks a sequence of per-group cursors back-to-back, presenting them to the caller as a single + * cursor over a clustered base table. On group transitions the wrapper {@link ColumnSelectorFactory} updates its + * underlying delegate and clustering values so previously-acquired delegating selectors observe the new group's data + * on their next access. + *

+ * Each entry in {@link #holderSuppliers} is a lazy producer of a {@link CursorHolder} for one cluster group. The + * outer {@link CursorHolder} owns the lifecycle of the per-group holders. + */ +public final class ConcatenatingCursor implements Cursor +{ + private final List> holderSuppliers; + private final List clusteringValuesByGroup; + private final ClusteringColumnSelectorFactory wrapperFactory; + + private int currentIdx; + @Nullable + private Cursor currentCursor; + private boolean initialized; + + public ConcatenatingCursor( + List> holderSuppliers, + List clusteringValuesByGroup, + ClusteringColumnSelectorFactory wrapperFactory + ) + { + if (holderSuppliers.size() != clusteringValuesByGroup.size()) { + throw DruidException.defensive( + "holderSuppliers size [%s] must equal clusteringValuesByGroup size [%s]", + holderSuppliers.size(), + clusteringValuesByGroup.size() + ); + } + if (holderSuppliers.isEmpty()) { + throw DruidException.defensive("ConcatenatingCursor requires at least one cluster group"); + } + this.holderSuppliers = holderSuppliers; + this.clusteringValuesByGroup = clusteringValuesByGroup; + this.wrapperFactory = wrapperFactory; + this.currentIdx = -1; + } + + private void initializeIfNeeded() + { + if (initialized) { + return; + } + initialized = true; + advanceToNextNonEmptyGroup(); + } + + /** + * Open the next group whose cursor has at least one row. Sets {@code currentCursor = null} when all groups are + * exhausted. + */ + private void advanceToNextNonEmptyGroup() + { + while (++currentIdx < holderSuppliers.size()) { + final CursorHolder holder = holderSuppliers.get(currentIdx).get(); + final Cursor cursor = holder.asCursor(); + if (cursor != null && !cursor.isDone()) { + currentCursor = cursor; + wrapperFactory.setDelegate(cursor.getColumnSelectorFactory(), clusteringValuesByGroup.get(currentIdx)); + return; + } + // Group has no rows after filter application; try the next. + } + currentCursor = null; + } + + @Override + public ColumnSelectorFactory getColumnSelectorFactory() + { + initializeIfNeeded(); + return wrapperFactory; + } + + @Override + public void advance() + { + initializeIfNeeded(); + if (currentCursor == null) { + return; + } + currentCursor.advance(); + if (currentCursor.isDone()) { + advanceToNextNonEmptyGroup(); + } + } + + @Override + public void advanceUninterruptibly() + { + initializeIfNeeded(); + if (currentCursor == null) { + return; + } + currentCursor.advanceUninterruptibly(); + if (currentCursor.isDone()) { + advanceToNextNonEmptyGroup(); + } + } + + @Override + public boolean isDone() + { + initializeIfNeeded(); + return currentCursor == null; + } + + @Override + public boolean isDoneOrInterrupted() + { + return isDone() || Thread.currentThread().isInterrupted(); + } + + @Override + public void reset() + { + currentIdx = -1; + currentCursor = null; + initialized = false; + } +} diff --git a/processing/src/main/java/org/apache/druid/segment/IndexIO.java b/processing/src/main/java/org/apache/druid/segment/IndexIO.java index f246421f8415..ba9d6d455378 100644 --- a/processing/src/main/java/org/apache/druid/segment/IndexIO.java +++ b/processing/src/main/java/org/apache/druid/segment/IndexIO.java @@ -72,9 +72,11 @@ import org.apache.druid.segment.file.SegmentFileMetadata; import org.apache.druid.segment.projections.AggregateProjectionSchema; import org.apache.druid.segment.projections.BaseTableProjectionSchema; +import org.apache.druid.segment.projections.ClusteredValueGroupsBaseTableSchema; import org.apache.druid.segment.projections.ConstantTimeColumn; import org.apache.druid.segment.projections.ProjectionMetadata; import org.apache.druid.segment.projections.Projections; +import org.apache.druid.segment.projections.TableClusterGroupSpec; import org.apache.druid.segment.serde.ComplexColumnPartSupplier; import org.apache.druid.segment.serde.FloatNumericColumnSupplier; import org.apache.druid.segment.serde.LongNumericColumnSupplier; @@ -952,16 +954,33 @@ public QueryableIndex load(File inDir, ObjectMapper mapper, boolean lazy, Segmen // projections can omit a __time column, but one still has to exist, so we use the interval start to make a // constant for this case final long intervalStartMillis = Intervals.of(metadata.getInterval()).getStartMillis(); - // read base table projection columns, which are shared with other projections - final Map> baseColumns = readProjectionColumns( - metadata, - baseProjection, - fileMapper, - Map.of(), - intervalStartMillis, - lazy, - loadFailed - ); + + // For clustered base tables the columns are always accessed through cluster groups, skip reading any base + // columns so we don't try to map files that don't exist + final boolean isClusteredSummary = baseSchema instanceof ClusteredValueGroupsBaseTableSchema; + final ClusteredValueGroupsBaseTableSchema clusteredBaseSummary; + final Map> baseColumns; + if (isClusteredSummary) { + clusteredBaseSummary = (ClusteredValueGroupsBaseTableSchema) baseSchema; + if (clusteredBaseSummary.getSharedColumns().isEmpty()) { + baseColumns = Map.of(); + } else { + throw DruidException.defensive( + "Reading clustered segments with non-empty sharedColumns is not yet supported" + ); + } + } else { + clusteredBaseSummary = null; + baseColumns = readProjectionColumns( + metadata, + baseProjection, + fileMapper, + Map.of(), + intervalStartMillis, + lazy, + loadFailed + ); + } final Map>> projectionsColumns = new LinkedHashMap<>(); final List aggProjections = new ArrayList<>(metadata.getProjections().size() - 1); @@ -972,6 +991,13 @@ public QueryableIndex load(File inDir, ObjectMapper mapper, boolean lazy, Segmen first = false; continue; } + if (!(projectionSpec.getSchema() instanceof AggregateProjectionSchema)) { + throw DruidException.defensive( + "Unexpected projection[%s] with type[%s]; only aggregate projections are valid as top-level entries", + projectionSpec.getSchema().getName(), + projectionSpec.getSchema().getClass() + ); + } final Map> projectionColumns = readProjectionColumns( metadata, projectionSpec, @@ -983,31 +1009,51 @@ public QueryableIndex load(File inDir, ObjectMapper mapper, boolean lazy, Segmen ); projectionsColumns.put(projectionSpec.getSchema().getName(), projectionColumns); - if (projectionSpec.getSchema() instanceof AggregateProjectionSchema) { - aggProjections.add( - new AggregateProjectionMetadata( - (AggregateProjectionSchema) projectionSpec.getSchema(), - projectionSpec.getNumRows() - ) - ); - } else { - throw DruidException.defensive( - "Unexpected projection[%s] with type[%s]", - projectionSpec.getSchema().getName(), - projectionSpec.getSchema().getClass() - ); + aggProjections.add( + new AggregateProjectionMetadata( + (AggregateProjectionSchema) projectionSpec.getSchema(), + projectionSpec.getNumRows() + ) + ); + } + + final List>> clusterGroupColumnsList; + if (isClusteredSummary) { + final List nestedGroups = clusteredBaseSummary.getClusterGroups(); + clusterGroupColumnsList = new ArrayList<>(nestedGroups.size()); + for (int i = 0; i < nestedGroups.size(); i++) { + clusterGroupColumnsList.add(readClusterGroupColumns( + metadata, + clusteredBaseSummary, + i, + fileMapper, + intervalStartMillis, + lazy, + loadFailed + )); } + } else { + clusterGroupColumnsList = List.of(); } + final Metadata reconstructedMetadata = baseSchema.asMetadata(aggProjections); + // For clustered segments, the top-level index has no per-column data of its own, so pass an empty dimensions + // list so the SimpleQueryableIndex precondition passes and dimension-handler init has nothing to materialize + final Indexed dimensionsIndex = isClusteredSummary + ? new ListIndexed<>(List.of()) + : new ListIndexed<>(baseSchema.getDimensionNames()); + return new SimpleQueryableIndex( Intervals.fromString(metadata.getInterval()), - new ListIndexed<>(baseSchema.getDimensionNames()), + dimensionsIndex, metadata.getBitmapEncoding().getBitmapFactory(), baseColumns, fileMapper, reconstructedMetadata, - projectionsColumns + projectionsColumns, + clusteredBaseSummary, + clusterGroupColumnsList ) { @Override @@ -1080,6 +1126,63 @@ private Map> readProjectionColumns( return projectionColumns; } + /** + * Read the per-column data for cluster group {@code groupIndex}. Mirrors {@link #readProjectionColumns} but + * with the dictionary-id-tuple smoosh prefix {@code __base$_.../}; the column set + * excludes clustering columns (constants, injected at query time). + */ + private Map> readClusterGroupColumns( + SegmentFileMetadata metadata, + ClusteredValueGroupsBaseTableSchema summary, + int groupIndex, + SegmentFileMapper segmentFileMapper, + long intervalStartMillis, + boolean lazy, + SegmentLazyLoadFailCallback loadFailed + ) throws IOException + { + final TableClusterGroupSpec spec = summary.getClusterGroups().get(groupIndex); + final List clusteringValueIds = spec.getClusteringValueIds(); + final String timeColumnName = summary.getTimeColumnName(); + final boolean renameTime = !ColumnHolder.TIME_COLUMN_NAME.equals(timeColumnName); + final Map> groupColumns = new LinkedHashMap<>(); + + for (String column : summary.getGroupColumnNames()) { + final String smooshName = Projections.getClusterGroupSegmentInternalFileName(clusteringValueIds, column); + final ByteBuffer colBuffer = segmentFileMapper.mapFile(smooshName); + final ColumnDescriptor columnDescriptor = metadata.getColumnDescriptors().get(smooshName); + if (columnDescriptor == null) { + continue; + } + + final String internedColumnName = SmooshedFileMapper.STRING_INTERNER.intern(column); + groupColumns.put( + internedColumnName, + makeColumnHolderSupplier( + internedColumnName, + columnDescriptor, + colBuffer, + segmentFileMapper, + null, + lazy, + loadFailed + ) + ); + + if (column.equals(timeColumnName) && renameTime) { + groupColumns.put(ColumnHolder.TIME_COLUMN_NAME, groupColumns.get(column)); + groupColumns.remove(column); + } + } + if (timeColumnName == null) { + groupColumns.put( + ColumnHolder.TIME_COLUMN_NAME, + ConstantTimeColumn.makeConstantTimeSupplier(spec.getNumRows(), intervalStartMillis) + ); + } + return groupColumns; + } + private Supplier makeColumnHolderSupplier( String columnName, ColumnDescriptor columnDescriptor, diff --git a/processing/src/main/java/org/apache/druid/segment/QueryableIndex.java b/processing/src/main/java/org/apache/druid/segment/QueryableIndex.java index ce9bee38ac19..6f604fbb18cc 100644 --- a/processing/src/main/java/org/apache/druid/segment/QueryableIndex.java +++ b/processing/src/main/java/org/apache/druid/segment/QueryableIndex.java @@ -25,12 +25,15 @@ import org.apache.druid.segment.column.ColumnCapabilities; import org.apache.druid.segment.column.ColumnHolder; import org.apache.druid.segment.data.Indexed; +import org.apache.druid.segment.projections.ClusteredValueGroupsBaseTableSchema; import org.apache.druid.segment.projections.QueryableProjection; +import org.apache.druid.segment.projections.TableClusterGroupSpec; import org.joda.time.Interval; import javax.annotation.Nullable; import java.io.Closeable; import java.io.IOException; +import java.util.Collections; import java.util.List; import java.util.Map; @@ -108,4 +111,37 @@ default QueryableIndex getProjectionQueryableIndex(String name) { return null; } + + /** + * Returns the {@link ClusteredValueGroupsBaseTableSchema} summary if this index represents a clustered base table, or + * {@code null} for a non-clustered segment. Default returns {@code null}; only V10-loaded clustered segments + * override. + */ + @Nullable + default ClusteredValueGroupsBaseTableSchema getClusteredBaseSummary() + { + return null; + } + + /** + * Returns the list of {@link TableClusterGroupSpec} entries on this index, one per cluster group. Empty for a + * non-clustered segment. For a clustered segment, this is the same list returned by + * {@code getClusteredBaseSummary().getClusterGroups()}, surfaced here so query-time dispatch can enumerate cluster + * groups (e.g. via {@code Projections.pruneClusterGroups}). + */ + default List getClusterGroupSchemas() + { + return Collections.emptyList(); + } + + /** + * Returns a {@link QueryableIndex} sub-view scoped to a single cluster group's column data. Mirrors + * {@link #getProjectionQueryableIndex(String)} but for cluster groups, addressed by reference rather than name. + * Default returns {@code null}; only clustered segments override. + */ + @Nullable + default QueryableIndex getClusterGroupQueryableIndex(TableClusterGroupSpec groupSpec) + { + return null; + } } diff --git a/processing/src/main/java/org/apache/druid/segment/QueryableIndexCursorFactory.java b/processing/src/main/java/org/apache/druid/segment/QueryableIndexCursorFactory.java index f1aa2bb7bf57..982c681cbafe 100644 --- a/processing/src/main/java/org/apache/druid/segment/QueryableIndexCursorFactory.java +++ b/processing/src/main/java/org/apache/druid/segment/QueryableIndexCursorFactory.java @@ -19,18 +19,41 @@ package org.apache.druid.segment; +import com.google.common.base.Supplier; +import com.google.common.base.Suppliers; +import org.apache.druid.error.DruidException; +import org.apache.druid.java.util.common.io.Closer; import org.apache.druid.query.OrderBy; import org.apache.druid.query.aggregation.AggregatorFactory; +import org.apache.druid.query.dimension.DimensionSpec; import org.apache.druid.segment.column.ColumnCapabilities; +import org.apache.druid.segment.column.ColumnCapabilitiesImpl; import org.apache.druid.segment.column.ColumnHolder; import org.apache.druid.segment.column.ColumnType; import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.column.ValueType; import org.apache.druid.segment.data.Offset; +import org.apache.druid.segment.projections.ClusteredValueGroupsBaseTableSchema; +import org.apache.druid.segment.projections.ClusteringColumnSelectorFactory; +import org.apache.druid.segment.projections.ClusteringVectorColumnSelectorFactory; +import org.apache.druid.segment.projections.Projections; import org.apache.druid.segment.projections.QueryableProjection; +import org.apache.druid.segment.projections.TableClusterGroupSpec; +import org.apache.druid.segment.vector.ConcatenatingVectorCursor; +import org.apache.druid.segment.vector.MultiValueDimensionVectorSelector; +import org.apache.druid.segment.vector.ReadableVectorInspector; +import org.apache.druid.segment.vector.SingleValueDimensionVectorSelector; import org.apache.druid.segment.vector.VectorColumnSelectorFactory; +import org.apache.druid.segment.vector.VectorCursor; +import org.apache.druid.segment.vector.VectorObjectSelector; import org.apache.druid.segment.vector.VectorOffset; +import org.apache.druid.segment.vector.VectorValueSelector; import javax.annotation.Nullable; +import java.io.IOException; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.Collections; import java.util.LinkedHashSet; import java.util.List; @@ -61,12 +84,22 @@ public QueryableIndexCursorFactory(QueryableIndex index) public CursorHolder makeCursorHolder(CursorBuildSpec spec) { QueryableProjection projection = index.getProjection(spec); - if (projection == null) { - // no projections, create regular cursor holder - return new QueryableIndexCursorHolder(index, spec, timeBoundaryInspector); + if (projection != null) { + return makeAggregateProjectionCursorHolder(projection); } - // create projection cursor holder + // Cluster-group dispatch runs after aggregate-projection match, before the regular base-table fallback + final ClusteredValueGroupsBaseTableSchema clusterSummary = index.getClusteredBaseSummary(); + if (clusterSummary != null) { + return makeClusteredCursorHolder(spec, clusterSummary); + } + + // No projections, no clustering, regular full-segment cursor. + return new QueryableIndexCursorHolder(index, spec, timeBoundaryInspector); + } + + private CursorHolder makeAggregateProjectionCursorHolder(QueryableProjection projection) + { return new QueryableIndexCursorHolder( projection.getRowSelector(), projection.getCursorBuildSpec(), @@ -110,9 +143,328 @@ public List getAggregatorsForPreAggregated() }; } + private CursorHolder makeClusteredCursorHolder(CursorBuildSpec spec, ClusteredValueGroupsBaseTableSchema clusterSummary) + { + final List matching = Projections.pruneClusterGroups( + new ArrayList<>(index.getClusterGroupSchemas()), + spec.getFilter(), + spec.getVirtualColumns() + ); + + if (matching.isEmpty()) { + return EmptyClusteredCursorHolder.INSTANCE; + } + + if (matching.size() == 1) { + return makeSingleGroupClusteredCursorHolder(spec, matching.get(0)); + } + return makeMultiGroupClusteredCursorHolder(spec, matching); + } + + private CursorHolder makeSingleGroupClusteredCursorHolder( + CursorBuildSpec spec, + TableClusterGroupSpec valueGroup + ) + { + final QueryableIndex groupIndex = index.getClusterGroupQueryableIndex(valueGroup); + if (groupIndex == null) { + throw DruidException.defensive( + "No cluster-group sub-index resolvable for clustering values " + + Arrays.toString(valueGroup.lookupClusteringValues()) + ); + } + + return new QueryableIndexCursorHolder( + groupIndex, + spec, + QueryableIndexTimeBoundaryInspector.create(groupIndex) + ) + { + @Override + protected ColumnSelectorFactory makeColumnSelectorFactoryForOffset( + ColumnCache columnCache, + Offset baseOffset + ) + { + return new ClusteringColumnSelectorFactory( + super.makeColumnSelectorFactoryForOffset(columnCache, baseOffset), + valueGroup.getSummary().getClusteringColumns(), + valueGroup.lookupClusteringValues() + ); + } + + @Override + protected VectorColumnSelectorFactory makeVectorColumnSelectorFactoryForOffset( + ColumnCache columnCache, + VectorOffset baseOffset + ) + { + return new ClusteringVectorColumnSelectorFactory( + super.makeVectorColumnSelectorFactoryForOffset(columnCache, baseOffset), + valueGroup.getSummary().getClusteringColumns(), + valueGroup.lookupClusteringValues() + ); + } + }; + } + + /** + * Build a cursor holder that walks multiple matching cluster groups back-to-back via + * {@link ConcatenatingCursor}. Each per-group {@link CursorHolder} is built lazily inside the cursor's group + * transition, so a query that finishes early (e.g., LIMIT-bounded) doesn't open every group's offset. + */ + private CursorHolder makeMultiGroupClusteredCursorHolder( + CursorBuildSpec spec, + List matching + ) + { + // All matching specs share the same parent summary (they came out of one segment); grab a reference for + // getOrdering() and clusteringColumns below. + final ClusteredValueGroupsBaseTableSchema clusterSummary = matching.get(0).getSummary(); + final RowSignature clusteringColumns = clusterSummary.getClusteringColumns(); + final List clusteringValuesByGroup = new ArrayList<>(matching.size()); + final List> holderSuppliers = new ArrayList<>(matching.size()); + // lifecycle management closer for per-group CursorHolders + final Closer closer = Closer.create(); + for (TableClusterGroupSpec valueGroup : matching) { + clusteringValuesByGroup.add(valueGroup.lookupClusteringValues()); + final QueryableIndex groupIndex = index.getClusterGroupQueryableIndex(valueGroup); + if (groupIndex == null) { + throw DruidException.defensive( + "No cluster-group sub-index resolvable for clustering values " + + Arrays.toString(valueGroup.lookupClusteringValues()) + ); + } + holderSuppliers.add( + Suppliers.memoize( + () -> closer.register( + new QueryableIndexCursorHolder( + groupIndex, + spec, + QueryableIndexTimeBoundaryInspector.create(groupIndex) + ) + ) + ) + ); + } + + // Initial wrapper state uses the first group's clustering values + a throwing placeholder delegate. The + // ConcatenatingCursor immediately calls setDelegate on init (before any selector is exposed). The vector + // wrapper carries the query-level max vector size from the build spec, the placeholder delegate can't be + // queried for sizing, and the value is constant across groups anyway. + final int vectorSize = spec.getQueryContext().getVectorSize(); + final ClusteringColumnSelectorFactory wrapperFactory = new ClusteringColumnSelectorFactory( + UNINITIALIZED_DELEGATE, + clusteringColumns, + clusteringValuesByGroup.get(0) + ); + final ClusteringVectorColumnSelectorFactory vectorWrapperFactory = new ClusteringVectorColumnSelectorFactory( + UNINITIALIZED_VECTOR_DELEGATE, + clusteringColumns, + clusteringValuesByGroup.get(0), + vectorSize + ); + + final ConcatenatingCursor cursor = new ConcatenatingCursor( + holderSuppliers, + clusteringValuesByGroup, + wrapperFactory + ); + final ConcatenatingVectorCursor vectorCursor = new ConcatenatingVectorCursor( + holderSuppliers, + clusteringValuesByGroup, + vectorWrapperFactory + ); + + // canVectorize() is determined by the per-group holders. Probe the first one (lazily, `Suppliers.memoize` + // means this opens it once and is reused by ConcatenatingVectorCursor). + final boolean canVectorize = holderSuppliers.get(0).get().canVectorize(); + + return new CursorHolder() + { + @Override + public Cursor asCursor() + { + return cursor; + } + + @Override + public VectorCursor asVectorCursor() + { + return vectorCursor; + } + + @Override + public boolean canVectorize() + { + return canVectorize; + } + + @Override + public List getOrdering() + { + // Cluster groups are written in clustering-value order (writer-enforced; see ClusteredValueGroupsBaseTableSchema), + // and within each group rows are sorted by the segment ordering's tail (clustering prefix dropped). So + // back-to-back walking yields rows in the full segment ordering; the writer-side contract makes the + // concatenation order-preserving without any merge work at read time. + return clusterSummary.getOrdering(); + } + + @Override + public void close() + { + try { + closer.close(); + } + catch (IOException e) { + throw new RuntimeException(e); + } + } + }; + } + + /** + * Placeholder delegate for the {@link ClusteringColumnSelectorFactory} constructed by + * {@link #makeMultiGroupClusteredCursorHolder}. Throws on any access; replaced by the concatenating cursor's + * lazy init before the wrapper is exposed to the caller. + */ + private static final ColumnSelectorFactory UNINITIALIZED_DELEGATE = new ColumnSelectorFactory() + { + @Override + public DimensionSelector makeDimensionSelector(DimensionSpec dimensionSpec) + { + throw DruidException.defensive("ConcatenatingCursor delegate accessed before initialization"); + } + + @Override + public ColumnValueSelector makeColumnValueSelector(String columnName) + { + throw DruidException.defensive("ConcatenatingCursor delegate accessed before initialization"); + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + return null; + } + }; + + /** + * Vector counterpart of {@link #UNINITIALIZED_DELEGATE}. Replaced by + * {@link ConcatenatingVectorCursor}'s lazy init before the wrapper is exposed. + */ + private static final VectorColumnSelectorFactory UNINITIALIZED_VECTOR_DELEGATE = new VectorColumnSelectorFactory() + { + @Override + public ReadableVectorInspector getReadableVectorInspector() + { + throw DruidException.defensive("ConcatenatingVectorCursor delegate accessed before initialization"); + } + + @Override + public SingleValueDimensionVectorSelector makeSingleValueDimensionSelector(DimensionSpec dimensionSpec) + { + throw DruidException.defensive("ConcatenatingVectorCursor delegate accessed before initialization"); + } + + @Override + public MultiValueDimensionVectorSelector makeMultiValueDimensionSelector(DimensionSpec dimensionSpec) + { + throw DruidException.defensive("ConcatenatingVectorCursor delegate accessed before initialization"); + } + + @Override + public VectorValueSelector makeValueSelector(String column) + { + throw DruidException.defensive("ConcatenatingVectorCursor delegate accessed before initialization"); + } + + @Override + public VectorObjectSelector makeObjectSelector(String column) + { + throw DruidException.defensive("ConcatenatingVectorCursor delegate accessed before initialization"); + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + return null; + } + }; + + /** + * CursorHolder that yields no rows. Used when {@link Projections#pruneClusterGroups} excludes every cluster + * group, so the filter is provably unsatisfiable on this segment. + */ + private static final class EmptyClusteredCursorHolder implements CursorHolder + { + static final EmptyClusteredCursorHolder INSTANCE = new EmptyClusteredCursorHolder(); + + @Override + public Cursor asCursor() + { + return new Cursor() + { + @Override + public ColumnSelectorFactory getColumnSelectorFactory() + { + throw DruidException.defensive( + "No column selector factory available on an empty cluster-group cursor" + ); + } + + @Override + public void advance() + { + } + + @Override + public void advanceUninterruptibly() + { + } + + @Override + public boolean isDone() + { + return true; + } + + @Override + public boolean isDoneOrInterrupted() + { + return true; + } + + @Override + public void reset() + { + } + }; + } + + @Override + public boolean canVectorize() + { + return false; + } + + @Override + public List getOrdering() + { + return Collections.emptyList(); + } + } + @Override public RowSignature getRowSignature() { + final ClusteredValueGroupsBaseTableSchema clusterSummary = index.getClusteredBaseSummary(); + if (clusterSummary != null) { + return getClusteredRowSignature(clusterSummary); + } + final LinkedHashSet columns = new LinkedHashSet<>(); for (final OrderBy orderBy : index.getOrdering()) { @@ -137,10 +489,82 @@ public RowSignature getRowSignature() return builder.build(); } + /** + * Build the row signature for a clustered segment. Top-level columns are empty, so column types are sourced from: + * - the summary's clustering {@link RowSignature} for clustering columns; + * - the first cluster group's sub-index for everything else (all groups share the same data-column shape). + */ + private RowSignature getClusteredRowSignature(ClusteredValueGroupsBaseTableSchema clusterSummary) + { + final LinkedHashSet columns = new LinkedHashSet<>(); + + for (final OrderBy orderBy : clusterSummary.getOrdering()) { + columns.add(orderBy.getColumnName()); + } + columns.add(ColumnHolder.TIME_COLUMN_NAME); + columns.addAll(clusterSummary.getColumnNames()); + + final RowSignature.Builder builder = RowSignature.builder(); + for (final String column : columns) { + final ColumnType columnType = resolveClusteredColumnType(column, clusterSummary); + if (columnType != null) { + builder.add(column, columnType); + } + } + return builder.build(); + } + + @Nullable + private ColumnType resolveClusteredColumnType(String column, ClusteredValueGroupsBaseTableSchema clusterSummary) + { + // 1. Clustering columns: typed RowSignature on the summary is authoritative. + final ColumnType clusteringType = clusterSummary.getClusteringColumns().getColumnType(column).orElse(null); + if (clusteringType != null) { + return clusteringType; + } + // 2. Data columns: ask the first cluster group's sub-index. All groups share the same column shape. + final List groups = index.getClusterGroupSchemas(); + if (groups.isEmpty()) { + return null; + } + final QueryableIndex firstGroupIndex = index.getClusterGroupQueryableIndex(groups.get(0)); + if (firstGroupIndex == null) { + return null; + } + return ColumnType.fromCapabilities(firstGroupIndex.getColumnCapabilities(column)); + } + @Nullable @Override public ColumnCapabilities getColumnCapabilities(String column) { + final ClusteredValueGroupsBaseTableSchema clusterSummary = index.getClusteredBaseSummary(); + if (clusterSummary != null) { + return getClusteredColumnCapabilities(column, clusterSummary); + } return index.getColumnCapabilities(column); } + + @Nullable + private ColumnCapabilities getClusteredColumnCapabilities(String column, ClusteredValueGroupsBaseTableSchema clusterSummary) + { + // synthesize capabilities from the typed RowSignature. + final ColumnType clusteringType = clusterSummary.getClusteringColumns().getColumnType(column).orElse(null); + if (clusteringType != null) { + if (clusteringType.is(ValueType.STRING)) { + return ColumnCapabilitiesImpl.createSimpleSingleValueStringColumnCapabilities(); + } + return ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(clusteringType); + } + // ask the first cluster group's sub-index. + final List groups = index.getClusterGroupSchemas(); + if (groups.isEmpty()) { + return null; + } + final QueryableIndex firstGroupIndex = index.getClusterGroupQueryableIndex(groups.get(0)); + if (firstGroupIndex == null) { + return null; + } + return firstGroupIndex.getColumnCapabilities(column); + } } diff --git a/processing/src/main/java/org/apache/druid/segment/SimpleQueryableIndex.java b/processing/src/main/java/org/apache/druid/segment/SimpleQueryableIndex.java index e6b1ab0c29c1..33386c77eb90 100644 --- a/processing/src/main/java/org/apache/druid/segment/SimpleQueryableIndex.java +++ b/processing/src/main/java/org/apache/druid/segment/SimpleQueryableIndex.java @@ -29,14 +29,17 @@ import com.google.common.collect.Maps; import it.unimi.dsi.fastutil.objects.ObjectAVLTreeSet; import org.apache.druid.collections.bitmap.BitmapFactory; +import org.apache.druid.error.DruidException; import org.apache.druid.query.OrderBy; import org.apache.druid.segment.column.BaseColumnHolder; import org.apache.druid.segment.column.ColumnHolder; import org.apache.druid.segment.data.Indexed; import org.apache.druid.segment.data.ListIndexed; import org.apache.druid.segment.file.SegmentFileMapper; +import org.apache.druid.segment.projections.ClusteredValueGroupsBaseTableSchema; import org.apache.druid.segment.projections.Projections; import org.apache.druid.segment.projections.QueryableProjection; +import org.apache.druid.segment.projections.TableClusterGroupSpec; import org.joda.time.Interval; import javax.annotation.Nullable; @@ -63,6 +66,9 @@ public abstract class SimpleQueryableIndex implements QueryableIndex private final Map projectionsMap; private final SortedSet projections; private final Map>> projectionColumns; + @Nullable + private final ClusteredValueGroupsBaseTableSchema clusteredBaseSummary; + private final List>> clusterGroupColumns; private final SegmentFileMapper fileMapper; private final Supplier> dimensionHandlers; @@ -74,7 +80,7 @@ public SimpleQueryableIndex( SegmentFileMapper fileMapper ) { - this(dataInterval, dimNames, bitmapFactory, columns, fileMapper, null, null); + this(dataInterval, dimNames, bitmapFactory, columns, fileMapper, null, null, null, null); } public SimpleQueryableIndex( @@ -87,7 +93,26 @@ public SimpleQueryableIndex( @Nullable Map>> projectionColumns ) { - Preconditions.checkNotNull(columns.get(ColumnHolder.TIME_COLUMN_NAME)); + this(dataInterval, dimNames, bitmapFactory, columns, fileMapper, metadata, projectionColumns, null, null); + } + + public SimpleQueryableIndex( + Interval dataInterval, + Indexed dimNames, + BitmapFactory bitmapFactory, + Map> columns, + SegmentFileMapper fileMapper, + @Nullable Metadata metadata, + @Nullable Map>> projectionColumns, + @Nullable ClusteredValueGroupsBaseTableSchema clusteredBaseSummary, + @Nullable List>> clusterGroupColumns + ) + { + // For clustered base tables, the top-level columns map is empty; all column data lives under per-cluster-group + // entries in clusterGroupColumns. For all other schema shapes, __time must be present in the top-level columns map + if (!columns.isEmpty()) { + Preconditions.checkNotNull(columns.get(ColumnHolder.TIME_COLUMN_NAME)); + } this.dataInterval = Preconditions.checkNotNull(dataInterval, "dataInterval"); ImmutableList.Builder columnNamesBuilder = ImmutableList.builder(); LinkedHashSet dimsFirst = new LinkedHashSet<>(); @@ -107,6 +132,10 @@ public SimpleQueryableIndex( this.fileMapper = fileMapper; this.projectionColumns = projectionColumns == null ? Collections.emptyMap() : projectionColumns; + this.clusteredBaseSummary = clusteredBaseSummary; + this.clusterGroupColumns = clusterGroupColumns == null + ? Collections.emptyList() + : List.copyOf(clusterGroupColumns); this.dimensionHandlers = Suppliers.memoize(() -> initDimensionHandlers(availableDimensions)); if (metadata != null) { @@ -127,7 +156,7 @@ public SimpleQueryableIndex( this.projections = Collections.emptySortedSet(); } } else { - // When sort order isn't set in metadata.drd, assume the segment is sorted by __time. + // When sort order isn't available from metadata, assume the segment is sorted by __time. this.ordering = Cursors.ascendingTimeOrder(); this.projections = Collections.emptySortedSet(); this.projectionsMap = Collections.emptyMap(); @@ -184,6 +213,73 @@ public Map> getColumns() return columns; } + /** + * Returns the {@link ClusteredValueGroupsBaseTableSchema} summary entry if this index is for a clustered segment, or null for + * a non-clustered segment. The summary owns segment-wide clustering config (clustering column signature, shared- + * column markers, naming-scheme version) + */ + @Override + @Nullable + public ClusteredValueGroupsBaseTableSchema getClusteredBaseSummary() + { + return clusteredBaseSummary; + } + + /** + * Returns the cluster groups nested in this index's summary, in their original order. Empty for a non-clustered + * segment. Used by query-time dispatch to enumerate groups and feed {@link Projections#pruneClusterGroups}. + */ + @Override + public List getClusterGroupSchemas() + { + return clusteredBaseSummary == null ? Collections.emptyList() : clusteredBaseSummary.getClusterGroups(); + } + + /** + * Returns a {@link QueryableIndex} sub-view scoped to a single cluster group's column data. Mirrors + * {@link #getProjectionQueryableIndex(String)} but for cluster groups: addressed by reference to the spec, not by + * name. The returned index's columns are the group's per-group columns; clustering columns are NOT present in + * the returned index, they're injected at the cursor-factory level via {@code ClusteringColumnSelectorFactory}. + */ + @Override + public QueryableIndex getClusterGroupQueryableIndex(TableClusterGroupSpec groupSpec) + { + if (clusteredBaseSummary == null) { + throw DruidException.defensive("getClusterGroupQueryableIndex called on a non-clustered segment"); + } + final List groups = clusteredBaseSummary.getClusterGroups(); + final int index = groups.indexOf(groupSpec); + if (index < 0) { + throw DruidException.defensive("Cluster group spec is not part of this segment"); + } + final Map> groupColumns = clusterGroupColumns.get(index); + final Metadata groupMetadata = new Metadata( + null, + clusteredBaseSummary.getAggregators(), + null, + clusteredBaseSummary.getEffectiveGranularity(), + false, + clusteredBaseSummary.getGroupOrdering(), + null + ); + return new SimpleQueryableIndex( + dataInterval, + new ListIndexed<>(clusteredBaseSummary.getGroupDimensionNames()), + bitmapFactory, + groupColumns, + fileMapper, + groupMetadata, + null + ) + { + @Override + public Metadata getMetadata() + { + return groupMetadata; + } + }; + } + @VisibleForTesting public SegmentFileMapper getFileMapper() { diff --git a/processing/src/main/java/org/apache/druid/segment/projections/ClusteredValueGroupsBaseTableSchema.java b/processing/src/main/java/org/apache/druid/segment/projections/ClusteredValueGroupsBaseTableSchema.java new file mode 100644 index 000000000000..e90502edc3f4 --- /dev/null +++ b/processing/src/main/java/org/apache/druid/segment/projections/ClusteredValueGroupsBaseTableSchema.java @@ -0,0 +1,410 @@ +/* + * 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.druid.segment.projections; + +import com.fasterxml.jackson.annotation.JsonCreator; +import com.fasterxml.jackson.annotation.JsonIgnore; +import com.fasterxml.jackson.annotation.JsonInclude; +import com.fasterxml.jackson.annotation.JsonProperty; +import com.google.common.collect.Lists; +import org.apache.druid.error.DruidException; +import org.apache.druid.java.util.common.granularity.Granularities; +import org.apache.druid.java.util.common.granularity.Granularity; +import org.apache.druid.query.OrderBy; +import org.apache.druid.query.aggregation.AggregatorFactory; +import org.apache.druid.segment.AggregateProjectionMetadata; +import org.apache.druid.segment.Metadata; +import org.apache.druid.segment.VirtualColumn; +import org.apache.druid.segment.VirtualColumns; +import org.apache.druid.segment.column.ColumnHolder; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.utils.CollectionUtils; + +import javax.annotation.Nullable; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.HashSet; +import java.util.List; +import java.util.Objects; +import java.util.Set; + +/** + * Top-level summary for a clustered base table whose groups are identified by discrete clustering-value tuples. Each + * tuple group is internally stored as a separate table without storing the cluster columns, which are pulled into this + * metadata. This is optimizing for use cases which typically only need to read from a single group via filters present + * on a query. Cluster groups nest inside as {@link #getClusterGroups()}; their column data live in the V10 segment + * file under dictionary-id-tuple prefixes ({@code __base$_.../}), where the ids index into + * {@link #getClusteringDictionaries()}. + */ +public class ClusteredValueGroupsBaseTableSchema implements BaseTableProjectionSchema +{ + public static final String TYPE_NAME = "clustered-value-groups-base-table"; + + private final VirtualColumns virtualColumns; + private final List columnNames; + private final AggregatorFactory[] aggregators; + private final List ordering; + private final RowSignature clusteringColumns; + private final List sharedColumns; + private final ClusteringDictionaries clusteringDictionaries; + private final List clusterGroups; + + // computed + private final int timeColumnPosition; + private final Granularity effectiveGranularity; + + @JsonCreator + public ClusteredValueGroupsBaseTableSchema( + @JsonProperty("virtualColumns") VirtualColumns virtualColumns, + @JsonProperty("columns") List columns, + @JsonProperty("aggregators") @Nullable AggregatorFactory[] aggregators, + @JsonProperty("ordering") List ordering, + @JsonProperty("clusteringColumns") RowSignature clusteringColumns, + @JsonProperty("sharedColumns") @Nullable List sharedColumns, + @JsonProperty("clusteringDictionaries") @Nullable ClusteringDictionaries clusteringDictionaries, + @JsonProperty("clusterGroups") @Nullable List clusterGroups + ) + { + if (CollectionUtils.isNullOrEmpty(columns)) { + throw DruidException.defensive("clustered base table schema columns must not be null or empty"); + } + if (ordering == null) { + throw DruidException.defensive("clustered base table schema ordering must not be null"); + } + if (clusteringColumns == null || clusteringColumns.size() == 0) { + throw DruidException.defensive( + "clustered base table schema clusteringColumns must not be null or empty" + ); + } + if (ordering.size() < clusteringColumns.size()) { + throw DruidException.defensive( + "ordering size [%s] must be at least clusteringColumns size [%s] (clustering columns must form a prefix" + + " of the segment ordering)", + ordering.size(), + clusteringColumns.size() + ); + } + for (int i = 0; i < clusteringColumns.size(); i++) { + final String clusteringColumn = clusteringColumns.getColumnName(i); + if (!columns.contains(clusteringColumn)) { + throw DruidException.defensive( + "clusteringColumn [%s] must appear in columns of the clustered base table summary", + clusteringColumn + ); + } + final ColumnType type = clusteringColumns.getColumnType(i).orElse(null); + if (!Projections.isAllowedClusteringType(type)) { + throw DruidException.defensive( + "clustering column [%s] has unsupported type [%s]; allowed types are STRING, LONG, DOUBLE, FLOAT", + clusteringColumn, + type + ); + } + // Per-group ordering is derived by dropping this prefix; pruning + cursor concatenation rely on it. + final String orderingColumn = ordering.get(i).getColumnName(); + if (!clusteringColumn.equals(orderingColumn)) { + throw DruidException.defensive( + "clustering column at position [%s] is [%s] but the segment ordering at the same position is [%s];" + + " clustering columns must form a prefix of the segment ordering", + i, + clusteringColumn, + orderingColumn + ); + } + } + final List resolvedSharedColumns = sharedColumns == null ? List.of() : sharedColumns; + for (String shared : resolvedSharedColumns) { + if (!columns.contains(shared)) { + throw DruidException.defensive( + "sharedColumn [%s] must appear in columns of the clustered base table summary", + shared + ); + } + } + this.virtualColumns = virtualColumns == null ? VirtualColumns.EMPTY : virtualColumns; + this.columnNames = columns; + this.aggregators = aggregators == null ? new AggregatorFactory[0] : aggregators; + this.ordering = ordering; + this.clusteringColumns = clusteringColumns; + this.sharedColumns = resolvedSharedColumns; + this.clusterGroups = clusterGroups == null ? List.of() : List.copyOf(clusterGroups); + this.clusteringDictionaries = clusteringDictionaries == null + ? ClusteringDictionaries.EMPTY + : clusteringDictionaries; + + int foundTimePosition = -1; + Granularity granularity = null; + for (int i = 0; i < ordering.size(); i++) { + OrderBy orderBy = ordering.get(i); + if (orderBy.getColumnName().equals(ColumnHolder.TIME_COLUMN_NAME)) { + foundTimePosition = i; + final VirtualColumn vc = this.virtualColumns.getVirtualColumn(Granularities.GRANULARITY_VIRTUAL_COLUMN_NAME); + if (vc != null) { + granularity = Granularities.fromVirtualColumn(vc); + } else { + granularity = Granularities.NONE; + } + } + } + if (granularity == null) { + throw DruidException.defensive( + "clustered base table doesn't have a [%s] column?", + ColumnHolder.TIME_COLUMN_NAME + ); + } + this.timeColumnPosition = foundTimePosition; + this.effectiveGranularity = granularity; + + // Specs always start unwired: there's a chicken-and-egg between the summary and its specs, resolved by + // deferring all summary-dependent state on the spec to setSummary, which we invoke here once the summary's + // own state is populated. + for (TableClusterGroupSpec spec : this.clusterGroups) { + spec.setSummary(this); + } + } + + @JsonIgnore + @Override + public List getColumnNames() + { + List columns = new ArrayList<>(columnNames.size() + aggregators.length); + columns.addAll(columnNames); + for (AggregatorFactory aggregator : aggregators) { + columns.add(aggregator.getName()); + } + return columns; + } + + @JsonProperty + @Override + public VirtualColumns getVirtualColumns() + { + return virtualColumns; + } + + @JsonProperty + public List getColumns() + { + return columnNames; + } + + @JsonProperty + @JsonInclude(JsonInclude.Include.NON_DEFAULT) + public AggregatorFactory[] getAggregators() + { + return aggregators; + } + + @JsonProperty + @Override + public List getOrdering() + { + return ordering; + } + + @JsonProperty + public RowSignature getClusteringColumns() + { + return clusteringColumns; + } + + /** + * Columns which have common data stored once under {@link Projections#BASE_TABLE_PROJECTION_NAME} and shared by + * all cluster-group entries + */ + @JsonProperty + @JsonInclude(JsonInclude.Include.NON_EMPTY) + public List getSharedColumns() + { + return sharedColumns; + } + + /** + * Per-type clustering value dictionaries; see {@link ClusteringDictionaries} for routing semantics. + */ + @JsonProperty + public ClusteringDictionaries getClusteringDictionaries() + { + return clusteringDictionaries; + } + + /** + * Materialize the typed value at clustering position {@code clusteringColumnIndex} with dictionary + * {@code dictionaryId}. + */ + Object lookupClusteringValue(int clusteringColumnIndex, int dictionaryId) + { + return clusteringDictionaries.lookupValue( + clusteringColumns.getColumnType(clusteringColumnIndex).orElseThrow(), + dictionaryId + ); + } + + /** + * The cluster groups nested in this summary, in clustering-value sort order. Walking groups back-to-back in + * this order yields rows in the segment's full declared ordering (clustering values monotonically advance + * across groups; within each group, rows follow the segment ordering with the clustering prefix dropped). + */ + @JsonProperty + @JsonInclude(JsonInclude.Include.NON_EMPTY) + public List getClusterGroups() + { + return clusterGroups; + } + + /** + * Per-group sort order: the segment ordering with the clustering-column prefix dropped. + */ + @JsonIgnore + public List getGroupOrdering() + { + return ordering.subList(clusteringColumns.size(), ordering.size()); + } + + /** + * Per-group column names: this summary's full column list (including aggregator names) minus the clustering columns. + */ + @JsonIgnore + public List getGroupColumnNames() + { + final Set clusteringNames = new HashSet<>(clusteringColumns.getColumnNames()); + final List all = getColumnNames(); + final List result = new ArrayList<>(all.size()); + for (String c : all) { + if (!clusteringNames.contains(c)) { + result.add(c); + } + } + return result; + } + + /** + * Per-group dimension names: {@link #getDimensionNames()} minus the clustering columns. + */ + @JsonIgnore + public List getGroupDimensionNames() + { + final Set clusteringNames = new HashSet<>(clusteringColumns.getColumnNames()); + final List dims = getDimensionNames(); + final List result = new ArrayList<>(dims.size()); + for (String d : dims) { + if (!clusteringNames.contains(d)) { + result.add(d); + } + } + return result; + } + + @JsonIgnore + @Override + public int getTimeColumnPosition() + { + return timeColumnPosition; + } + + @JsonIgnore + @Override + public Granularity getEffectiveGranularity() + { + return effectiveGranularity; + } + + @JsonIgnore + @Override + public List getDimensionNames() + { + if (timeColumnPosition == 0) { + return columnNames.subList(1, columnNames.size()); + } + final List dimsWithoutTime = Lists.newArrayListWithCapacity(columnNames.size() - 1); + for (String column : columnNames) { + if (ColumnHolder.TIME_COLUMN_NAME.equals(column)) { + continue; + } + dimsWithoutTime.add(column); + } + return dimsWithoutTime; + } + + @Override + public Metadata asMetadata(@Nullable List projections) + { + return new Metadata( + null, + aggregators, + null, + effectiveGranularity, + false, + ordering, + projections + ); + } + + @Override + public boolean equals(Object o) + { + if (this == o) { + return true; + } + if (o == null || getClass() != o.getClass()) { + return false; + } + ClusteredValueGroupsBaseTableSchema that = (ClusteredValueGroupsBaseTableSchema) o; + return Objects.equals(virtualColumns, that.virtualColumns) + && Objects.equals(columnNames, that.columnNames) + && Objects.deepEquals(aggregators, that.aggregators) + && Objects.equals(ordering, that.ordering) + && Objects.equals(clusteringColumns, that.clusteringColumns) + && Objects.equals(sharedColumns, that.sharedColumns) + && Objects.equals(clusteringDictionaries, that.clusteringDictionaries) + && Objects.equals(clusterGroups, that.clusterGroups); + } + + @Override + public int hashCode() + { + return Objects.hash( + virtualColumns, + columnNames, + Arrays.hashCode(aggregators), + ordering, + clusteringColumns, + sharedColumns, + clusteringDictionaries, + clusterGroups + ); + } + + @Override + public String toString() + { + return "ClusteredValueGroupsBaseTableSchema{" + + "virtualColumns=" + virtualColumns + + ", columnNames=" + columnNames + + ", aggregators=" + Arrays.toString(aggregators) + + ", ordering=" + ordering + + ", clusteringColumns=" + clusteringColumns + + ", sharedColumns=" + sharedColumns + + ", clusteringDictionaries=" + clusteringDictionaries + + ", clusterGroups=" + clusterGroups + + '}'; + } +} diff --git a/processing/src/main/java/org/apache/druid/segment/projections/ClusteringColumnSelectorFactory.java b/processing/src/main/java/org/apache/druid/segment/projections/ClusteringColumnSelectorFactory.java new file mode 100644 index 000000000000..a48610278db1 --- /dev/null +++ b/processing/src/main/java/org/apache/druid/segment/projections/ClusteringColumnSelectorFactory.java @@ -0,0 +1,566 @@ +/* + * 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.druid.segment.projections; + +import org.apache.druid.error.DruidException; +import org.apache.druid.math.expr.ExprEval; +import org.apache.druid.math.expr.ExpressionType; +import org.apache.druid.query.dimension.DimensionSpec; +import org.apache.druid.query.filter.DruidPredicateFactory; +import org.apache.druid.query.filter.ValueMatcher; +import org.apache.druid.query.monomorphicprocessing.RuntimeShapeInspector; +import org.apache.druid.segment.ColumnSelectorFactory; +import org.apache.druid.segment.ColumnValueSelector; +import org.apache.druid.segment.ConstantExprEvalSelector; +import org.apache.druid.segment.DimensionSelector; +import org.apache.druid.segment.IdLookup; +import org.apache.druid.segment.RowIdSupplier; +import org.apache.druid.segment.column.ColumnCapabilities; +import org.apache.druid.segment.column.ColumnCapabilitiesImpl; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.column.ValueType; +import org.apache.druid.segment.data.IndexedInts; + +import javax.annotation.Nullable; +import java.util.function.Supplier; + +/** + * {@link ColumnSelectorFactory} wrapper that intercepts requests for clustering columns and returns selectors + * carrying the group's constant value, while delegating all other column lookups to a wrapped factory. This is the + * mechanism by which a cluster group's clustering columns, which are NOT stored in the per-group column data since + * they're constant across the group, are made visible to query engines as if they were ordinary columns. + */ +public class ClusteringColumnSelectorFactory implements ColumnSelectorFactory +{ + private final RowSignature clusteringColumns; + private ColumnSelectorFactory delegate; + private Object[] clusteringValues; + // Bumped on every setDelegate(...) so per-call selector wrappers can detect group transitions and rebuild their + // cached inner state + private long generation; + + public ClusteringColumnSelectorFactory( + ColumnSelectorFactory delegate, + RowSignature clusteringColumns, + Object[] clusteringValues + ) + { + this.clusteringColumns = clusteringColumns; + setDelegate(delegate, clusteringValues); + } + + /** + * Update the underlying factory and the constant values for the current cluster group. Called by a multi-group + * concatenating cursor on each group transition. Selectors previously returned by this factory will, on their next + * invocation, observe the updated state; see the per-call indirection in the inner selector classes. + */ + public void setDelegate(ColumnSelectorFactory delegate, Object[] clusteringValues) + { + if (clusteringValues == null || clusteringValues.length != clusteringColumns.size()) { + throw DruidException.defensive( + "clusteringValues length [%s] must match clusteringColumns size [%s]", + clusteringValues == null ? "null" : clusteringValues.length, + clusteringColumns.size() + ); + } + this.delegate = delegate; + this.clusteringValues = clusteringValues; + this.generation++; + } + + ColumnSelectorFactory getDelegate() + { + return delegate; + } + + long getGeneration() + { + return generation; + } + + @Override + public DimensionSelector makeDimensionSelector(DimensionSpec dimensionSpec) + { + final int idx = clusteringColumns.indexOf(dimensionSpec.getDimension()); + if (idx < 0) { + return new DelegatingDimensionSelector(this, dimensionSpec); + } + return new ClusteringDimensionSelector(this, idx, dimensionSpec); + } + + @Override + public ColumnValueSelector makeColumnValueSelector(String columnName) + { + final int idx = clusteringColumns.indexOf(columnName); + if (idx < 0) { + return new DelegatingColumnValueSelector(this, columnName); + } + return new ClusteringColumnValueSelector(this, idx, clusteringColumns.getColumnType(idx).orElseThrow()); + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + final int idx = clusteringColumns.indexOf(column); + if (idx < 0) { + return delegate.getColumnCapabilities(column); + } + final ColumnType type = clusteringColumns.getColumnType(idx).orElseThrow(); + if (type.is(ValueType.STRING)) { + return ColumnCapabilitiesImpl.createSimpleSingleValueStringColumnCapabilities(); + } + return ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(type); + } + + @Nullable + @Override + public RowIdSupplier getRowIdSupplier() + { + return delegate.getRowIdSupplier(); + } + + Object currentValue(int idx) + { + return clusteringValues[idx]; + } + + /** + * Dimension selector for a clustering column. Delegates the value lookup back to the parent factory each call so + * that group transitions (which mutate the parent's clustering values) are observed immediately. Internally + * decorates a {@link DimensionSelector#constant(String)} re-built when the underlying value changes. + */ + private static final class ClusteringDimensionSelector implements DimensionSelector + { + private final ClusteringColumnSelectorFactory parent; + private final int idx; + private final DimensionSpec spec; + private DimensionSelector cachedSelector; + private long cachedGeneration = -1; + + private ClusteringDimensionSelector(ClusteringColumnSelectorFactory parent, int idx, DimensionSpec spec) + { + this.parent = parent; + this.idx = idx; + this.spec = spec; + } + + private DimensionSelector currentSelector() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration == currentGeneration) { + return cachedSelector; + } + final Object raw = parent.currentValue(idx); + final String stringValue = raw == null ? null : String.valueOf(raw); + cachedSelector = DimensionSelector.constant(stringValue, spec.getExtractionFn()); + cachedGeneration = currentGeneration; + return cachedSelector; + } + + @Override + public IndexedInts getRow() + { + return currentSelector().getRow(); + } + + @Override + public ValueMatcher makeValueMatcher(@Nullable String value) + { + // Generation-aware: rebuild the matcher when the clustering value itself changes (group transition). + return new ClusteringValueMatcher(() -> currentSelector().makeValueMatcher(value)); + } + + @Override + public ValueMatcher makeValueMatcher( + DruidPredicateFactory predicateFactory + ) + { + return new ClusteringValueMatcher(() -> currentSelector().makeValueMatcher(predicateFactory)); + } + + /** + * Generation-aware matcher for the clustering-column path. The constant value itself changes between groups, + * so a held matcher must re-resolve from the current per-generation constant selector. + */ + private final class ClusteringValueMatcher implements ValueMatcher + { + private final Supplier factory; + private long cachedGeneration = -1; + private ValueMatcher cachedMatcher; + + private ClusteringValueMatcher(Supplier factory) + { + this.factory = factory; + } + + private ValueMatcher current() + { + final long gen = parent.getGeneration(); + if (cachedGeneration != gen) { + cachedMatcher = factory.get(); + cachedGeneration = gen; + } + return cachedMatcher; + } + + @Override + public boolean matches(boolean includeUnknown) + { + return current().matches(includeUnknown); + } + + @Override + public void inspectRuntimeShape(RuntimeShapeInspector inspector) + { + inspector.visit("clustering-matcher", idx); + } + } + + @Override + public int getValueCardinality() + { + return currentSelector().getValueCardinality(); + } + + @Nullable + @Override + public String lookupName(int id) + { + return currentSelector().lookupName(id); + } + + @Override + public boolean nameLookupPossibleInAdvance() + { + return currentSelector().nameLookupPossibleInAdvance(); + } + + @Nullable + @Override + public IdLookup idLookup() + { + return currentSelector().idLookup(); + } + + @Nullable + @Override + public Object getObject() + { + return currentSelector().getObject(); + } + + @Override + public Class classOfObject() + { + return currentSelector().classOfObject(); + } + + @Override + public void inspectRuntimeShape(RuntimeShapeInspector inspector) + { + inspector.visit("clusteringIndex", idx); + } + } + + /** + * Value selector for a clustering column. Caches a {@link ConstantExprEvalSelector} built from the group's typed + * value; on group transition, the cache rebuilds against the new value. Mirrors + * {@code EvalUnwrappingColumnValueSelector} in {@code ExpressionSelectors}. + */ + private static final class ClusteringColumnValueSelector implements ColumnValueSelector + { + private final ClusteringColumnSelectorFactory parent; + private final int idx; + private final ExpressionType expressionType; + private long cachedGeneration = -1; + private ConstantExprEvalSelector cachedSelector; + + private ClusteringColumnValueSelector(ClusteringColumnSelectorFactory parent, int idx, ColumnType columnType) + { + this.parent = parent; + this.idx = idx; + this.expressionType = ExpressionType.fromColumnTypeStrict(columnType); + } + + private ConstantExprEvalSelector currentSelector() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration == currentGeneration) { + return cachedSelector; + } + cachedSelector = new ConstantExprEvalSelector(ExprEval.ofType(expressionType, parent.currentValue(idx))); + cachedGeneration = currentGeneration; + return cachedSelector; + } + + @Override + public double getDouble() + { + return currentSelector().getDouble(); + } + + @Override + public float getFloat() + { + return currentSelector().getFloat(); + } + + @Override + public long getLong() + { + return currentSelector().getLong(); + } + + @Override + public boolean isNull() + { + return currentSelector().isNull(); + } + + @Nullable + @Override + public Object getObject() + { + return currentSelector().getObject().value(); + } + + @Override + public Class classOfObject() + { + return Object.class; + } + + @Override + public void inspectRuntimeShape(RuntimeShapeInspector inspector) + { + inspector.visit("clusteringIndex", idx); + } + } + + /** + * Stable {@link DimensionSelector} for a non-clustering column. Each call delegates to the parent factory's + * current underlying delegate; on group transition, the cached inner selector is rebuilt against the new delegate. + * For the single-group case, the cache fills once and never invalidates. + * + *

Value matchers returned by {@link #makeValueMatcher(String)} / + * {@link #makeValueMatcher(DruidPredicateFactory)} are also generation-aware; they re-resolve their inner matcher + * from the current delegate on group transition, so callers that hold a matcher across transitions observe the + * new group's data the same way the selector itself does. + */ + private static final class DelegatingDimensionSelector implements DimensionSelector + { + private final ClusteringColumnSelectorFactory parent; + private final DimensionSpec spec; + private long cachedGeneration = -1; + private DimensionSelector cachedInner; + + private DelegatingDimensionSelector(ClusteringColumnSelectorFactory parent, DimensionSpec spec) + { + this.parent = parent; + this.spec = spec; + } + + private DimensionSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration != currentGeneration) { + cachedInner = parent.getDelegate().makeDimensionSelector(spec); + cachedGeneration = currentGeneration; + } + return cachedInner; + } + + @Override + public IndexedInts getRow() + { + return currentInner().getRow(); + } + + @Override + public ValueMatcher makeValueMatcher(@Nullable String value) + { + return new DelegatingValueMatcher(() -> currentInner().makeValueMatcher(value)); + } + + @Override + public ValueMatcher makeValueMatcher(DruidPredicateFactory predicateFactory) + { + return new DelegatingValueMatcher(() -> currentInner().makeValueMatcher(predicateFactory)); + } + + /** + * Generation-aware {@link ValueMatcher}: re-resolves its inner matcher from the current + * {@link DimensionSelector} on each group transition. Non-static inner class so it can read the outer + * selector's {@code parent.getGeneration()} and trigger a rebuild via the supplier. + */ + private final class DelegatingValueMatcher implements ValueMatcher + { + private final Supplier factory; + private long cachedGeneration = -1; + private ValueMatcher cachedMatcher; + + private DelegatingValueMatcher(Supplier factory) + { + this.factory = factory; + } + + private ValueMatcher current() + { + final long gen = parent.getGeneration(); + if (cachedGeneration != gen) { + cachedMatcher = factory.get(); + cachedGeneration = gen; + } + return cachedMatcher; + } + + @Override + public boolean matches(boolean includeUnknown) + { + return current().matches(includeUnknown); + } + + @Override + public void inspectRuntimeShape(RuntimeShapeInspector inspector) + { + inspector.visit("delegating-matcher", spec.getDimension()); + } + } + + @Override + public int getValueCardinality() + { + return currentInner().getValueCardinality(); + } + + @Nullable + @Override + public String lookupName(int id) + { + return currentInner().lookupName(id); + } + + @Override + public boolean nameLookupPossibleInAdvance() + { + return currentInner().nameLookupPossibleInAdvance(); + } + + @Nullable + @Override + public IdLookup idLookup() + { + return currentInner().idLookup(); + } + + @Nullable + @Override + public Object getObject() + { + return currentInner().getObject(); + } + + @Override + public Class classOfObject() + { + return currentInner().classOfObject(); + } + + @Override + public void inspectRuntimeShape(RuntimeShapeInspector inspector) + { + inspector.visit("delegating", spec.getDimension()); + } + } + + /** + * Stable {@link ColumnValueSelector} for a non-clustering column. Same delegating-with-generation-cache pattern as + * {@link DelegatingDimensionSelector}. + */ + private static final class DelegatingColumnValueSelector implements ColumnValueSelector + { + private final ClusteringColumnSelectorFactory parent; + private final String columnName; + private long cachedGeneration = -1; + private ColumnValueSelector cachedInner; + + private DelegatingColumnValueSelector(ClusteringColumnSelectorFactory parent, String columnName) + { + this.parent = parent; + this.columnName = columnName; + } + + @SuppressWarnings("unchecked") + private ColumnValueSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration != currentGeneration) { + cachedInner = parent.getDelegate().makeColumnValueSelector(columnName); + cachedGeneration = currentGeneration; + } + return (ColumnValueSelector) cachedInner; + } + + @Override + public double getDouble() + { + return currentInner().getDouble(); + } + + @Override + public float getFloat() + { + return currentInner().getFloat(); + } + + @Override + public long getLong() + { + return currentInner().getLong(); + } + + @Override + public boolean isNull() + { + return currentInner().isNull(); + } + + @Nullable + @Override + public Object getObject() + { + return currentInner().getObject(); + } + + @Override + public Class classOfObject() + { + return currentInner().classOfObject(); + } + + @Override + public void inspectRuntimeShape(RuntimeShapeInspector inspector) + { + inspector.visit("delegating", columnName); + } + } +} diff --git a/processing/src/main/java/org/apache/druid/segment/projections/ClusteringDictionaries.java b/processing/src/main/java/org/apache/druid/segment/projections/ClusteringDictionaries.java new file mode 100644 index 000000000000..4f94ee45793f --- /dev/null +++ b/processing/src/main/java/org/apache/druid/segment/projections/ClusteringDictionaries.java @@ -0,0 +1,180 @@ +/* + * 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.druid.segment.projections; + +import com.fasterxml.jackson.annotation.JsonCreator; +import com.fasterxml.jackson.annotation.JsonInclude; +import com.fasterxml.jackson.annotation.JsonProperty; +import com.google.common.collect.Interner; +import org.apache.druid.error.DruidException; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.ValueType; +import org.apache.druid.timeline.DataSegment; + +import javax.annotation.Nullable; +import java.util.ArrayList; +import java.util.Collections; +import java.util.List; +import java.util.Objects; + +/** + * Per-type sorted-nulls-first dictionaries for a clustered base table's clustering values. Each + * {@link TableClusterGroupSpec#getClusteringValueIds()} entry at position {@code i} indexes into the dictionary + * for the column type at position {@code i}; columns of the same type share a dictionary. + * + *

String entries are interned via {@link DataSegment#stringInterner()} on construction since clustering + * strings repeat heavily across cached segments. Numeric types aren't interned. + */ +public class ClusteringDictionaries +{ + public static final ClusteringDictionaries EMPTY = new ClusteringDictionaries(null, null, null, null); + + private final List stringDictionary; + private final List longDictionary; + private final List doubleDictionary; + private final List floatDictionary; + + @JsonCreator + public ClusteringDictionaries( + @JsonProperty("string") @Nullable List stringDictionary, + @JsonProperty("long") @Nullable List longDictionary, + @JsonProperty("double") @Nullable List doubleDictionary, + @JsonProperty("float") @Nullable List floatDictionary + ) + { + this.stringDictionary = internStringDictionary(stringDictionary); + this.longDictionary = longDictionary == null ? List.of() : Collections.unmodifiableList(longDictionary); + this.doubleDictionary = doubleDictionary == null ? List.of() : Collections.unmodifiableList(doubleDictionary); + this.floatDictionary = floatDictionary == null ? List.of() : Collections.unmodifiableList(floatDictionary); + } + + @JsonProperty("string") + @JsonInclude(JsonInclude.Include.NON_EMPTY) + public List getStringDictionary() + { + return stringDictionary; + } + + @JsonProperty("long") + @JsonInclude(JsonInclude.Include.NON_EMPTY) + public List getLongDictionary() + { + return longDictionary; + } + + @JsonProperty("double") + @JsonInclude(JsonInclude.Include.NON_EMPTY) + public List getDoubleDictionary() + { + return doubleDictionary; + } + + @JsonProperty("float") + @JsonInclude(JsonInclude.Include.NON_EMPTY) + public List getFloatDictionary() + { + return floatDictionary; + } + + /** + * Look up the typed value at position {@code id} in the dictionary for {@code type}. Throws on out-of-range + * {@code id} or unsupported {@code type}. + */ + @Nullable + public Object lookupValue(ColumnType type, int id) + { + final List dict = dictionaryForType(type); + if (id < 0 || id >= dict.size()) { + throw DruidException.defensive( + "dictionary id [%s] is out of range for clustering type [%s] (size [%s])", + id, + type, + dict.size() + ); + } + return dict.get(id); + } + + public List dictionaryForType(ColumnType type) + { + if (type == null) { + throw DruidException.defensive("clustering type must not be null"); + } + if (type.is(ValueType.STRING)) { + return stringDictionary; + } + if (type.is(ValueType.LONG)) { + return longDictionary; + } + if (type.is(ValueType.DOUBLE)) { + return doubleDictionary; + } + if (type.is(ValueType.FLOAT)) { + return floatDictionary; + } + throw DruidException.defensive("unsupported clustering type [%s]", type); + } + + @Override + public boolean equals(Object o) + { + if (this == o) { + return true; + } + if (o == null || getClass() != o.getClass()) { + return false; + } + ClusteringDictionaries that = (ClusteringDictionaries) o; + return Objects.equals(stringDictionary, that.stringDictionary) + && Objects.equals(longDictionary, that.longDictionary) + && Objects.equals(doubleDictionary, that.doubleDictionary) + && Objects.equals(floatDictionary, that.floatDictionary); + } + + @Override + public int hashCode() + { + return Objects.hash(stringDictionary, longDictionary, doubleDictionary, floatDictionary); + } + + @Override + public String toString() + { + return "ClusteringDictionaries{" + + "string=" + stringDictionary + + ", long=" + longDictionary + + ", double=" + doubleDictionary + + ", float=" + floatDictionary + + '}'; + } + + private static List internStringDictionary(@Nullable List dict) + { + if (dict == null || dict.isEmpty()) { + return List.of(); + } + final Interner interner = DataSegment.stringInterner(); + final List out = new ArrayList<>(dict.size()); + for (String s : dict) { + out.add(s == null ? null : interner.intern(s)); + } + return Collections.unmodifiableList(out); + } +} diff --git a/processing/src/main/java/org/apache/druid/segment/projections/ClusteringVectorColumnSelectorFactory.java b/processing/src/main/java/org/apache/druid/segment/projections/ClusteringVectorColumnSelectorFactory.java new file mode 100644 index 000000000000..450bd859815d --- /dev/null +++ b/processing/src/main/java/org/apache/druid/segment/projections/ClusteringVectorColumnSelectorFactory.java @@ -0,0 +1,631 @@ +/* + * 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.druid.segment.projections; + +import org.apache.druid.error.DruidException; +import org.apache.druid.query.dimension.DimensionSpec; +import org.apache.druid.segment.IdLookup; +import org.apache.druid.segment.column.ColumnCapabilities; +import org.apache.druid.segment.column.ColumnCapabilitiesImpl; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.column.ValueType; +import org.apache.druid.segment.data.IndexedInts; +import org.apache.druid.segment.vector.ConstantVectorSelectors; +import org.apache.druid.segment.vector.MultiValueDimensionVectorSelector; +import org.apache.druid.segment.vector.ReadableVectorInspector; +import org.apache.druid.segment.vector.SingleValueDimensionVectorSelector; +import org.apache.druid.segment.vector.VectorColumnSelectorFactory; +import org.apache.druid.segment.vector.VectorObjectSelector; +import org.apache.druid.segment.vector.VectorValueSelector; + +import javax.annotation.Nullable; + +/** + * Vectorized counterpart of {@link ClusteringColumnSelectorFactory}. Wraps a delegate + * {@link VectorColumnSelectorFactory} and intercepts requests for clustering columns, returning constant-typed + * vector selectors (via {@link ConstantVectorSelectors}). Other column requests pass through to delegating wrappers. + * + * The factory is mutable via {@link #setDelegate(VectorColumnSelectorFactory, Object[])}: a multi-group + * {@code ConcatenatingVectorCursor} swaps the underlying delegate + clustering values on each group transition. + * Selectors previously returned by this factory observe the new state on their next call thanks to a generation + * counter cache invalidation. + * + * For single-group dispatch, the factory is constructed once and {@code setDelegate} is never called; selectors' + * caches fill on first access and never invalidate. + */ +public class ClusteringVectorColumnSelectorFactory implements VectorColumnSelectorFactory +{ + private final RowSignature clusteringColumns; + private final int maxVectorSize; + private VectorColumnSelectorFactory delegate; + private Object[] clusteringValues; + // Bumped on every setDelegate(...) so per-call selector wrappers can detect group transitions and rebuild their + // cached inner state. + private long generation; + + /** + * Convenience overload that derives {@code maxVectorSize} from the supplied delegate. Used by single-group + * dispatch where the delegate is the per-group {@link VectorColumnSelectorFactory} from the cursor itself. + */ + public ClusteringVectorColumnSelectorFactory( + VectorColumnSelectorFactory delegate, + RowSignature clusteringColumns, + Object[] clusteringValues + ) + { + this(delegate, clusteringColumns, clusteringValues, delegate.getMaxVectorSize()); + } + + public ClusteringVectorColumnSelectorFactory( + VectorColumnSelectorFactory delegate, + RowSignature clusteringColumns, + Object[] clusteringValues, + int maxVectorSize + ) + { + this.clusteringColumns = clusteringColumns; + this.maxVectorSize = maxVectorSize; + setDelegate(delegate, clusteringValues); + } + + /** + * Update the underlying delegate and the constant clustering values for the current cluster group. Called by a + * multi-group {@code ConcatenatingVectorCursor} on each group transition. + */ + public void setDelegate(VectorColumnSelectorFactory delegate, Object[] clusteringValues) + { + if (clusteringValues == null || clusteringValues.length != clusteringColumns.size()) { + throw DruidException.defensive( + "clusteringValues length [%s] must match clusteringColumns size [%s]", + clusteringValues == null ? "null" : clusteringValues.length, + clusteringColumns.size() + ); + } + this.delegate = delegate; + this.clusteringValues = clusteringValues; + this.generation++; + } + + VectorColumnSelectorFactory getDelegate() + { + return delegate; + } + + long getGeneration() + { + return generation; + } + + Object currentValue(int idx) + { + return clusteringValues[idx]; + } + + @Override + public ReadableVectorInspector getReadableVectorInspector() + { + return delegate.getReadableVectorInspector(); + } + + @Override + public int getMaxVectorSize() + { + return maxVectorSize; + } + + @Override + public SingleValueDimensionVectorSelector makeSingleValueDimensionSelector(DimensionSpec dimensionSpec) + { + final int idx = clusteringColumns.indexOf(dimensionSpec.getDimension()); + if (idx < 0) { + return new DelegatingSingleValueDimensionVectorSelector(this, dimensionSpec); + } + return new ClusteringSingleValueDimensionVectorSelector(this, idx, dimensionSpec); + } + + @Override + public MultiValueDimensionVectorSelector makeMultiValueDimensionSelector(DimensionSpec dimensionSpec) + { + final int idx = clusteringColumns.indexOf(dimensionSpec.getDimension()); + if (idx < 0) { + return new DelegatingMultiValueDimensionVectorSelector(this, dimensionSpec); + } + // Clustering values are single-typed primitives. Multi-value requests on a clustering column shouldn't happen + // in practice; throw to surface caller bugs rather than silently misbehave. + throw DruidException.defensive( + "multi-value vector selector not supported for clustering column [" + dimensionSpec.getDimension() + "]" + ); + } + + @Override + public VectorValueSelector makeValueSelector(String column) + { + final int idx = clusteringColumns.indexOf(column); + if (idx < 0) { + return new DelegatingVectorValueSelector(this, column); + } + return new ClusteringVectorValueSelector(this, idx); + } + + @Override + public VectorObjectSelector makeObjectSelector(String column) + { + final int idx = clusteringColumns.indexOf(column); + if (idx < 0) { + return new DelegatingVectorObjectSelector(this, column); + } + return new ClusteringVectorObjectSelector(this, idx); + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + final int idx = clusteringColumns.indexOf(column); + if (idx < 0) { + return delegate.getColumnCapabilities(column); + } + final ColumnType type = clusteringColumns.getColumnType(idx).orElseThrow(); + if (type.is(ValueType.STRING)) { + return ColumnCapabilitiesImpl.createSimpleSingleValueStringColumnCapabilities(); + } + return ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(type); + } + + private static final class ClusteringSingleValueDimensionVectorSelector implements SingleValueDimensionVectorSelector + { + private final ClusteringVectorColumnSelectorFactory parent; + private final int idx; + private final DimensionSpec spec; + private long cachedGeneration = -1; + private SingleValueDimensionVectorSelector cachedInner; + + private ClusteringSingleValueDimensionVectorSelector( + ClusteringVectorColumnSelectorFactory parent, + int idx, + DimensionSpec spec + ) + { + this.parent = parent; + this.idx = idx; + this.spec = spec; + } + + private SingleValueDimensionVectorSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration == currentGeneration) { + return cachedInner; + } + final Object raw = parent.currentValue(idx); + final String stringValue = raw == null ? null : String.valueOf(raw); + final String afterExtraction = + spec.getExtractionFn() == null ? stringValue : spec.getExtractionFn().apply(stringValue); + cachedInner = ConstantVectorSelectors.singleValueDimensionVectorSelector( + parent.getReadableVectorInspector(), + afterExtraction + ); + cachedGeneration = currentGeneration; + return cachedInner; + } + + @Override + public int[] getRowVector() + { + return currentInner().getRowVector(); + } + + @Override + public int getValueCardinality() + { + return currentInner().getValueCardinality(); + } + + @Nullable + @Override + public String lookupName(int id) + { + return currentInner().lookupName(id); + } + + @Override + public boolean nameLookupPossibleInAdvance() + { + return currentInner().nameLookupPossibleInAdvance(); + } + + @Nullable + @Override + public IdLookup idLookup() + { + return currentInner().idLookup(); + } + + @Override + public int getMaxVectorSize() + { + return parent.getMaxVectorSize(); + } + + @Override + public int getCurrentVectorSize() + { + return parent.getReadableVectorInspector().getCurrentVectorSize(); + } + } + + private static final class ClusteringVectorValueSelector implements VectorValueSelector + { + private final ClusteringVectorColumnSelectorFactory parent; + private final int idx; + private long cachedGeneration = -1; + private VectorValueSelector cachedInner; + + private ClusteringVectorValueSelector(ClusteringVectorColumnSelectorFactory parent, int idx) + { + this.parent = parent; + this.idx = idx; + } + + private VectorValueSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration == currentGeneration) { + return cachedInner; + } + final Object raw = parent.currentValue(idx); + // VectorValueSelector is meant for DOUBLE/FLOAT/LONG. STRING clustering shouldn't reach this method per the + // VectorColumnSelectorFactory contract; pass null (treated as null-numeric) to be conservative. + final Number number = (raw instanceof Number) ? (Number) raw : null; + cachedInner = ConstantVectorSelectors.vectorValueSelector(parent.getReadableVectorInspector(), number); + cachedGeneration = currentGeneration; + return cachedInner; + } + + @Override + public long[] getLongVector() + { + return currentInner().getLongVector(); + } + + @Override + public float[] getFloatVector() + { + return currentInner().getFloatVector(); + } + + @Override + public double[] getDoubleVector() + { + return currentInner().getDoubleVector(); + } + + @Nullable + @Override + public boolean[] getNullVector() + { + return currentInner().getNullVector(); + } + + @Override + public int getMaxVectorSize() + { + return parent.getMaxVectorSize(); + } + + @Override + public int getCurrentVectorSize() + { + return parent.getReadableVectorInspector().getCurrentVectorSize(); + } + } + + private static final class ClusteringVectorObjectSelector implements VectorObjectSelector + { + private final ClusteringVectorColumnSelectorFactory parent; + private final int idx; + private long cachedGeneration = -1; + private VectorObjectSelector cachedInner; + + private ClusteringVectorObjectSelector(ClusteringVectorColumnSelectorFactory parent, int idx) + { + this.parent = parent; + this.idx = idx; + } + + private VectorObjectSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration == currentGeneration) { + return cachedInner; + } + cachedInner = ConstantVectorSelectors.vectorObjectSelector( + parent.getReadableVectorInspector(), + parent.currentValue(idx) + ); + cachedGeneration = currentGeneration; + return cachedInner; + } + + @Override + public Object[] getObjectVector() + { + return currentInner().getObjectVector(); + } + + @Override + public int getMaxVectorSize() + { + return parent.getMaxVectorSize(); + } + + @Override + public int getCurrentVectorSize() + { + return parent.getReadableVectorInspector().getCurrentVectorSize(); + } + } + + private static final class DelegatingSingleValueDimensionVectorSelector implements SingleValueDimensionVectorSelector + { + private final ClusteringVectorColumnSelectorFactory parent; + private final DimensionSpec spec; + private long cachedGeneration = -1; + private SingleValueDimensionVectorSelector cachedInner; + + private DelegatingSingleValueDimensionVectorSelector( + ClusteringVectorColumnSelectorFactory parent, + DimensionSpec spec + ) + { + this.parent = parent; + this.spec = spec; + } + + private SingleValueDimensionVectorSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration != currentGeneration) { + cachedInner = parent.getDelegate().makeSingleValueDimensionSelector(spec); + cachedGeneration = currentGeneration; + } + return cachedInner; + } + + @Override + public int[] getRowVector() + { + return currentInner().getRowVector(); + } + + @Override + public int getValueCardinality() + { + return currentInner().getValueCardinality(); + } + + @Nullable + @Override + public String lookupName(int id) + { + return currentInner().lookupName(id); + } + + @Override + public boolean nameLookupPossibleInAdvance() + { + return currentInner().nameLookupPossibleInAdvance(); + } + + @Nullable + @Override + public IdLookup idLookup() + { + return currentInner().idLookup(); + } + + @Override + public int getMaxVectorSize() + { + return currentInner().getMaxVectorSize(); + } + + @Override + public int getCurrentVectorSize() + { + return currentInner().getCurrentVectorSize(); + } + } + + private static final class DelegatingMultiValueDimensionVectorSelector implements MultiValueDimensionVectorSelector + { + private final ClusteringVectorColumnSelectorFactory parent; + private final DimensionSpec spec; + private long cachedGeneration = -1; + private MultiValueDimensionVectorSelector cachedInner; + + private DelegatingMultiValueDimensionVectorSelector( + ClusteringVectorColumnSelectorFactory parent, + DimensionSpec spec + ) + { + this.parent = parent; + this.spec = spec; + } + + private MultiValueDimensionVectorSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration != currentGeneration) { + cachedInner = parent.getDelegate().makeMultiValueDimensionSelector(spec); + cachedGeneration = currentGeneration; + } + return cachedInner; + } + + @Override + public IndexedInts[] getRowVector() + { + return currentInner().getRowVector(); + } + + @Override + public int getValueCardinality() + { + return currentInner().getValueCardinality(); + } + + @Nullable + @Override + public String lookupName(int id) + { + return currentInner().lookupName(id); + } + + @Override + public boolean nameLookupPossibleInAdvance() + { + return currentInner().nameLookupPossibleInAdvance(); + } + + @Nullable + @Override + public IdLookup idLookup() + { + return currentInner().idLookup(); + } + + @Override + public int getMaxVectorSize() + { + return currentInner().getMaxVectorSize(); + } + + @Override + public int getCurrentVectorSize() + { + return currentInner().getCurrentVectorSize(); + } + } + + private static final class DelegatingVectorValueSelector implements VectorValueSelector + { + private final ClusteringVectorColumnSelectorFactory parent; + private final String column; + private long cachedGeneration = -1; + private VectorValueSelector cachedInner; + + private DelegatingVectorValueSelector(ClusteringVectorColumnSelectorFactory parent, String column) + { + this.parent = parent; + this.column = column; + } + + private VectorValueSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration != currentGeneration) { + cachedInner = parent.getDelegate().makeValueSelector(column); + cachedGeneration = currentGeneration; + } + return cachedInner; + } + + @Override + public long[] getLongVector() + { + return currentInner().getLongVector(); + } + + @Override + public float[] getFloatVector() + { + return currentInner().getFloatVector(); + } + + @Override + public double[] getDoubleVector() + { + return currentInner().getDoubleVector(); + } + + @Nullable + @Override + public boolean[] getNullVector() + { + return currentInner().getNullVector(); + } + + @Override + public int getMaxVectorSize() + { + return currentInner().getMaxVectorSize(); + } + + @Override + public int getCurrentVectorSize() + { + return currentInner().getCurrentVectorSize(); + } + } + + private static final class DelegatingVectorObjectSelector implements VectorObjectSelector + { + private final ClusteringVectorColumnSelectorFactory parent; + private final String column; + private long cachedGeneration = -1; + private VectorObjectSelector cachedInner; + + private DelegatingVectorObjectSelector(ClusteringVectorColumnSelectorFactory parent, String column) + { + this.parent = parent; + this.column = column; + } + + private VectorObjectSelector currentInner() + { + final long currentGeneration = parent.getGeneration(); + if (cachedGeneration != currentGeneration) { + cachedInner = parent.getDelegate().makeObjectSelector(column); + cachedGeneration = currentGeneration; + } + return cachedInner; + } + + @Override + public Object[] getObjectVector() + { + return currentInner().getObjectVector(); + } + + @Override + public int getMaxVectorSize() + { + return currentInner().getMaxVectorSize(); + } + + @Override + public int getCurrentVectorSize() + { + return currentInner().getCurrentVectorSize(); + } + } +} diff --git a/processing/src/main/java/org/apache/druid/segment/projections/ProjectionSchema.java b/processing/src/main/java/org/apache/druid/segment/projections/ProjectionSchema.java index e8c6b249de6c..2d9c28dd1be9 100644 --- a/processing/src/main/java/org/apache/druid/segment/projections/ProjectionSchema.java +++ b/processing/src/main/java/org/apache/druid/segment/projections/ProjectionSchema.java @@ -43,7 +43,8 @@ @JsonSubTypes(value = { @JsonSubTypes.Type(name = AggregateProjectionSpec.TYPE_NAME, value = AggregateProjectionSchema.class), @JsonSubTypes.Type(name = TableProjectionSchema.TYPE_NAME, value = TableProjectionSchema.class), - @JsonSubTypes.Type(name = RollupTableProjectionSchema.TYPE_NAME, value = RollupTableProjectionSchema.class) + @JsonSubTypes.Type(name = RollupTableProjectionSchema.TYPE_NAME, value = RollupTableProjectionSchema.class), + @JsonSubTypes.Type(name = ClusteredValueGroupsBaseTableSchema.TYPE_NAME, value = ClusteredValueGroupsBaseTableSchema.class) }) public interface ProjectionSchema { diff --git a/processing/src/main/java/org/apache/druid/segment/projections/Projections.java b/processing/src/main/java/org/apache/druid/segment/projections/Projections.java index cc037d730dae..19ce75b8cb2b 100644 --- a/processing/src/main/java/org/apache/druid/segment/projections/Projections.java +++ b/processing/src/main/java/org/apache/druid/segment/projections/Projections.java @@ -19,8 +19,11 @@ package org.apache.druid.segment.projections; +import com.google.common.base.Supplier; +import com.google.common.base.Suppliers; import com.google.common.collect.RangeSet; import org.apache.druid.data.input.impl.AggregateProjectionSpec; +import org.apache.druid.error.DruidException; import org.apache.druid.error.InvalidInput; import org.apache.druid.java.util.common.granularity.Granularities; import org.apache.druid.java.util.common.granularity.Granularity; @@ -30,18 +33,30 @@ import org.apache.druid.query.aggregation.FilteredAggregatorFactory; import org.apache.druid.query.cache.CacheKeyBuilder; import org.apache.druid.query.filter.DimFilter; +import org.apache.druid.query.filter.DruidPredicateMatch; +import org.apache.druid.query.filter.EqualityFilter; import org.apache.druid.query.filter.Filter; +import org.apache.druid.query.filter.NullFilter; +import org.apache.druid.query.filter.TypedInFilter; import org.apache.druid.segment.AggregateProjectionMetadata; import org.apache.druid.segment.CursorBuildSpec; import org.apache.druid.segment.CursorHolder; import org.apache.druid.segment.VirtualColumn; import org.apache.druid.segment.VirtualColumns; import org.apache.druid.segment.column.ColumnHolder; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.column.ValueType; +import org.apache.druid.segment.filter.AndFilter; +import org.apache.druid.segment.filter.NotFilter; +import org.apache.druid.segment.filter.OrFilter; import org.apache.druid.utils.CollectionUtils; import org.joda.time.DateTime; import org.joda.time.Interval; import javax.annotation.Nullable; +import java.util.ArrayList; +import java.util.Collections; import java.util.HashMap; import java.util.List; import java.util.Map; @@ -55,8 +70,11 @@ public class Projections { public static final String BASE_TABLE_PROJECTION_NAME = "__base"; + private static final String CLUSTER_GROUP_PREFIX = BASE_TABLE_PROJECTION_NAME + "$"; + private static final ConcurrentHashMap PERIOD_GRAN_CACHE = new ConcurrentHashMap<>(); + public static String validateProjectionName(@Nullable String name) { if (name == null || name.isEmpty()) { @@ -530,6 +548,241 @@ public static String getProjectionSegmentInternalFilePrefix(ProjectionSchema pro return projectionSchema.getName() + "/"; } + /** + * Check whether {@code type} is an allowed cluster group clustering-column type. Clustering is restricted to the + * primitive scalar types: {@link ValueType#STRING}, {@link ValueType#LONG}, {@link ValueType#DOUBLE}, + * {@link ValueType#FLOAT}. Complex and array types are rejected. + */ + public static boolean isAllowedClusteringType(@Nullable ColumnType type) + { + return type != null && type.anyOf(ValueType.STRING, ValueType.LONG, ValueType.DOUBLE, ValueType.FLOAT); + } + + /** + * Segment internal file prefix + column for a cluster group's per-group column data: + * {@code __base$_.../} + */ + public static String getClusterGroupSegmentInternalFileName(List clusteringValueIds, String column) + { + return getClusterGroupSegmentInternalFilePrefix(clusteringValueIds) + column; + } + + public static String getClusterGroupSegmentInternalFilePrefix(List clusteringValueIds) + { + if (clusteringValueIds == null || clusteringValueIds.isEmpty()) { + throw DruidException.defensive("clusteringValueIds must not be null or empty"); + } + final StringBuilder sb = new StringBuilder(CLUSTER_GROUP_PREFIX); + for (int i = 0; i < clusteringValueIds.size(); i++) { + if (i > 0) { + sb.append('_'); + } + sb.append(clusteringValueIds.get(i)); + } + sb.append('/'); + return sb.toString(); + } + + /** + * Returns the subset of {@code groups} that a query filter can't rule out from clustering values alone. + * Filters not referencing any clustering column are conservatively retained for every group. + */ + public static List pruneClusterGroups( + List groups, + @Nullable Filter filter, + @Nullable VirtualColumns queryVirtualColumns + ) + { + if (filter == null || groups.isEmpty()) { + return groups; + } + final VirtualColumns queryVcs = queryVirtualColumns == null ? VirtualColumns.EMPTY : queryVirtualColumns; + final List kept = new ArrayList<>(groups.size()); + for (TableClusterGroupSpec group : groups) { + if (matchesClusterGroup(filter, group, queryVcs)) { + kept.add(group); + } + } + return kept; + } + + private static boolean matchesClusterGroup( + Filter filter, + TableClusterGroupSpec group, + VirtualColumns queryVcs + ) + { + final ClusteredValueGroupsBaseTableSchema summary = group.getSummary(); + final RowSignature clusteringColumns = summary.getClusteringColumns(); + final VirtualColumns groupVcs = summary.getVirtualColumns(); + + // remap query-side column names that are virtual-column-equivalent to a clustering column + final Map remap = buildClusterGroupRemap( + filter.getRequiredColumns(), + clusteringColumns, + queryVcs, + groupVcs + ); + final Filter rewritten = remap.isEmpty() ? filter : filter.rewriteRequiredColumns(remap); + + // keep unless provably FALSE; UNKNOWN (filter references non-clustering data) keeps conservatively + return matchesClusterGroupFilter(rewritten, clusteringColumns, group.lookupClusteringValues()) + != DruidPredicateMatch.FALSE; + } + + /** + * Build a name-rewrite map so the pruner can walk the filter against a group's clustering tuple. Three cases per + * filter column: + *

    + *
  1. Query virtual column by that name (shadows any physical of the same name): prunable iff the group has an + * equivalent VC whose output is a clustering column. Same-name equivalence is an identity entry; different-name + * equivalence remaps. Otherwise, the column is remapped to a non-clustering sentinel so the pruner returns + * UNKNOWN, without that, a query virtual sharing a clustering column's name would be mis-matched against the + * clustering value.
    2. + *
  2. No query virtual column, but the name is a clustering column: identity entry, filter walks it directly.
    3. + *
  3. Neither query virtual column nor clustering column: identity entry, pruner returns UNKNOWN at that leaf.
    4. + *
+ * The query virtual column check must come first because query VC names are allowed to shadow physical/clustering + * column names. + *

+ * If no column needs a non-identity rewrite the result is empty and the caller skips the rewrite call. When any + * non-identity rewrite is present, identity entries are populated for all remaining required columns because + * {@link Filter#rewriteRequiredColumns} requires an entry for every referenced column (missing entries throw). + */ + private static Map buildClusterGroupRemap( + Set requiredColumns, + RowSignature clusteringColumns, + VirtualColumns queryVcs, + VirtualColumns groupVcs + ) + { + // if Query virtual column isn't equivalent to a clustering column, we remap to a sentinel name that is guaranteed + // not to be a clustering column so the pruner returns UNKNOWN for any leaf referencing it; compute it lazily + final Supplier sentinel = Suppliers.memoize(() -> { + String candidate = "__nonClusteringRef"; + while (clusteringColumns.indexOf(candidate) >= 0) { + candidate = "_" + candidate; + } + return candidate; + }); + + Map remap = null; + for (String col : requiredColumns) { + final VirtualColumns.Node queryNode = queryVcs.getNode(col); + if (queryNode == null) { + continue; + } + // query treats `col` as a virtual column, shadowing any physical/clustering column of the same name + final VirtualColumn equivalent = groupVcs.findEquivalent(queryNode); + final String target = equivalent != null && clusteringColumns.contains(equivalent.getOutputName()) + ? equivalent.getOutputName() + : null; + if (target == null) { + // query column has same name as a clustering column, but has no equivalent clustering virtual column; remap to + // use sentinel to be safe + if (remap == null) { + remap = new HashMap<>(); + } + remap.put(col, sentinel.get()); + } else if (!col.equals(target)) { + if (remap == null) { + remap = new HashMap<>(); + } + remap.put(col, target); + } + // else: same name, handled below if remap ends up non-empty + } + if (remap == null) { + return Collections.emptyMap(); + } + // fill identity entries for the remaining required columns; rewriteRequiredColumns rejects partial maps + for (String col : requiredColumns) { + remap.putIfAbsent(col, col); + } + return remap; + } + + /** + * Walk a (remapped) filter against a group's constant clustering values. Returns a {@link DruidPredicateMatch} + * 3VL result: TRUE = filter holds for every row in the group, FALSE = filter holds for no row (group can be + * pruned), UNKNOWN = can't decide from clustering values alone (filter references non-clustering data, or + * unrecognized filter type). UNKNOWN is distinct from FALSE so that {@code NOT(can't-tell)} stays + * {@code can't-tell} rather than flipping to "definitely false" and silently pruning live groups. + */ + private static DruidPredicateMatch matchesClusterGroupFilter( + Filter filter, + RowSignature clusteringColumns, + Object[] clusteringValues + ) + { + if (filter instanceof AndFilter andFilter) { + DruidPredicateMatch result = DruidPredicateMatch.TRUE; // identity for AND + for (Filter sub : andFilter.getFilters()) { + result = DruidPredicateMatch.and(result, matchesClusterGroupFilter(sub, clusteringColumns, clusteringValues)); + if (result == DruidPredicateMatch.FALSE) { + return result; // short-circuit: AND with false stays false + } + } + return result; + } + + if (filter instanceof OrFilter orFilter) { + DruidPredicateMatch result = DruidPredicateMatch.FALSE; // identity for OR + for (Filter sub : orFilter.getFilters()) { + result = DruidPredicateMatch.or(result, matchesClusterGroupFilter(sub, clusteringColumns, clusteringValues)); + if (result == DruidPredicateMatch.TRUE) { + return result; // short-circuit: OR with true stays true + } + } + return result; + } + + if (filter instanceof NotFilter notFilter) { + return DruidPredicateMatch.not( + matchesClusterGroupFilter(notFilter.getBaseFilter(), clusteringColumns, clusteringValues) + ); + } + + if (filter instanceof NullFilter isNull) { + final int idx = clusteringColumns.indexOf(isNull.getColumn()); + if (idx < 0) { + return DruidPredicateMatch.UNKNOWN; + } + return DruidPredicateMatch.of(clusteringValues[idx] == null); + } + + if (filter instanceof EqualityFilter eq) { + final int idx = clusteringColumns.indexOf(eq.getColumn()); + if (idx < 0) { + return DruidPredicateMatch.UNKNOWN; + } + // EqualityFilter doesn't match nulls; constructor also rejects null match values. + if (clusteringValues[idx] == null) { + return DruidPredicateMatch.FALSE; + } + return DruidPredicateMatch.of(Objects.equals(clusteringValues[idx], eq.getMatchValue())); + } + + if (filter instanceof TypedInFilter in) { + final int idx = clusteringColumns.indexOf(in.getColumn()); + if (idx < 0) { + return DruidPredicateMatch.UNKNOWN; + } + // TypedInFilter matches nulls if present in the values list. Iterate explicitly — immutable List impls + // (List.of, ImmutableList) NPE on contains(null). + final Object val = clusteringValues[idx]; + for (Object v : in.getSortedValues()) { + if (Objects.equals(v, val)) { + return DruidPredicateMatch.TRUE; + } + } + return DruidPredicateMatch.FALSE; + } + + // anything else, fall through as UNKNOWN (this could be improved in the future) + return DruidPredicateMatch.UNKNOWN; + } + /** * Check that the query {@link CursorBuildSpec} either contains the entire data interval, or that the query interval * is aligned with {@link AggregateProjectionSchema#getEffectiveGranularity()} diff --git a/processing/src/main/java/org/apache/druid/segment/projections/TableClusterGroupSpec.java b/processing/src/main/java/org/apache/druid/segment/projections/TableClusterGroupSpec.java new file mode 100644 index 000000000000..96d1196cab76 --- /dev/null +++ b/processing/src/main/java/org/apache/druid/segment/projections/TableClusterGroupSpec.java @@ -0,0 +1,176 @@ +/* + * 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.druid.segment.projections; + +import com.fasterxml.jackson.annotation.JsonCreator; +import com.fasterxml.jackson.annotation.JsonInclude; +import com.fasterxml.jackson.annotation.JsonProperty; +import com.fasterxml.jackson.annotation.JsonTypeInfo; +import org.apache.druid.error.DruidException; +import org.apache.druid.segment.column.RowSignature; + +import javax.annotation.Nullable; +import java.util.Arrays; +import java.util.List; +import java.util.Objects; + +/** + * A single cluster group within a clustered base table, with all rows sharing the clustering-column tuple given by + * {@link #lookupClusteringValues()}. + *

+ * Shape is very minimal containing only clustering value ids and row count; other things such as columns, ordering, + * dictionaries, etc. live on the parent {@link ClusteredValueGroupsBaseTableSchema} as they are identical from group + * to group within a segment. The parent summary is attached after construction via {@link #setSummary} and reached + * via {@link #getSummary}. + */ +@JsonTypeInfo(use = JsonTypeInfo.Id.NAME, property = "type") +public class TableClusterGroupSpec +{ + public static final String TYPE_NAME = "cluster-base-table"; + + /** + * Clustering value id tuple (positions in the parent summary's per-column dictionaries). + */ + private final List clusteringValueIds; + + /** + * Number of rows in this cluster group's per-group column data. + */ + private final int numRows; + + /** + * Parent back-reference, set once via {@link #setSummary} during the schema constructor. Readers observe the + * spec only after the holding {@code SimpleQueryableIndex}'s final field publishes it, so non-final is safe. + */ + @Nullable + private ClusteredValueGroupsBaseTableSchema summary; + + @JsonCreator + public TableClusterGroupSpec( + @JsonProperty("clusteringValueIds") List clusteringValueIds, + @JsonProperty("numRows") @Nullable Integer numRows + ) + { + if (clusteringValueIds == null) { + throw DruidException.defensive("clusteringValueIds must not be null"); + } + this.clusteringValueIds = List.copyOf(clusteringValueIds); + this.numRows = numRows == null ? 0 : numRows; + this.summary = null; + } + + /** + * Wire up the parent summary. Called once per spec by {@link ClusteredValueGroupsBaseTableSchema}'s constructor. + */ + public void setSummary(ClusteredValueGroupsBaseTableSchema summary) + { + DruidException.conditionalDefensive(summary != null, "summary must not be null"); + DruidException.conditionalDefensive(this.summary == null, "summary already set"); + + final RowSignature clusteringColumns = summary.getClusteringColumns(); + final int numCols = clusteringColumns.size(); + + DruidException.conditionalDefensive( + clusteringValueIds.size() == numCols, + "clusteringValueIds size [%s] does not match summary clusteringColumns size [%s]", + clusteringValueIds.size(), + numCols + ); + + this.summary = summary; + } + + /** + * Returns the parent summary. Throws if {@link #setSummary} hasn't been called yet. + */ + public ClusteredValueGroupsBaseTableSchema getSummary() + { + DruidException.conditionalDefensive( + this.summary != null, + "TableClusterGroupSpec.setSummary must be called before this method" + ); + return summary; + } + + /** + * Typed clustering values for this group, materialized fresh from the summary's dictionaries on each call. + * Length and order match {@link ClusteredValueGroupsBaseTableSchema#getClusteringColumns()}. Callers that walk + * the array repeatedly should cache the returned reference. + */ + public Object[] lookupClusteringValues() + { + final ClusteredValueGroupsBaseTableSchema s = getSummary(); + final int numCols = clusteringValueIds.size(); + final Object[] out = new Object[numCols]; + for (int i = 0; i < numCols; i++) { + out[i] = s.lookupClusteringValue(i, clusteringValueIds.get(i)); + } + return out; + } + + /** + * Dictionary ids identifying this group, one per clustering column. + */ + @JsonProperty + public List getClusteringValueIds() + { + return clusteringValueIds; + } + + @JsonProperty + @JsonInclude(JsonInclude.Include.NON_DEFAULT) + public int getNumRows() + { + return numRows; + } + + @Override + public boolean equals(Object o) + { + if (this == o) { + return true; + } + if (o == null || getClass() != o.getClass()) { + return false; + } + TableClusterGroupSpec that = (TableClusterGroupSpec) o; + // Specs are only compared within a single segment context, where segment-local IDs uniquely identify a tuple. + return Objects.equals(clusteringValueIds, that.clusteringValueIds); + } + + @Override + public int hashCode() + { + return Objects.hashCode(clusteringValueIds); + } + + @Override + public String toString() + { + if (summary == null) { + return "TableClusterGroupSpec{" + + "clusteringValues=" + "" + + '}'; + } + return "TableClusterGroupSpec{" + + "clusteringValues=" + Arrays.toString(lookupClusteringValues()) + + '}'; + } +} diff --git a/processing/src/main/java/org/apache/druid/segment/vector/ConcatenatingVectorCursor.java b/processing/src/main/java/org/apache/druid/segment/vector/ConcatenatingVectorCursor.java new file mode 100644 index 000000000000..55f374cd4a58 --- /dev/null +++ b/processing/src/main/java/org/apache/druid/segment/vector/ConcatenatingVectorCursor.java @@ -0,0 +1,158 @@ +/* + * 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.druid.segment.vector; + +import com.google.common.base.Supplier; +import org.apache.druid.error.DruidException; +import org.apache.druid.segment.CursorHolder; +import org.apache.druid.segment.projections.ClusteringVectorColumnSelectorFactory; + +import java.util.List; + +/** + * Vector-cursor counterpart of {@link org.apache.druid.segment.ConcatenatingCursor}. Walks a sequence of per-group + * {@link VectorCursor}s back-to-back, presenting them to the caller as a single vector cursor over a clustered base + * table. Vector boundaries do not align with cluster-group boundaries; the last vector of each group is naturally + * partial (its {@link #getCurrentVectorSize()} returns less than max), and the next {@link #advance()} swaps in the + * next group's vector cursor without trying to merge across groups. + *

On each group transition the wrapper {@link ClusteringVectorColumnSelectorFactory} updates its underlying + * delegate and clustering values, so previously-acquired delegating vector selectors observe the new group's data via + * generation-counter cache invalidation. + *

+ * The outer {@link CursorHolder} owns the lifetime of the per-group holders (typically by registering each one + * with a {@link org.apache.druid.java.util.common.io.Closer} as part of the supplier itself) + */ +public final class ConcatenatingVectorCursor implements VectorCursor +{ + private final List> holderSuppliers; + private final List clusteringValuesByGroup; + private final ClusteringVectorColumnSelectorFactory wrapperFactory; + + private int currentIdx; + private VectorCursor currentCursor; + private boolean initialized; + + public ConcatenatingVectorCursor( + List> holderSuppliers, + List clusteringValuesByGroup, + ClusteringVectorColumnSelectorFactory wrapperFactory + ) + { + if (holderSuppliers.size() != clusteringValuesByGroup.size()) { + throw DruidException.defensive( + "holderSuppliers size [%s] must equal clusteringValuesByGroup size [%s]", + holderSuppliers.size(), + clusteringValuesByGroup.size() + ); + } + if (holderSuppliers.isEmpty()) { + throw DruidException.defensive("ConcatenatingVectorCursor requires at least one cluster group"); + } + this.holderSuppliers = holderSuppliers; + this.clusteringValuesByGroup = clusteringValuesByGroup; + this.wrapperFactory = wrapperFactory; + this.currentIdx = -1; + } + + private void initializeIfNeeded() + { + if (initialized) { + return; + } + initialized = true; + advanceToNextNonEmptyGroup(); + } + + /** + * Open the next group whose vector cursor has at least one row. Skips empty groups. Sets {@code currentCursor = + * null} when all groups are exhausted. + */ + private void advanceToNextNonEmptyGroup() + { + while (++currentIdx < holderSuppliers.size()) { + final CursorHolder holder = holderSuppliers.get(currentIdx).get(); + final VectorCursor cursor = holder.asVectorCursor(); + if (cursor != null && !cursor.isDone()) { + currentCursor = cursor; + wrapperFactory.setDelegate(cursor.getColumnSelectorFactory(), clusteringValuesByGroup.get(currentIdx)); + return; + } + // Group has no rows after filter application; try the next. + } + currentCursor = null; + } + + @Override + public VectorColumnSelectorFactory getColumnSelectorFactory() + { + initializeIfNeeded(); + return wrapperFactory; + } + + @Override + public void advance() + { + initializeIfNeeded(); + if (currentCursor == null) { + return; + } + currentCursor.advance(); + if (currentCursor.isDone()) { + advanceToNextNonEmptyGroup(); + } + } + + @Override + public boolean isDone() + { + initializeIfNeeded(); + return currentCursor == null; + } + + /** + * Rewind to before the first group. Does not close any per-group holders, those are owned by the outer + * {@link CursorHolder}. Subsequent {@link #advance()} / {@link #isDone()} re-fetch each group's vector cursor via + * {@link CursorHolder#asVectorCursor}. + */ + @Override + public void reset() + { + currentIdx = -1; + currentCursor = null; + initialized = false; + } + + @Override + public int getMaxVectorSize() + { + return wrapperFactory.getMaxVectorSize(); + } + + @Override + public int getCurrentVectorSize() + { + initializeIfNeeded(); + if (currentCursor == null) { + return 0; + } + return currentCursor.getCurrentVectorSize(); + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/ConcatenatingCursorTest.java b/processing/src/test/java/org/apache/druid/segment/ConcatenatingCursorTest.java new file mode 100644 index 000000000000..36d2f8c547fc --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/ConcatenatingCursorTest.java @@ -0,0 +1,478 @@ +/* + * 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.druid.segment; + +import com.google.common.base.Supplier; +import com.google.common.base.Suppliers; +import org.apache.druid.java.util.common.io.Closer; +import org.apache.druid.query.dimension.DimensionSpec; +import org.apache.druid.query.monomorphicprocessing.RuntimeShapeInspector; +import org.apache.druid.segment.column.ColumnCapabilities; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.projections.ClusteringColumnSelectorFactory; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import javax.annotation.Nullable; +import java.util.ArrayList; +import java.util.List; + +class ConcatenatingCursorTest +{ + private static final RowSignature CLUSTER_SIGNATURE = RowSignature.builder().add("tenant", ColumnType.STRING).build(); + + private final Closer closer = Closer.create(); + + @Test + void testWalksTwoNonEmptyGroupsBackToBack() + { + FakeCursorHolder a = new FakeCursorHolder(List.of("a1", "a2")); + FakeCursorHolder b = new FakeCursorHolder(List.of("b1")); + + ClusteringColumnSelectorFactory wrapper = new ClusteringColumnSelectorFactory( + new FakeFactory(List.of(), new int[]{0}), + CLUSTER_SIGNATURE, + new Object[]{"acme"} + ); + + ConcatenatingCursor c = new ConcatenatingCursor( + List.of(holderSupplier(a), holderSupplier(b)), + List.of(new Object[]{"acme"}, new Object[]{"globex"}), + wrapper + ); + + ColumnValueSelector tenant = c.getColumnSelectorFactory().makeColumnValueSelector("tenant"); + ColumnValueSelector metric = c.getColumnSelectorFactory().makeColumnValueSelector("metric"); + + // Group 1: tenant=acme, metric=a1 + Assertions.assertEquals("acme", tenant.getObject()); + Assertions.assertEquals("a1", metric.getObject()); + c.advance(); + + // Group 1: tenant=acme, metric=a2 + Assertions.assertEquals("acme", tenant.getObject()); + Assertions.assertEquals("a2", metric.getObject()); + c.advance(); + + // Group transition → tenant=globex, metric=b1 + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals("globex", tenant.getObject()); + Assertions.assertEquals("b1", metric.getObject()); + c.advance(); + + Assertions.assertTrue(c.isDone()); + Assertions.assertDoesNotThrow(closer::close); + Assertions.assertTrue(a.closed); + Assertions.assertTrue(b.closed); + } + + @Test + void testSkipsLeadingEmptyGroup() + { + FakeCursorHolder empty = new FakeCursorHolder(List.of()); + FakeCursorHolder full = new FakeCursorHolder(List.of("x")); + + ClusteringColumnSelectorFactory wrapper = new ClusteringColumnSelectorFactory( + new FakeFactory(List.of(), new int[]{0}), + CLUSTER_SIGNATURE, + new Object[]{"placeholder"} + ); + + ConcatenatingCursor c = new ConcatenatingCursor( + List.of(holderSupplier(empty), holderSupplier(full)), + List.of(new Object[]{"a"}, new Object[]{"b"}), + wrapper + ); + + ColumnValueSelector tenant = c.getColumnSelectorFactory().makeColumnValueSelector("tenant"); + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals("b", tenant.getObject()); + c.advance(); + Assertions.assertTrue(c.isDone()); + Assertions.assertDoesNotThrow(closer::close); + Assertions.assertTrue(empty.closed); + Assertions.assertTrue(full.closed); + } + + @Test + void testSkipsTrailingEmptyGroup() + { + FakeCursorHolder full = new FakeCursorHolder(List.of("x")); + FakeCursorHolder empty = new FakeCursorHolder(List.of()); + + ClusteringColumnSelectorFactory wrapper = new ClusteringColumnSelectorFactory( + new FakeFactory(List.of(), new int[]{0}), + CLUSTER_SIGNATURE, + new Object[]{"a"} + ); + + ConcatenatingCursor c = new ConcatenatingCursor( + List.of(holderSupplier(full), holderSupplier(empty)), + List.of(new Object[]{"a"}, new Object[]{"b"}), + wrapper + ); + + ColumnValueSelector tenant = c.getColumnSelectorFactory().makeColumnValueSelector("tenant"); + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals("a", tenant.getObject()); + c.advance(); + Assertions.assertTrue(c.isDone()); + } + + @Test + void testAllEmptyGroups() + { + FakeCursorHolder e1 = new FakeCursorHolder(List.of()); + FakeCursorHolder e2 = new FakeCursorHolder(List.of()); + + ClusteringColumnSelectorFactory wrapper = new ClusteringColumnSelectorFactory( + new FakeFactory(List.of(), new int[]{0}), + CLUSTER_SIGNATURE, + new Object[]{"placeholder"} + ); + + ConcatenatingCursor c = new ConcatenatingCursor( + List.of(holderSupplier(e1), holderSupplier(e2)), + List.of(new Object[]{"a"}, new Object[]{"b"}), + wrapper + ); + + Assertions.assertTrue(c.isDone()); + } + + @Test + void testSingleGroupDegenerateCase() + { + FakeCursorHolder only = new FakeCursorHolder(List.of("x", "y")); + + ClusteringColumnSelectorFactory wrapper = new ClusteringColumnSelectorFactory( + new FakeFactory(List.of(), new int[]{0}), + CLUSTER_SIGNATURE, + new Object[]{"a"} + ); + + ConcatenatingCursor c = new ConcatenatingCursor( + List.of(holderSupplier(only)), + List.of(new Object[]{"a"}), + wrapper + ); + + ColumnValueSelector tenant = c.getColumnSelectorFactory().makeColumnValueSelector("tenant"); + ColumnValueSelector metric = c.getColumnSelectorFactory().makeColumnValueSelector("metric"); + + Assertions.assertEquals("a", tenant.getObject()); + Assertions.assertEquals("x", metric.getObject()); + c.advance(); + Assertions.assertEquals("a", tenant.getObject()); + Assertions.assertEquals("y", metric.getObject()); + c.advance(); + Assertions.assertTrue(c.isDone()); + } + + @Test + void testCloseClosesAllOpenedHolders() + { + FakeCursorHolder a = new FakeCursorHolder(List.of("a1")); + FakeCursorHolder b = new FakeCursorHolder(List.of("b1")); + + ClusteringColumnSelectorFactory wrapper = new ClusteringColumnSelectorFactory( + new FakeFactory(List.of(), new int[]{0}), + CLUSTER_SIGNATURE, + new Object[]{"x"} + ); + + ConcatenatingCursor c = new ConcatenatingCursor( + List.of(holderSupplier(a), holderSupplier(b)), + List.of(new Object[]{"x"}, new Object[]{"y"}), + wrapper + ); + + // Walk through, opens both holders. + c.getColumnSelectorFactory(); + c.advance(); + c.advance(); // exhausts + Assertions.assertDoesNotThrow(closer::close); + Assertions.assertTrue(a.closed); + Assertions.assertTrue(b.closed); + } + + @Test + void testGroupsAreOpenedLazilyOnTransitionNotEagerly() + { + final boolean[] secondOpened = {false}; + FakeCursorHolder first = new FakeCursorHolder(List.of("x")); + + ClusteringColumnSelectorFactory wrapper = new ClusteringColumnSelectorFactory( + new FakeFactory(List.of(), new int[]{0}), + CLUSTER_SIGNATURE, + new Object[]{"a"} + ); + + List> suppliers = new ArrayList<>(); + suppliers.add(holderSupplier(first)); + suppliers.add(() -> { + secondOpened[0] = true; + return new FakeCursorHolder(List.of("y")); + }); + + ConcatenatingCursor c = new ConcatenatingCursor( + suppliers, + List.of(new Object[]{"a"}, new Object[]{"b"}), + wrapper + ); + + // Init opens first group only. Second is still untouched. + c.getColumnSelectorFactory(); + Assertions.assertFalse(secondOpened[0]); + + c.advance(); // exhausts first → opens second + Assertions.assertTrue(secondOpened[0]); + } + + @Test + void testResetReIteratesWithoutClosingHolders() + { + // reset() must not close per-group holders, those are owned by the outer Closer. After reset, the cursor + // should re-iterate from the start; the suppliers are memoized so they return the same holders, and + // CursorHolder.asCursor() returns a fresh cursor each call. + FakeCursorHolder a = new FakeCursorHolder(List.of("a1", "a2")); + FakeCursorHolder b = new FakeCursorHolder(List.of("b1")); + + ClusteringColumnSelectorFactory wrapper = new ClusteringColumnSelectorFactory( + new FakeFactory(List.of(), new int[]{0}), + CLUSTER_SIGNATURE, + new Object[]{"acme"} + ); + + ConcatenatingCursor c = new ConcatenatingCursor( + List.of(holderSupplier(a), holderSupplier(b)), + List.of(new Object[]{"acme"}, new Object[]{"globex"}), + wrapper + ); + + ColumnValueSelector tenant = c.getColumnSelectorFactory().makeColumnValueSelector("tenant"); + ColumnValueSelector metric = c.getColumnSelectorFactory().makeColumnValueSelector("metric"); + + // Walk to exhaustion. + Assertions.assertEquals("a1", metric.getObject()); + c.advance(); + Assertions.assertEquals("a2", metric.getObject()); + c.advance(); + Assertions.assertEquals("b1", metric.getObject()); + c.advance(); + Assertions.assertTrue(c.isDone()); + + // Holders MUST remain open across reset; only the outer Closer closes them. + Assertions.assertFalse(a.closed); + Assertions.assertFalse(b.closed); + + c.reset(); + + // Re-walk from the start, exercising both groups again. + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals("acme", tenant.getObject()); + Assertions.assertEquals("a1", metric.getObject()); + c.advance(); + Assertions.assertEquals("a2", metric.getObject()); + c.advance(); + Assertions.assertEquals("globex", tenant.getObject()); + Assertions.assertEquals("b1", metric.getObject()); + c.advance(); + Assertions.assertTrue(c.isDone()); + } + + private Supplier holderSupplier(FakeCursorHolder h) + { + return Suppliers.memoize(() -> closer.register(h))::get; + } + + private static class FakeStringValueSelector implements ColumnValueSelector + { + private final List rows; + private final int[] offset; + + FakeStringValueSelector(List rows, int[] offset) + { + this.rows = rows; + this.offset = offset; + } + + @Override + public double getDouble() + { + return 0; + } + + @Override + public float getFloat() + { + return 0; + } + + @Override + public long getLong() + { + return 0; + } + + @Override + public boolean isNull() + { + return rows.get(offset[0]) == null; + } + + @Nullable + @Override + public Object getObject() + { + return rows.get(offset[0]); + } + + @Override + public Class classOfObject() + { + return String.class; + } + + @Override + public void inspectRuntimeShape(RuntimeShapeInspector inspector) + { + } + } + + private static class FakeFactory implements ColumnSelectorFactory + { + private final List metricRows; + private final int[] offset; + + FakeFactory(List metricRows, int[] offset) + { + this.metricRows = metricRows; + this.offset = offset; + } + + @Override + public DimensionSelector makeDimensionSelector(DimensionSpec dimensionSpec) + { + throw new UnsupportedOperationException("not used in these tests"); + } + + @Override + public ColumnValueSelector makeColumnValueSelector(String columnName) + { + if ("metric".equals(columnName)) { + return new FakeStringValueSelector(metricRows, offset); + } + return NilColumnValueSelector.instance(); + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + return null; + } + } + + private static class FakeCursor implements Cursor + { + private final List rows; + private final int[] offset = new int[]{0}; + private final FakeFactory factory; + + FakeCursor(List rows) + { + this.rows = rows; + this.factory = new FakeFactory(rows, offset); + } + + @Override + public ColumnSelectorFactory getColumnSelectorFactory() + { + return factory; + } + + @Override + public void advance() + { + offset[0]++; + } + + @Override + public void advanceUninterruptibly() + { + offset[0]++; + } + + @Override + public boolean isDone() + { + return offset[0] >= rows.size(); + } + + @Override + public boolean isDoneOrInterrupted() + { + return isDone(); + } + + @Override + public void reset() + { + offset[0] = 0; + } + } + + private static class FakeCursorHolder implements CursorHolder + { + private final List rows; + private boolean closed; + + FakeCursorHolder(List rows) + { + this.rows = rows; + } + + @Override + public Cursor asCursor() + { + return new FakeCursor(rows); + } + + @Override + public boolean canVectorize() + { + return false; + } + + @Override + public List getOrdering() + { + return List.of(); + } + + @Override + public void close() + { + closed = true; + } + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/QueryableIndexCursorFactoryClusteredTest.java b/processing/src/test/java/org/apache/druid/segment/QueryableIndexCursorFactoryClusteredTest.java new file mode 100644 index 000000000000..881ddae87720 --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/QueryableIndexCursorFactoryClusteredTest.java @@ -0,0 +1,340 @@ +/* + * 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.druid.segment; + +import org.apache.druid.collections.bitmap.BitmapFactory; +import org.apache.druid.collections.bitmap.RoaringBitmapFactory; +import org.apache.druid.java.util.common.Intervals; +import org.apache.druid.query.OrderBy; +import org.apache.druid.query.aggregation.AggregatorFactory; +import org.apache.druid.query.aggregation.CountAggregatorFactory; +import org.apache.druid.segment.column.BaseColumnHolder; +import org.apache.druid.segment.column.ColumnCapabilities; +import org.apache.druid.segment.column.ColumnCapabilitiesImpl; +import org.apache.druid.segment.column.ColumnHolder; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.column.ValueType; +import org.apache.druid.segment.data.Indexed; +import org.apache.druid.segment.data.ListIndexed; +import org.apache.druid.segment.projections.ClusterGroupSchemaTestHelpers; +import org.apache.druid.segment.projections.ClusteredValueGroupsBaseTableSchema; +import org.apache.druid.segment.projections.TableClusterGroupSpec; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import javax.annotation.Nullable; +import java.util.Collections; +import java.util.List; +import java.util.Map; +import java.util.function.Function; + +class QueryableIndexCursorFactoryClusteredTest +{ + /** + * Build a clustered summary from typed tenant values. The helper derives dictionaries and dictionary-IDs for each + * group; the resulting specs are nested in the summary and wired via setSummary inside the schema constructor. + */ + private static ClusteredValueGroupsBaseTableSchema summary(String... tenants) + { + final RowSignature clustering = RowSignature.builder().add("tenant", ColumnType.STRING).build(); + final List> tuples = new java.util.ArrayList<>(tenants.length); + for (String t : tenants) { + tuples.add(java.util.Arrays.asList(t)); + } + final ClusterGroupSchemaTestHelpers.Built built = ClusterGroupSchemaTestHelpers.buildClusterGroups( + clustering, + tuples + ); + return new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of("tenant", ColumnHolder.TIME_COLUMN_NAME, "metric"), + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + List.of(OrderBy.ascending("tenant"), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + clustering, + null, + built.dictionaries(), + built.specs() + ); + } + + /** + * Bare-bones {@link QueryableIndex} that exposes the clustered cluster summary + groups via the interface + * defaults, plus a stubbed {@link #getClusterGroupQueryableIndex} that returns a fake group sub-index whose + * column capabilities are programmable. Used to drive {@link QueryableIndexCursorFactory#getRowSignature} and + * {@link QueryableIndexCursorFactory#getColumnCapabilities} for clustered segments without standing up a real + * V10 segment file. + */ + private static class FakeClusteredIndex implements QueryableIndex + { + private final ClusteredValueGroupsBaseTableSchema summary; + private final Function dataColumnCaps; + + FakeClusteredIndex( + ClusteredValueGroupsBaseTableSchema summary, + Function dataColumnCaps + ) + { + this.summary = summary; + this.dataColumnCaps = dataColumnCaps; + } + + @Override + public org.joda.time.Interval getDataInterval() + { + return Intervals.of("2025-01-01/2025-01-02"); + } + + @Override + public int getNumRows() + { + return 0; + } + + @Override + public Indexed getAvailableDimensions() + { + return new ListIndexed<>(List.of()); + } + + @Override + public BitmapFactory getBitmapFactoryForDimensions() + { + return new RoaringBitmapFactory(); + } + + @Nullable + @Override + public Metadata getMetadata() + { + return null; + } + + @Override + public Map getDimensionHandlers() + { + return Collections.emptyMap(); + } + + @Override + public List getColumnNames() + { + return Collections.emptyList(); + } + + @Nullable + @Override + public BaseColumnHolder getColumnHolder(String columnName) + { + return null; + } + + @Override + public List getOrdering() + { + return summary.getOrdering(); + } + + @Override + public void close() + { + } + + @Nullable + @Override + public ClusteredValueGroupsBaseTableSchema getClusteredBaseSummary() + { + return summary; + } + + @Override + public List getClusterGroupSchemas() + { + return summary.getClusterGroups(); + } + + @Nullable + @Override + public QueryableIndex getClusterGroupQueryableIndex(TableClusterGroupSpec groupSpec) + { + // Return a fake sub-index that answers getColumnCapabilities from the supplied function. Other methods + // throw if reached so we'd notice if the cursor factory's clustered paths start needing more. + return new QueryableIndex() + { + @Override + public org.joda.time.Interval getDataInterval() + { + return Intervals.of("2025-01-01/2025-01-02"); + } + + @Override + public int getNumRows() + { + return 0; + } + + @Override + public Indexed getAvailableDimensions() + { + return new ListIndexed<>(List.of()); + } + + @Override + public BitmapFactory getBitmapFactoryForDimensions() + { + return new RoaringBitmapFactory(); + } + + @Nullable + @Override + public Metadata getMetadata() + { + return null; + } + + @Override + public Map getDimensionHandlers() + { + return Collections.emptyMap(); + } + + @Override + public List getColumnNames() + { + return Collections.emptyList(); + } + + @Nullable + @Override + public BaseColumnHolder getColumnHolder(String columnName) + { + return null; + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + return dataColumnCaps.apply(column); + } + + @Override + public List getOrdering() + { + return Collections.emptyList(); + } + + @Override + public void close() + { + } + }; + } + } + + private static ColumnCapabilities longCaps() + { + return ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(ColumnType.LONG); + } + + private static ColumnCapabilities timeCaps() + { + return ColumnCapabilitiesImpl.createSimpleNumericColumnCapabilities(ColumnType.LONG); + } + + @Test + void testGetRowSignatureSourcesClusteringColumnFromSummaryAndDataColumnFromGroup() + { + final FakeClusteredIndex index = new FakeClusteredIndex( + summary("acme", "globex"), + col -> { + if (ColumnHolder.TIME_COLUMN_NAME.equals(col)) { + return timeCaps(); + } + if ("metric".equals(col)) { + return longCaps(); + } + return null; + } + ); + + QueryableIndexCursorFactory factory = new QueryableIndexCursorFactory( + index, + QueryableIndexTimeBoundaryInspector.create(index) + ); + + RowSignature sig = factory.getRowSignature(); + Assertions.assertEquals(ColumnType.STRING, sig.getColumnType("tenant").orElseThrow()); + Assertions.assertEquals(ColumnType.LONG, sig.getColumnType(ColumnHolder.TIME_COLUMN_NAME).orElseThrow()); + Assertions.assertEquals(ColumnType.LONG, sig.getColumnType("metric").orElseThrow()); + } + + @Test + void testGetColumnCapabilitiesForClusteringColumnFromSummary() + { + final FakeClusteredIndex index = new FakeClusteredIndex( + summary("acme"), + col -> null + ); + + QueryableIndexCursorFactory factory = new QueryableIndexCursorFactory( + index, + QueryableIndexTimeBoundaryInspector.create(index) + ); + + ColumnCapabilities tenantCaps = factory.getColumnCapabilities("tenant"); + Assertions.assertNotNull(tenantCaps); + Assertions.assertTrue(tenantCaps.is(ValueType.STRING)); + } + + @Test + void testGetColumnCapabilitiesForDataColumnFromFirstGroup() + { + final FakeClusteredIndex index = new FakeClusteredIndex( + summary("acme", "globex"), + col -> "metric".equals(col) ? longCaps() : null + ); + + QueryableIndexCursorFactory factory = new QueryableIndexCursorFactory( + index, + QueryableIndexTimeBoundaryInspector.create(index) + ); + + ColumnCapabilities metricCaps = factory.getColumnCapabilities("metric"); + Assertions.assertNotNull(metricCaps); + Assertions.assertTrue(metricCaps.is(ValueType.LONG)); + } + + @Test + void testGetColumnCapabilitiesForUnknownColumnIsNull() + { + final FakeClusteredIndex index = new FakeClusteredIndex( + summary("acme"), + col -> null + ); + + QueryableIndexCursorFactory factory = new QueryableIndexCursorFactory( + index, + QueryableIndexTimeBoundaryInspector.create(index) + ); + + Assertions.assertNull(factory.getColumnCapabilities("unknown")); + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/SimpleQueryableIndexClusteredTest.java b/processing/src/test/java/org/apache/druid/segment/SimpleQueryableIndexClusteredTest.java new file mode 100644 index 000000000000..8860c08e2686 --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/SimpleQueryableIndexClusteredTest.java @@ -0,0 +1,196 @@ +/* + * 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.druid.segment; + +import com.google.common.base.Supplier; +import org.apache.druid.collections.bitmap.RoaringBitmapFactory; +import org.apache.druid.java.util.common.Intervals; +import org.apache.druid.query.OrderBy; +import org.apache.druid.query.aggregation.AggregatorFactory; +import org.apache.druid.query.aggregation.CountAggregatorFactory; +import org.apache.druid.segment.column.BaseColumnHolder; +import org.apache.druid.segment.column.ColumnHolder; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.data.ListIndexed; +import org.apache.druid.segment.projections.ClusterGroupSchemaTestHelpers; +import org.apache.druid.segment.projections.ClusteredValueGroupsBaseTableSchema; +import org.apache.druid.segment.projections.TableClusterGroupSpec; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import java.util.List; +import java.util.Map; + +/** + * Coverage for {@link SimpleQueryableIndex}'s clustered base table awareness: the constructor accepts an empty + * columns map for clustered segments, and the cluster-aware helpers surface the summary and per-group schemas that + * {@link org.apache.druid.segment.IndexIO}'s V10 loader collects from a clustered segment. + */ +class SimpleQueryableIndexClusteredTest +{ + /** + * Build a one-column STRING-clustered summary plus its specs from typed tenant values. Returns the helper's + * {@link ClusterGroupSchemaTestHelpers.Built} bundle plus the constructed schema (whose {@code clusterGroups} are + * the same spec instances as {@code built.specs()}, wired via setSummary by the schema constructor). + */ + private static ClusterGroupSchemaTestHelpers.Built summary(String... tenants) + { + final RowSignature clustering = RowSignature.builder().add("tenant", ColumnType.STRING).build(); + final List> tuples = new java.util.ArrayList<>(tenants.length); + for (String t : tenants) { + tuples.add(java.util.Arrays.asList(t)); + } + final ClusterGroupSchemaTestHelpers.Built built = ClusterGroupSchemaTestHelpers.buildClusterGroups( + clustering, + tuples + ); + new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of("tenant", ColumnHolder.TIME_COLUMN_NAME, "region", "metric"), + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + List.of( + OrderBy.ascending("tenant"), + OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME), + OrderBy.ascending("region") + ), + clustering, + null, + built.dictionaries(), + built.specs() + ); + return built; + } + + private static ClusteredValueGroupsBaseTableSchema summarySchema(String... tenants) + { + return summary(tenants).specs().get(0).getSummary(); + } + + private static SimpleQueryableIndex buildClusteredIndex( + ClusteredValueGroupsBaseTableSchema summary, + List>> clusterGroupColumns, + Map>> projectionColumns + ) + { + final Metadata reconstructed = summary.asMetadata(null); + return new SimpleQueryableIndex( + Intervals.of("2025-01-01/2025-01-02"), + new ListIndexed<>(List.of()), + new RoaringBitmapFactory(), + Map.of(), // clustered summary has no top-level columns + null, // no SegmentFileMapper for in-memory test + reconstructed, + projectionColumns, + summary, + clusterGroupColumns + ) + { + @Override + public Metadata getMetadata() + { + return reconstructed; + } + + @Override + public int getNumRows() + { + return 0; + } + }; + } + + @Test + void testConstructorAcceptsEmptyColumnsForClusteredSegment() + { + // Pre-Phase-2 the constructor would NPE here because __time isn't in the columns map. + Assertions.assertDoesNotThrow( + () -> buildClusteredIndex(summarySchema("acme"), List.of(Map.of()), Map.of()) + ); + } + + @Test + void testGetClusteredBaseSummaryReturnsSummary() + { + final ClusteredValueGroupsBaseTableSchema s = summarySchema("acme"); + SimpleQueryableIndex index = buildClusteredIndex(s, List.of(Map.of()), Map.of()); + Assertions.assertSame(s, index.getClusteredBaseSummary()); + } + + @Test + void testGetClusterGroupSchemasReturnsGroupsInOrder() + { + final ClusterGroupSchemaTestHelpers.Built built = summary("acme", "globex"); + final ClusteredValueGroupsBaseTableSchema sum = built.specs().get(0).getSummary(); + SimpleQueryableIndex index = buildClusteredIndex(sum, List.of(Map.of(), Map.of()), Map.of()); + List result = index.getClusterGroupSchemas(); + Assertions.assertEquals(2, result.size()); + Assertions.assertSame(built.specs().get(0), result.get(0)); + Assertions.assertSame(built.specs().get(1), result.get(1)); + } + + @Test + void testNonClusteredIndexHasNullSummaryAndEmptyGroups() + { + // Use the no-cluster-args constructor variant directly via the overload that takes no cluster params. + SimpleQueryableIndex index = new SimpleQueryableIndex( + Intervals.of("2025-01-01/2025-01-02"), + new ListIndexed<>(List.of()), + new RoaringBitmapFactory(), + Map.of(), + null, + null, + null + ) + { + @Override + public Metadata getMetadata() + { + return null; + } + + @Override + public int getNumRows() + { + return 0; + } + }; + + Assertions.assertNull(index.getClusteredBaseSummary()); + Assertions.assertTrue(index.getClusterGroupSchemas().isEmpty()); + } + + @Test + void testGetProjectionQueryableIndexResolvesAggregatesNotClusterGroups() + { + // Sanity: cluster groups don't pollute the aggregate-projection map. getProjection(spec) is the + // aggregate-only path; cluster-group dispatch is QueryableIndexCursorFactory's job. + SimpleQueryableIndex index = buildClusteredIndex( + summarySchema("acme"), + List.of(Map.of()), + Map.of() + ); + // No aggregate projections were added, so no aggregate by name "tenant=acme" exists either. + Assertions.assertThrows( + NullPointerException.class, + () -> index.getProjectionQueryableIndex("does-not-exist") + ); + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/projections/ClusterGroupSchemaTestHelpers.java b/processing/src/test/java/org/apache/druid/segment/projections/ClusterGroupSchemaTestHelpers.java new file mode 100644 index 000000000000..73e921f19181 --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/projections/ClusterGroupSchemaTestHelpers.java @@ -0,0 +1,187 @@ +/* + * 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.druid.segment.projections; + +import org.apache.druid.error.DruidException; +import org.apache.druid.segment.DimensionHandlerUtils; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.NullableTypeStrategy; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.column.ValueType; + +import javax.annotation.Nullable; +import java.util.ArrayList; +import java.util.Collections; +import java.util.EnumMap; +import java.util.List; +import java.util.Map; +import java.util.TreeSet; + +/** + * Test-only utility: build {@link ClusteredValueGroupsBaseTableSchema} inputs from typed clustering tuples. + * Derives the per-type dictionaries + per-spec dictionary IDs that the writer will emit at ingest time. + */ +public final class ClusterGroupSchemaTestHelpers +{ + private ClusterGroupSchemaTestHelpers() + { + } + + /** Bundle to pass into the {@link ClusteredValueGroupsBaseTableSchema} constructor's last two args. */ + public record Built(ClusteringDictionaries dictionaries, List specs) + { + } + + /** + * From typed clustering tuples (one per cluster group), build the per-type dictionaries and the matching specs. + */ + public static Built buildClusterGroups( + RowSignature clusteringColumns, + List> tuples + ) + { + final int numCols = clusteringColumns.size(); + + // 1. Coerce each tuple's raw values to the typed primitive expected per column. + final List coercedTuples = new ArrayList<>(tuples.size()); + for (List tuple : tuples) { + if (tuple == null) { + throw DruidException.defensive("tuple must not be null"); + } + if (tuple.size() != numCols) { + throw DruidException.defensive( + "tuple size [%s] must match clusteringColumns size [%s]", + tuple.size(), + numCols + ); + } + final Object[] coerced = new Object[numCols]; + for (int i = 0; i < numCols; i++) { + coerced[i] = coerceClusterGroupValue( + clusteringColumns.getColumnName(i), + clusteringColumns.getColumnType(i).orElseThrow(), + tuple.get(i) + ); + } + coercedTuples.add(coerced); + } + + // 2. Build per-type dictionaries; TreeSet on the type's strategy gives sort + dedupe in one shot. + final Map> perType = new EnumMap<>(ValueType.class); + for (int i = 0; i < numCols; i++) { + final ColumnType colType = clusteringColumns.getColumnType(i).orElseThrow(); + perType.computeIfAbsent(typeKey(colType), k -> newTypedTreeSet(colType)); + } + for (Object[] coerced : coercedTuples) { + for (int i = 0; i < numCols; i++) { + final ColumnType colType = clusteringColumns.getColumnType(i).orElseThrow(); + perType.get(typeKey(colType)).add(coerced[i]); + } + } + final List stringDict = materialize(perType.get(ValueType.STRING)); + final List longDict = materialize(perType.get(ValueType.LONG)); + final List doubleDict = materialize(perType.get(ValueType.DOUBLE)); + final List floatDict = materialize(perType.get(ValueType.FLOAT)); + final ClusteringDictionaries dictionaries = + new ClusteringDictionaries(stringDict, longDict, doubleDict, floatDict); + + // 3. For each tuple, look each value up in its column type's dictionary to derive the ID list. + final List specs = new ArrayList<>(coercedTuples.size()); + for (Object[] coerced : coercedTuples) { + final List ids = new ArrayList<>(numCols); + for (int i = 0; i < numCols; i++) { + final ColumnType colType = clusteringColumns.getColumnType(i).orElseThrow(); + @SuppressWarnings({"unchecked", "rawtypes"}) + final int id = Collections.binarySearch( + (List) dictionaries.dictionaryForType(colType), + coerced[i], + (NullableTypeStrategy) colType.getNullableStrategy() + ); + if (id < 0) { + throw DruidException.defensive( + "value [%s] not found in dictionary for clustering column [%s]; this is a bug in the helper", + coerced[i], + clusteringColumns.getColumnName(i) + ); + } + ids.add(id); + } + specs.add(new TableClusterGroupSpec(ids, null)); + } + + return new Built(dictionaries, Collections.unmodifiableList(specs)); + } + + private static ValueType typeKey(ColumnType columnType) + { + return columnType.getType(); + } + + @SuppressWarnings({"unchecked", "rawtypes"}) + private static TreeSet newTypedTreeSet(ColumnType columnType) + { + return new TreeSet<>((NullableTypeStrategy) columnType.getNullableStrategy()); + } + + @SuppressWarnings("unchecked") + private static List materialize(TreeSet set) + { + if (set == null || set.isEmpty()) { + return List.of(); + } + final List out = new ArrayList<>(set.size()); + for (Object v : set) { + out.add((T) v); + } + return Collections.unmodifiableList(out); + } + + /** + * Coerce a raw value (from JSON or test code) to the native Java type for {@code columnType}: STRING → String, + * LONG → Long, DOUBLE → Double, FLOAT → Float. Null passes through; unsupported column types are rejected. + */ + @Nullable + public static Object coerceClusterGroupValue( + String columnName, + @Nullable ColumnType columnType, + @Nullable Object raw + ) + { + if (!Projections.isAllowedClusteringType(columnType)) { + throw DruidException.defensive( + "clustering column [%s] has unsupported type [%s]; allowed types are STRING, LONG, DOUBLE, FLOAT", + columnName, + columnType + ); + } + if (raw == null) { + return null; + } + if (columnType.is(ValueType.STRING)) { + return DimensionHandlerUtils.convertObjectToString(raw); + } else if (columnType.is(ValueType.LONG)) { + return DimensionHandlerUtils.convertObjectToLong(raw); + } else if (columnType.is(ValueType.DOUBLE)) { + return DimensionHandlerUtils.convertObjectToDouble(raw); + } else { + return DimensionHandlerUtils.convertObjectToFloat(raw); + } + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/projections/ClusteredValueGroupsBaseTableSchemaTest.java b/processing/src/test/java/org/apache/druid/segment/projections/ClusteredValueGroupsBaseTableSchemaTest.java new file mode 100644 index 000000000000..67d664e3622c --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/projections/ClusteredValueGroupsBaseTableSchemaTest.java @@ -0,0 +1,391 @@ +/* + * 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.druid.segment.projections; + +import com.fasterxml.jackson.core.JsonProcessingException; +import com.fasterxml.jackson.databind.ObjectMapper; +import nl.jqno.equalsverifier.EqualsVerifier; +import org.apache.druid.error.DruidException; +import org.apache.druid.query.OrderBy; +import org.apache.druid.query.aggregation.AggregatorFactory; +import org.apache.druid.query.aggregation.CountAggregatorFactory; +import org.apache.druid.query.aggregation.LongSumAggregatorFactory; +import org.apache.druid.segment.TestHelper; +import org.apache.druid.segment.VirtualColumns; +import org.apache.druid.segment.column.ColumnHolder; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import java.util.ArrayList; +import java.util.List; + +class ClusteredValueGroupsBaseTableSchemaTest +{ + private static final ObjectMapper JSON_MAPPER = TestHelper.makeJsonMapper(); + private static final String TENANT_COL = "tenant"; + private static final String REGION_COL = "region"; + private static final String PRIORITY_COL = "priority"; + private static final String METRIC_COL = "metric"; + private static final RowSignature TENANT_SIGNATURE = RowSignature.builder() + .add(TENANT_COL, ColumnType.STRING) + .build(); + + @Test + void testSerdeMultiColumnClustering() throws JsonProcessingException + { + ClusteredValueGroupsBaseTableSchema schema = newSchema( + RowSignature.builder() + .add(TENANT_COL, ColumnType.STRING) + .add(PRIORITY_COL, ColumnType.LONG) + .build() + ); + + Assertions.assertEquals( + schema, + JSON_MAPPER.readValue(JSON_MAPPER.writeValueAsString(schema), ProjectionSchema.class) + ); + } + + @Test + void testColumnNamesIncludeClusteringColumnsAndAggregators() + { + // Summary carries the full logical signature, including clustering columns. + Assertions.assertEquals( + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL, PRIORITY_COL, REGION_COL, METRIC_COL, "count", "c"), + newSchema(TENANT_SIGNATURE).getColumnNames() + ); + } + + @Test + void testDimensionNamesIncludeClusteringColumnsExceptTime() + { + // Summary's dimension names exclude __time but keep clustering columns (they're logical dims). + Assertions.assertEquals( + List.of(TENANT_COL, PRIORITY_COL, REGION_COL, METRIC_COL), + newSchema(TENANT_SIGNATURE).getDimensionNames() + ); + } + + @Test + void testNullColumnsRejected() + { + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> new ClusteredValueGroupsBaseTableSchema( + null, + null, + null, + List.of(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + TENANT_SIGNATURE, + null, + null, + null + ) + ); + Assertions.assertEquals( + "clustered base table schema columns must not be null or empty", + t.getMessage() + ); + } + + @Test + void testNullOrderingRejected() + { + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> new ClusteredValueGroupsBaseTableSchema( + null, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL), + null, + null, + TENANT_SIGNATURE, + null, + null, + null + ) + ); + Assertions.assertEquals( + "clustered base table schema ordering must not be null", + t.getMessage() + ); + } + + @Test + void testEmptyClusteringColumnsRejected() + { + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> new ClusteredValueGroupsBaseTableSchema( + null, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL), + null, + List.of(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + RowSignature.empty(), + null, + null, + null + ) + ); + Assertions.assertEquals( + "clustered base table schema clusteringColumns must not be null or empty", + t.getMessage() + ); + } + + @Test + void testClusteringColumnMustAppearInColumns() + { + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> new ClusteredValueGroupsBaseTableSchema( + null, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL), + null, + List.of(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + RowSignature.builder().add("unknown", ColumnType.STRING).build(), + null, + null, + null + ) + ); + Assertions.assertEquals( + "clusteringColumn [unknown] must appear in columns of the clustered base table summary", + t.getMessage() + ); + } + + @Test + void testUnsupportedClusteringTypeRejected() + { + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> new ClusteredValueGroupsBaseTableSchema( + null, + List.of(ColumnHolder.TIME_COLUMN_NAME, "tenants"), + null, + List.of(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + RowSignature.builder().add("tenants", ColumnType.STRING_ARRAY).build(), + null, + null, + null + ) + ); + Assertions.assertEquals( + "clustering column [tenants] has unsupported type [ARRAY]; " + + "allowed types are STRING, LONG, DOUBLE, FLOAT", + t.getMessage() + ); + } + + @Test + void testSharedColumnsRoundTrip() throws JsonProcessingException + { + ClusteredValueGroupsBaseTableSchema schema = new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL, REGION_COL), + null, + List.of(OrderBy.ascending(TENANT_COL), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + TENANT_SIGNATURE, + List.of(REGION_COL), + ClusteringDictionaries.EMPTY, + null + ); + + ClusteredValueGroupsBaseTableSchema roundTripped = (ClusteredValueGroupsBaseTableSchema) JSON_MAPPER.readValue( + JSON_MAPPER.writeValueAsString(schema), + ProjectionSchema.class + ); + Assertions.assertEquals(schema, roundTripped); + Assertions.assertEquals(List.of(REGION_COL), roundTripped.getSharedColumns()); + } + + @Test + void testSharedColumnMustAppearInColumns() + { + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> new ClusteredValueGroupsBaseTableSchema( + null, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL), + null, + List.of(OrderBy.ascending(TENANT_COL), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + TENANT_SIGNATURE, + List.of("unknown"), + null, + null + ) + ); + Assertions.assertEquals( + "sharedColumn [unknown] must appear in columns of the clustered base table summary", + t.getMessage() + ); + } + + @Test + void testClusteringPrefixOfOrderingRequired() + { + // The first ordering position is __time, but the clustering column is tenant, must reject. + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> new ClusteredValueGroupsBaseTableSchema( + null, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL), + null, + List.of(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME), OrderBy.ascending(TENANT_COL)), + TENANT_SIGNATURE, + null, + null, + null + ) + ); + Assertions.assertTrue(t.getMessage().contains("clustering columns must form a prefix of the segment ordering")); + } + + @Test + void testClusteringRequiresOrderingAtLeastAsLongAsClusteringSize() + { + // Two clustering columns but ordering only has one entry, must reject. + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> new ClusteredValueGroupsBaseTableSchema( + null, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL, REGION_COL), + null, + List.of(OrderBy.ascending(TENANT_COL)), + RowSignature.builder() + .add(TENANT_COL, ColumnType.STRING) + .add(REGION_COL, ColumnType.STRING) + .build(), + null, + null, + null + ) + ); + Assertions.assertTrue( + t.getMessage().contains("ordering size [1] must be at least clusteringColumns size [2]") + ); + } + + @Test + void testNullDictionariesDefaultsToEmpty() + { + // A null clusteringDictionaries is interpreted as "no values yet" (empty per-type dicts); valid for a + // freshly-constructed clustered schema that doesn't have any groups defined. + final ClusteredValueGroupsBaseTableSchema schema = new ClusteredValueGroupsBaseTableSchema( + null, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL), + null, + List.of(OrderBy.ascending(TENANT_COL), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + TENANT_SIGNATURE, + null, + null, + null + ); + Assertions.assertEquals(ClusteringDictionaries.EMPTY, schema.getClusteringDictionaries()); + } + + @Test + void testJsonCreatorAcceptsUnsortedDictionary() + { + // The trusting JsonCreator path does NOT validate; segments produced by a well-behaved writer have already + // passed validation, so re-checking on every load would just burn CPU. + final ClusteringDictionaries dicts = new ClusteringDictionaries(List.of("b", "a"), null, null, null); + Assertions.assertEquals(List.of("b", "a"), dicts.getStringDictionary()); + } + + @Test + void testGetGroupOrderingDropsClusteringPrefix() + { + // Single clustering column → group ordering is the rest of the segment ordering. + Assertions.assertEquals( + List.of(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME), OrderBy.ascending(REGION_COL)), + newSchema(TENANT_SIGNATURE).getGroupOrdering() + ); + + // Two clustering columns → both prefix entries are dropped. + final RowSignature multi = RowSignature.builder() + .add(TENANT_COL, ColumnType.STRING) + .add(PRIORITY_COL, ColumnType.LONG) + .build(); + Assertions.assertEquals( + List.of(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME), OrderBy.ascending(REGION_COL)), + newSchema(multi).getGroupOrdering() + ); + } + + @Test + void testGetGroupColumnNamesExcludesClusteringColumns() + { + // Summary's columns are (__time, tenant, priority, region, metric); clustering=tenant → group view drops tenant + // (and adds aggregator names: count, c). + Assertions.assertEquals( + List.of(ColumnHolder.TIME_COLUMN_NAME, PRIORITY_COL, REGION_COL, METRIC_COL, "count", "c"), + newSchema(TENANT_SIGNATURE).getGroupColumnNames() + ); + } + + @Test + void testGetGroupDimensionNamesExcludesClusteringColumnsAndTime() + { + // Summary's dimension names are (__time excluded) (tenant, priority, region, metric); + // clustering=tenant → drops tenant. + Assertions.assertEquals( + List.of(PRIORITY_COL, REGION_COL, METRIC_COL), + newSchema(TENANT_SIGNATURE).getGroupDimensionNames() + ); + } + + @Test + void testEqualsAndHashcode() + { + EqualsVerifier.forClass(ClusteredValueGroupsBaseTableSchema.class) + .withIgnoredFields("timeColumnPosition", "effectiveGranularity") + .usingGetClass() + .verify(); + } + + private static ClusteredValueGroupsBaseTableSchema newSchema(RowSignature clusteringColumns) + { + // clustering columns must be a prefix of the segment ordering. Build the ordering by prefixing every clustering + // column in order, then __time and the remaining data columns. + final ArrayList ordering = new ArrayList<>(clusteringColumns.size() + 2); + for (String col : clusteringColumns.getColumnNames()) { + ordering.add(OrderBy.ascending(col)); + } + ordering.add(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)); + ordering.add(OrderBy.ascending(REGION_COL)); + // No cluster groups in this fixture; dictionaries are empty (null is also acceptable, defaults to EMPTY). + return new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of(ColumnHolder.TIME_COLUMN_NAME, TENANT_COL, PRIORITY_COL, REGION_COL, METRIC_COL), + new AggregatorFactory[]{ + new CountAggregatorFactory("count"), + new LongSumAggregatorFactory("c", "c") + }, + ordering, + clusteringColumns, + null, + ClusteringDictionaries.EMPTY, + null + ); + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/projections/ClusteringColumnSelectorFactoryTest.java b/processing/src/test/java/org/apache/druid/segment/projections/ClusteringColumnSelectorFactoryTest.java new file mode 100644 index 000000000000..03f0b9fc3953 --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/projections/ClusteringColumnSelectorFactoryTest.java @@ -0,0 +1,385 @@ +/* + * 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.druid.segment.projections; + +import org.apache.druid.query.dimension.DefaultDimensionSpec; +import org.apache.druid.segment.ColumnSelectorFactory; +import org.apache.druid.segment.ColumnValueSelector; +import org.apache.druid.segment.DimensionSelector; +import org.apache.druid.segment.NilColumnValueSelector; +import org.apache.druid.segment.RowIdSupplier; +import org.apache.druid.segment.column.ColumnCapabilities; +import org.apache.druid.segment.column.ColumnCapabilitiesImpl; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.column.ValueType; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import javax.annotation.Nullable; + +class ClusteringColumnSelectorFactoryTest +{ + private static final RowSignature SIGNATURE = RowSignature.builder().add("tenant", ColumnType.STRING).build(); + + @Test + void testStringClusteringColumnDimensionSelector() + { + RecordingDelegate delegate = new RecordingDelegate(); + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + delegate, + SIGNATURE, + new Object[]{"acme"} + ); + + DimensionSelector sel = f.makeDimensionSelector(DefaultDimensionSpec.of("tenant")); + Assertions.assertEquals("acme", sel.lookupName(0)); + Assertions.assertNull(delegate.lastDimSelectorName, "delegate must not be hit for clustering columns"); + } + + @Test + void testStringClusteringColumnValueSelector() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + SIGNATURE, + new Object[]{"acme"} + ); + + ColumnValueSelector sel = f.makeColumnValueSelector("tenant"); + Assertions.assertEquals("acme", sel.getObject()); + // Non-numeric string: ExprEval reports it as numerically null (matches a real string column read via getLong). + Assertions.assertTrue(sel.isNull()); + } + + @Test + void testLongClusteringColumnValueSelectorTypedMethods() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + RowSignature.builder().add("priority", ColumnType.LONG).build(), + new Object[]{42L} + ); + + ColumnValueSelector sel = f.makeColumnValueSelector("priority"); + Assertions.assertEquals(42L, sel.getLong()); + Assertions.assertEquals(42.0, sel.getDouble()); + Assertions.assertEquals(42.0f, sel.getFloat()); + Assertions.assertEquals(42L, sel.getObject()); + Assertions.assertFalse(sel.isNull()); + } + + @Test + void testDoubleClusteringColumnValueSelector() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + RowSignature.builder().add("price", ColumnType.DOUBLE).build(), + new Object[]{3.14} + ); + + ColumnValueSelector sel = f.makeColumnValueSelector("price"); + Assertions.assertEquals(3.14, sel.getDouble()); + Assertions.assertEquals(3.14, sel.getObject()); + } + + @Test + void testFloatClusteringColumnValueSelector() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + RowSignature.builder().add("ratio", ColumnType.FLOAT).build(), + new Object[]{0.5f} + ); + + ColumnValueSelector sel = f.makeColumnValueSelector("ratio"); + Assertions.assertEquals(0.5f, sel.getFloat()); + // ExpressionType collapses FLOAT to DOUBLE, so the unwrapped value is a Double. + Assertions.assertEquals(0.5d, sel.getObject()); + } + + @Test + void testNullClusteringValueDimensionSelectorIsNil() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + SIGNATURE, + new Object[]{null} + ); + + DimensionSelector sel = f.makeDimensionSelector(DefaultDimensionSpec.of("tenant")); + Assertions.assertNull(sel.lookupName(sel.getRow().get(0))); + } + + @Test + void testNullClusteringValueValueSelectorIsNull() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + RowSignature.builder().add("priority", ColumnType.LONG).build(), + new Object[]{null} + ); + + ColumnValueSelector sel = f.makeColumnValueSelector("priority"); + Assertions.assertTrue(sel.isNull()); + Assertions.assertNull(sel.getObject()); + } + + @Test + void testNonClusteringDimensionSelectorDelegated() + { + RecordingDelegate delegate = new RecordingDelegate(); + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + delegate, + SIGNATURE, + new Object[]{"acme"} + ); + + // Non-clustering selectors are wrapped in a lazy delegating wrapper; the delegate is only consulted on first + // use (this enables ConcatenatingCursor to swap delegates between cluster groups without recreating selectors). + DimensionSelector sel = f.makeDimensionSelector(DefaultDimensionSpec.of("region")); + Assertions.assertNull(delegate.lastDimSelectorName, "delegate must not be hit until selector is used"); + + Assertions.assertEquals("delegated:region", sel.lookupName(0)); + Assertions.assertEquals("region", delegate.lastDimSelectorName); + } + + @Test + void testNonClusteringValueSelectorDelegated() + { + RecordingDelegate delegate = new RecordingDelegate(); + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + delegate, + SIGNATURE, + new Object[]{"acme"} + ); + + ColumnValueSelector sel = f.makeColumnValueSelector("region"); + Assertions.assertNull(delegate.lastValueSelectorName, "delegate must not be hit until selector is used"); + + sel.getObject(); + Assertions.assertEquals("region", delegate.lastValueSelectorName); + } + + @Test + void testNonClusteringSelectorObservesNewDelegateAfterSetDelegate() + { + RecordingDelegate first = new RecordingDelegate(); + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + first, + SIGNATURE, + new Object[]{"acme"} + ); + + DimensionSelector sel = f.makeDimensionSelector(DefaultDimensionSpec.of("region")); + Assertions.assertEquals("delegated:region", sel.lookupName(0)); // routed to first delegate + + RecordingDelegate second = new RecordingDelegate(); + f.setDelegate(second, new Object[]{"globex"}); + + // Same selector instance, now backed by the new delegate's makeDimensionSelector result. + Assertions.assertEquals("delegated:region", sel.lookupName(0)); + Assertions.assertEquals("region", second.lastDimSelectorName); + } + + @Test + void testGetColumnCapabilitiesForClusteringColumns() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + RowSignature.builder() + .add("tenant", ColumnType.STRING) + .add("priority", ColumnType.LONG) + .build(), + new Object[]{"acme", 5L} + ); + + ColumnCapabilities tenantCaps = f.getColumnCapabilities("tenant"); + Assertions.assertEquals(ColumnType.STRING, ColumnType.fromCapabilities(tenantCaps)); + + ColumnCapabilities priorityCaps = f.getColumnCapabilities("priority"); + Assertions.assertEquals(ColumnType.LONG, ColumnType.fromCapabilities(priorityCaps)); + } + + @Test + void testGetColumnCapabilitiesForNonClusteringDelegated() + { + RecordingDelegate delegate = new RecordingDelegate(); + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + delegate, + SIGNATURE, + new Object[]{"acme"} + ); + + f.getColumnCapabilities("metric"); + Assertions.assertEquals("metric", delegate.lastCapabilitiesColumn); + } + + @Test + void testSetDelegateUpdatesClusteringValues() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + SIGNATURE, + new Object[]{"acme"} + ); + + // Acquire a selector tied to the first group's value. + ColumnValueSelector sel = f.makeColumnValueSelector("tenant"); + Assertions.assertEquals("acme", sel.getObject()); + + // Simulate group transition. + f.setDelegate(new RecordingDelegate(), new Object[]{"globex"}); + + // The same selector should observe the new value via the generation-counter cache invalidation. + Assertions.assertEquals("globex", sel.getObject()); + } + + @Test + void testSetDelegateUpdatesDimensionSelector() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + SIGNATURE, + new Object[]{"acme"} + ); + + DimensionSelector sel = f.makeDimensionSelector(DefaultDimensionSpec.of("tenant")); + Assertions.assertEquals("acme", sel.lookupName(sel.getRow().get(0))); + + f.setDelegate(new RecordingDelegate(), new Object[]{"globex"}); + + Assertions.assertEquals("globex", sel.lookupName(sel.getRow().get(0))); + } + + @Test + void testClusteringMatcherObservesGroupTransition() + { + // The clustering-column path: a matcher built before the group transition still gives the right verdict + // afterwards because it re-resolves through the generation-aware wrapper. + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + SIGNATURE, + new Object[]{"acme"} + ); + DimensionSelector sel = f.makeDimensionSelector(DefaultDimensionSpec.of("tenant")); + org.apache.druid.query.filter.ValueMatcher acmeMatcher = sel.makeValueMatcher("acme"); + + // Pre-transition: the constant clustering value is "acme"; matcher returns true. + Assertions.assertTrue(acmeMatcher.matches(false)); + + f.setDelegate(new RecordingDelegate(), new Object[]{"globex"}); + + // Post-transition: the constant is now "globex"; the same matcher (held across the transition) returns false. + Assertions.assertFalse(acmeMatcher.matches(false)); + } + + @Test + void testNonClusteringDelegatingMatcherObservesGroupTransition() + { + // The non-clustering "delegating" path: a matcher built before the transition still routes through the + // current delegate's matcher afterwards. We use two RecordingDelegates whose makeDimensionSelector returns + // selectors with predictable lookupName, then drive the matcher across a delegate swap. + final RecordingDelegate first = new RecordingDelegate(); + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + first, + SIGNATURE, + new Object[]{"acme"} + ); + DimensionSelector sel = f.makeDimensionSelector(DefaultDimensionSpec.of("region")); + // Force the inner selector to materialize so RecordingDelegate.lastDimSelectorName is "region". + sel.lookupName(0); + org.apache.druid.query.filter.ValueMatcher matcher = sel.makeValueMatcher("delegated:region"); + + // Pre-transition: RecordingDelegate's stub selector returns "delegated:region" for any id, so the matcher + // is true. + Assertions.assertTrue(matcher.matches(false)); + + final RecordingDelegate second = new RecordingDelegate(); + f.setDelegate(second, new Object[]{"acme"}); + + // Post-transition: still true (RecordingDelegate's stub is identical), but the important thing is that + // calling matches() now invokes the second delegate's selector (verified by lastDimSelectorName getting set). + Assertions.assertTrue(matcher.matches(false)); + Assertions.assertEquals("region", second.lastDimSelectorName); + } + + @Test + void testStringDimensionCapabilitiesFlavorIsSingleValue() + { + ClusteringColumnSelectorFactory f = new ClusteringColumnSelectorFactory( + new RecordingDelegate(), + SIGNATURE, + new Object[]{"acme"} + ); + ColumnCapabilities caps = f.getColumnCapabilities("tenant"); + // Sanity: the simple-string-capabilities helper returns a STRING type with no multi-value flag set. + Assertions.assertNotNull(caps); + Assertions.assertTrue(caps.is(ValueType.STRING)); + Assertions.assertEquals( + ColumnCapabilities.Capable.FALSE, + caps.hasMultipleValues() + ); + Assertions.assertEquals( + ColumnCapabilitiesImpl.createSimpleSingleValueStringColumnCapabilities().getType(), + caps.getType() + ); + } + + /** + * Stub delegate that returns nothing; used to verify that calls for clustering columns are intercepted before + * ever reaching it. For non-clustering columns, methods record the requested name and return Nil/null shapes. + */ + private static class RecordingDelegate implements ColumnSelectorFactory + { + String lastDimSelectorName; + String lastValueSelectorName; + String lastCapabilitiesColumn; + + @Override + public DimensionSelector makeDimensionSelector(org.apache.druid.query.dimension.DimensionSpec dimensionSpec) + { + lastDimSelectorName = dimensionSpec.getDimension(); + return DimensionSelector.constant("delegated:" + dimensionSpec.getDimension()); + } + + @Override + public ColumnValueSelector makeColumnValueSelector(String columnName) + { + lastValueSelectorName = columnName; + return NilColumnValueSelector.instance(); + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + lastCapabilitiesColumn = column; + return null; + } + + @Nullable + @Override + public RowIdSupplier getRowIdSupplier() + { + return null; + } + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/projections/ClusteringVectorColumnSelectorFactoryTest.java b/processing/src/test/java/org/apache/druid/segment/projections/ClusteringVectorColumnSelectorFactoryTest.java new file mode 100644 index 000000000000..9a76566b3e1b --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/projections/ClusteringVectorColumnSelectorFactoryTest.java @@ -0,0 +1,333 @@ +/* + * 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.druid.segment.projections; + +import org.apache.druid.error.DruidException; +import org.apache.druid.query.dimension.DefaultDimensionSpec; +import org.apache.druid.segment.column.ColumnCapabilities; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.column.ValueType; +import org.apache.druid.segment.vector.MultiValueDimensionVectorSelector; +import org.apache.druid.segment.vector.NoFilterVectorOffset; +import org.apache.druid.segment.vector.ReadableVectorInspector; +import org.apache.druid.segment.vector.SingleValueDimensionVectorSelector; +import org.apache.druid.segment.vector.VectorColumnSelectorFactory; +import org.apache.druid.segment.vector.VectorObjectSelector; +import org.apache.druid.segment.vector.VectorValueSelector; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import javax.annotation.Nullable; + +class ClusteringVectorColumnSelectorFactoryTest +{ + private static final RowSignature CLUSTER_SIGNATURE = RowSignature.builder().add("tenant", ColumnType.STRING).build(); + + @Test + void testStringClusteringSingleValueDimensionSelector() + { + StubDelegate delegate = new StubDelegate(inspectorFor(8)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + delegate, + CLUSTER_SIGNATURE, + new Object[]{"acme"} + ); + + SingleValueDimensionVectorSelector sel = f.makeSingleValueDimensionSelector(DefaultDimensionSpec.of("tenant")); + Assertions.assertNull(delegate.lastSingleValDimRequest, "delegate must not be hit for clustering column"); + Assertions.assertEquals("acme", sel.lookupName(0)); + Assertions.assertEquals(1, sel.getValueCardinality()); + } + + @Test + void testLongClusteringValueSelector() + { + StubDelegate delegate = new StubDelegate(inspectorFor(4)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + delegate, + RowSignature.builder().add("priority", ColumnType.LONG).build(), + new Object[]{42L} + ); + + VectorValueSelector sel = f.makeValueSelector("priority"); + Assertions.assertNull(delegate.lastValueRequest); + long[] vec = sel.getLongVector(); + for (long v : vec) { + Assertions.assertEquals(42L, v); + } + } + + @Test + void testDoubleClusteringObjectSelector() + { + StubDelegate delegate = new StubDelegate(inspectorFor(4)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + delegate, + RowSignature.builder().add("price", ColumnType.DOUBLE).build(), + new Object[]{3.14} + ); + + VectorObjectSelector sel = f.makeObjectSelector("price"); + Object[] vec = sel.getObjectVector(); + for (Object v : vec) { + Assertions.assertEquals(3.14, v); + } + } + + @Test + void testFloatClusteringValueSelector() + { + StubDelegate delegate = new StubDelegate(inspectorFor(4)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + delegate, + RowSignature.builder().add("ratio", ColumnType.FLOAT).build(), + new Object[]{0.5f} + ); + + VectorValueSelector sel = f.makeValueSelector("ratio"); + float[] vec = sel.getFloatVector(); + for (float v : vec) { + Assertions.assertEquals(0.5f, v); + } + } + + @Test + void testNullStringClusteringValueDimensionSelectorIsNil() + { + StubDelegate delegate = new StubDelegate(inspectorFor(4)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + delegate, + CLUSTER_SIGNATURE, + new Object[]{null} + ); + + SingleValueDimensionVectorSelector sel = f.makeSingleValueDimensionSelector(DefaultDimensionSpec.of("tenant")); + // Nil vector selector returns null on lookupName(0) regardless of id. + Assertions.assertNull(sel.lookupName(0)); + } + + @Test + void testNonClusteringColumnDelegated() + { + StubDelegate delegate = new StubDelegate(inspectorFor(4)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + delegate, + CLUSTER_SIGNATURE, + new Object[]{"acme"} + ); + + // Non-clustering selectors are wrapped in lazy delegating wrappers; the delegate is only consulted on first + // use, so a multi-group ConcatenatingVectorCursor can swap delegates between groups without recreating the + // selector instance. + SingleValueDimensionVectorSelector svdSel = + f.makeSingleValueDimensionSelector(DefaultDimensionSpec.of("region")); + Assertions.assertNull(delegate.lastSingleValDimRequest, "delegate must not be hit until selector is used"); + try { + svdSel.getRowVector(); + } + catch (NullPointerException expected) { + // StubDelegate returns null for the inner selector; the wrapper forwards to it. We just want to confirm + // the delegate's makeSingleValueDimensionSelector was invoked. + } + Assertions.assertEquals("region", delegate.lastSingleValDimRequest); + + VectorValueSelector vvSel = f.makeValueSelector("metric"); + Assertions.assertNull(delegate.lastValueRequest); + try { + vvSel.getLongVector(); + } + catch (NullPointerException expected) { + // same ^, confirming the delegate was reached + } + Assertions.assertEquals("metric", delegate.lastValueRequest); + + VectorObjectSelector voSel = f.makeObjectSelector("region"); + try { + voSel.getObjectVector(); + } + catch (NullPointerException expected) { + // same + } + Assertions.assertEquals("region", delegate.lastObjectRequest); + + // getColumnCapabilities is NOT lazy; it returns the result directly. + f.getColumnCapabilities("metric"); + Assertions.assertEquals("metric", delegate.lastCapsRequest); + } + + @Test + void testSetDelegateUpdatesClusteringValueOnExistingSelector() + { + StubDelegate firstDelegate = new StubDelegate(inspectorFor(4)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + firstDelegate, + RowSignature.builder().add("priority", ColumnType.LONG).build(), + new Object[]{5L} + ); + + VectorValueSelector sel = f.makeValueSelector("priority"); + long[] firstVec = sel.getLongVector(); + for (long v : firstVec) { + Assertions.assertEquals(5L, v); + } + + // Simulate group transition. + StubDelegate secondDelegate = new StubDelegate(inspectorFor(4)); + f.setDelegate(secondDelegate, new Object[]{42L}); + + long[] secondVec = sel.getLongVector(); + for (long v : secondVec) { + Assertions.assertEquals(42L, v); + } + } + + @Test + void testSetDelegateUpdatesNonClusteringSelector() + { + StubDelegate first = new StubDelegate(inspectorFor(4)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + first, + CLUSTER_SIGNATURE, + new Object[]{"acme"} + ); + + SingleValueDimensionVectorSelector sel = + f.makeSingleValueDimensionSelector(DefaultDimensionSpec.of("region")); + try { + sel.getRowVector(); // warms the cache against the first delegate + } + catch (NullPointerException expected) { + // expected; just confirming the route + } + Assertions.assertEquals("region", first.lastSingleValDimRequest); + + StubDelegate second = new StubDelegate(inspectorFor(4)); + f.setDelegate(second, new Object[]{"globex"}); + + try { + sel.getRowVector(); // generation bumped → re-fetches against second delegate + } + catch (NullPointerException expected) { + // expected + } + Assertions.assertEquals("region", second.lastSingleValDimRequest); + } + + @Test + void testGetColumnCapabilitiesForClusteringColumns() + { + StubDelegate delegate = new StubDelegate(inspectorFor(4)); + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + delegate, + RowSignature.builder() + .add("tenant", ColumnType.STRING) + .add("priority", ColumnType.LONG) + .build(), + new Object[]{"acme", 5L} + ); + + ColumnCapabilities tenantCaps = f.getColumnCapabilities("tenant"); + Assertions.assertTrue(tenantCaps.is(ValueType.STRING)); + + ColumnCapabilities priorityCaps = f.getColumnCapabilities("priority"); + Assertions.assertTrue(priorityCaps.is(ValueType.LONG)); + + Assertions.assertNull(delegate.lastCapsRequest, "delegate must not be hit for clustering capabilities"); + } + + @Test + void testMultiValueDimensionSelectorOnClusteringRejected() + { + ClusteringVectorColumnSelectorFactory f = new ClusteringVectorColumnSelectorFactory( + new StubDelegate(inspectorFor(4)), + CLUSTER_SIGNATURE, + new Object[]{"acme"} + ); + Assertions.assertThrows( + DruidException.class, + () -> f.makeMultiValueDimensionSelector(DefaultDimensionSpec.of("tenant")) + ); + } + + private static ReadableVectorInspector inspectorFor(int size) + { + return new NoFilterVectorOffset(size, 0, size); + } + + private static class StubDelegate implements VectorColumnSelectorFactory + { + final ReadableVectorInspector inspector; + String lastSingleValDimRequest; + String lastValueRequest; + String lastObjectRequest; + String lastCapsRequest; + + StubDelegate(ReadableVectorInspector inspector) + { + this.inspector = inspector; + } + + @Override + public ReadableVectorInspector getReadableVectorInspector() + { + return inspector; + } + + @Override + public SingleValueDimensionVectorSelector makeSingleValueDimensionSelector( + org.apache.druid.query.dimension.DimensionSpec dimensionSpec + ) + { + lastSingleValDimRequest = dimensionSpec.getDimension(); + return null; + } + + @Override + public MultiValueDimensionVectorSelector makeMultiValueDimensionSelector( + org.apache.druid.query.dimension.DimensionSpec dimensionSpec + ) + { + throw new UnsupportedOperationException("not used"); + } + + @Override + public VectorValueSelector makeValueSelector(String column) + { + lastValueRequest = column; + return null; + } + + @Override + public VectorObjectSelector makeObjectSelector(String column) + { + lastObjectRequest = column; + return null; + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + lastCapsRequest = column; + return null; + } + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/projections/ProjectionsPruneClusterGroupsTest.java b/processing/src/test/java/org/apache/druid/segment/projections/ProjectionsPruneClusterGroupsTest.java new file mode 100644 index 000000000000..624f00a0e08d --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/projections/ProjectionsPruneClusterGroupsTest.java @@ -0,0 +1,721 @@ +/* + * 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.druid.segment.projections; + +import com.google.common.collect.ImmutableList; +import org.apache.druid.query.OrderBy; +import org.apache.druid.query.aggregation.AggregatorFactory; +import org.apache.druid.query.aggregation.CountAggregatorFactory; +import org.apache.druid.query.expression.TestExprMacroTable; +import org.apache.druid.query.filter.EqualityFilter; +import org.apache.druid.query.filter.Filter; +import org.apache.druid.query.filter.LikeDimFilter; +import org.apache.druid.query.filter.NullFilter; +import org.apache.druid.query.filter.TypedInFilter; +import org.apache.druid.segment.VirtualColumns; +import org.apache.druid.segment.column.ColumnHolder; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.filter.AndFilter; +import org.apache.druid.segment.filter.NotFilter; +import org.apache.druid.segment.filter.OrFilter; +import org.apache.druid.segment.virtual.ExpressionVirtualColumn; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import javax.annotation.Nullable; +import java.util.Arrays; +import java.util.LinkedHashSet; +import java.util.List; + +class ProjectionsPruneClusterGroupsTest +{ + /** Build a fresh single-spec summary for a STRING-clustered ``tenant`` group; each call gets its own summary. */ + private static TableClusterGroupSpec stringGroup(String tenant) + { + final RowSignature clustering = RowSignature.builder().add("tenant", ColumnType.STRING).build(); + final ClusterGroupSchemaTestHelpers.Built built = ClusterGroupSchemaTestHelpers.buildClusterGroups( + clustering, + List.of(Arrays.asList(tenant)) // Arrays.asList allows tenant == null + ); + new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of("tenant", ColumnHolder.TIME_COLUMN_NAME, "metric"), + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + List.of(OrderBy.ascending("tenant"), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + clustering, + null, + built.dictionaries(), + built.specs() + ); + return built.specs().get(0); + } + + private static TableClusterGroupSpec longGroup(long priority) + { + final RowSignature clustering = RowSignature.builder().add("priority", ColumnType.LONG).build(); + final ClusterGroupSchemaTestHelpers.Built built = ClusterGroupSchemaTestHelpers.buildClusterGroups( + clustering, + List.of(List.of(priority)) + ); + new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of("priority", ColumnHolder.TIME_COLUMN_NAME, "metric"), + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + List.of(OrderBy.ascending("priority"), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + clustering, + null, + built.dictionaries(), + built.specs() + ); + return built.specs().get(0); + } + + private static TableClusterGroupSpec multiGroup(String tenant, String region) + { + final RowSignature clustering = RowSignature.builder() + .add("tenant", ColumnType.STRING) + .add("region", ColumnType.STRING) + .build(); + final ClusterGroupSchemaTestHelpers.Built built = ClusterGroupSchemaTestHelpers.buildClusterGroups( + clustering, + List.of(Arrays.asList(tenant, region)) + ); + new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of("tenant", "region", ColumnHolder.TIME_COLUMN_NAME, "metric"), + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + List.of( + OrderBy.ascending("tenant"), + OrderBy.ascending("region"), + OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME) + ), + clustering, + null, + built.dictionaries(), + built.specs() + ); + return built.specs().get(0); + } + + /** IoT-style mixed-type clustering: {@code (device_id LONG, region STRING)}. */ + private static TableClusterGroupSpec deviceRegionGroup(long deviceId, String region) + { + final RowSignature clustering = RowSignature.builder() + .add("device_id", ColumnType.LONG) + .add("region", ColumnType.STRING) + .build(); + final ClusterGroupSchemaTestHelpers.Built built = ClusterGroupSchemaTestHelpers.buildClusterGroups( + clustering, + List.of(Arrays.asList(deviceId, region)) + ); + new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of("device_id", "region", ColumnHolder.TIME_COLUMN_NAME, "temperature"), + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + List.of( + OrderBy.ascending("device_id"), + OrderBy.ascending("region"), + OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME) + ), + clustering, + null, + built.dictionaries(), + built.specs() + ); + return built.specs().get(0); + } + + /** Generic single-STRING-column clustering with a domain-agnostic column name. */ + private static TableClusterGroupSpec partitionGroup(String key) + { + final RowSignature clustering = RowSignature.builder().add("partition", ColumnType.STRING).build(); + final ClusterGroupSchemaTestHelpers.Built built = ClusterGroupSchemaTestHelpers.buildClusterGroups( + clustering, + List.of(Arrays.asList(key)) + ); + new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of("partition", ColumnHolder.TIME_COLUMN_NAME, "payload"), + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + List.of(OrderBy.ascending("partition"), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + clustering, + null, + built.dictionaries(), + built.specs() + ); + return built.specs().get(0); + } + + private static List groups(TableClusterGroupSpec... gs) + { + return ImmutableList.copyOf(gs); + } + + /** Test convenience: prune without query-side virtual columns. */ + private static List pruneClusterGroups( + List groups, + @Nullable Filter filter + ) + { + return Projections.pruneClusterGroups(groups, filter, VirtualColumns.EMPTY); + } + + private static LinkedHashSet filters(Filter... fs) + { + LinkedHashSet out = new LinkedHashSet<>(); + for (Filter f : fs) { + out.add(f); + } + return out; + } + + @Test + void testNullFilterReturnsAllGroups() + { + List all = groups(stringGroup("acme"), stringGroup("globex")); + Assertions.assertSame(all, pruneClusterGroups(all, null)); + } + + @Test + void testEmptyGroupsReturnsAllGroups() + { + List empty = List.of(); + Assertions.assertSame(empty, pruneClusterGroups(empty, new EqualityFilter("tenant", ColumnType.STRING, "acme", null))); + } + + @Test + void testEqualityFilterOnStringClusteringColumnSelectsOneGroup() + { + List all = groups(stringGroup("acme"), stringGroup("globex")); + Filter f = new EqualityFilter("tenant", ColumnType.STRING, "acme", null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + } + + @Test + void testEqualityFilterOnLongClusteringColumn() + { + List all = groups(longGroup(5L), longGroup(10L), longGroup(20L)); + Filter f = new EqualityFilter("priority", ColumnType.LONG, 10L, null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(1), kept.get(0)); + } + + @Test + void testTypedInFilterOnLongClusteringColumn() + { + List all = groups(longGroup(5L), longGroup(10L), longGroup(20L)); + Filter f = new TypedInFilter("priority", ColumnType.LONG, List.of(5L, 20L), null, null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(2, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + Assertions.assertSame(all.get(2), kept.get(1)); + } + + @Test + void testFilterOnNonClusteringColumnKeepsAllGroups() + { + // Filter doesn't reference a clustering column → can't prune anything. + List all = groups(stringGroup("acme"), stringGroup("globex")); + Filter f = new EqualityFilter("metric", ColumnType.LONG, 42L, null); + Assertions.assertEquals(all, pruneClusterGroups(all, f)); + } + + @Test + void testAndFilterIntersectsSubFilterResults() + { + List all = groups( + multiGroup("acme", "us-east-1"), + multiGroup("acme", "us-west-2"), + multiGroup("globex", "us-east-1") + ); + Filter f = new AndFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "acme", null), + new EqualityFilter("region", ColumnType.STRING, "us-west-2", null) + )); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(1), kept.get(0)); + } + + @Test + void testOrFilterUnionsSubFilterResults() + { + List all = groups( + stringGroup("acme"), + stringGroup("globex"), + stringGroup("oscorp") + ); + Filter f = new OrFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "acme", null), + new EqualityFilter("tenant", ColumnType.STRING, "oscorp", null) + )); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(2, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + Assertions.assertSame(all.get(2), kept.get(1)); + } + + @Test + void testNotFilterInvertsMatch() + { + List all = groups(stringGroup("acme"), stringGroup("globex")); + Filter f = new NotFilter(new EqualityFilter("tenant", ColumnType.STRING, "acme", null)); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(1), kept.get(0)); + } + + /** Regression: a boolean (kept/pruned) matcher would flip UNKNOWN to "pruned" under NOT, dropping live data. */ + @Test + void testNotOverNonClusteringColumnKeepsAllGroups() + { + List all = groups(stringGroup("acme"), stringGroup("globex")); + Filter f = new NotFilter(new EqualityFilter("metric", ColumnType.LONG, 42L, null)); + Assertions.assertEquals(all, pruneClusterGroups(all, f)); + } + + @Test + void testDoubleNotPreservesPositive() + { + List all = groups(stringGroup("acme"), stringGroup("globex")); + Filter f = new NotFilter(new NotFilter(new EqualityFilter("tenant", ColumnType.STRING, "acme", null))); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + } + + @Test + void testAndOfMustNotMatchAndUnknownPrunes() + { + // tenant=globex is MUST_NOT_MATCH for tenant=acme group; AND with anything stays MUST_NOT_MATCH → prune. + List all = groups(stringGroup("acme")); + Filter f = new AndFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "globex", null), // MUST_NOT_MATCH for acme + new EqualityFilter("metric", ColumnType.LONG, 42L, null) // UNKNOWN + )); + Assertions.assertTrue(pruneClusterGroups(all, f).isEmpty()); + } + + @Test + void testOrOfMustMatchAndUnknownKeeps() + { + // tenant=acme is MUST_MATCH for tenant=acme group; OR short-circuits to MUST_MATCH regardless of the other. + List all = groups(stringGroup("acme")); + Filter f = new OrFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "acme", null), // MUST_MATCH + new EqualityFilter("metric", ColumnType.LONG, 42L, null) // UNKNOWN + )); + Assertions.assertEquals(all, pruneClusterGroups(all, f)); + } + + @Test + void testOrOfMustNotMatchAndUnknownKeeps() + { + // tenant=globex is MUST_NOT_MATCH for tenant=acme group; metric=42 is UNKNOWN. + // OR(MUST_NOT_MATCH, UNKNOWN) = UNKNOWN → keep (rows with metric=42 might exist). + List all = groups(stringGroup("acme")); + Filter f = new OrFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "globex", null), + new EqualityFilter("metric", ColumnType.LONG, 42L, null) + )); + Assertions.assertEquals(all, pruneClusterGroups(all, f)); + } + + @Test + void testNotOfAndWithMustNotMatchKeeps() + { + // inner: AND(tenant=globex on tenant=acme group, otherCol=foo) + // AND(MUST_NOT_MATCH, UNKNOWN) = MUST_NOT_MATCH (no row has tenant=globex, so AND is always false) + // NOT(MUST_NOT_MATCH) = MUST_MATCH → keep (NOT(false) is true for every row). + List all = groups(stringGroup("acme")); + Filter f = new NotFilter(new AndFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "globex", null), + new EqualityFilter("metric", ColumnType.LONG, 42L, null) + ))); + Assertions.assertEquals(all, pruneClusterGroups(all, f)); + } + + @Test + void testNotOfAndWithUnknownKeeps() + { + // inner: AND(tenant=acme, otherCol=foo) on tenant=acme group + // AND(MUST_MATCH, UNKNOWN) = UNKNOWN + // NOT(UNKNOWN) = UNKNOWN → keep. + List all = groups(stringGroup("acme")); + Filter f = new NotFilter(new AndFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "acme", null), + new EqualityFilter("metric", ColumnType.LONG, 42L, null) + ))); + Assertions.assertEquals(all, pruneClusterGroups(all, f)); + } + + @Test + void testMixedClusteringAndNonClusteringWithAndFilter() + { + // AND combines a clustering filter with a non-clustering filter; the latter contributes no info → result is + // determined purely by the clustering filter. + List all = groups(stringGroup("acme"), stringGroup("globex")); + Filter f = new AndFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "acme", null), + new EqualityFilter("metric", ColumnType.LONG, 42L, null) + )); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + } + + @Test + void testEmptyResultWhenNoGroupMatches() + { + List all = groups(stringGroup("acme"), stringGroup("globex")); + Filter f = new EqualityFilter("tenant", ColumnType.STRING, "unknown", null); + Assertions.assertTrue(pruneClusterGroups(all, f).isEmpty()); + } + + @Test + void testUnsupportedFilterTypeIsConservative() + { + // LikeDimFilter isn't recognized by the pruner → keep all groups (no pruning info). + List all = groups(stringGroup("acme"), stringGroup("globex")); + Filter f = new LikeDimFilter("tenant", "acm%", null, null).toFilter(); + Assertions.assertEquals(all, pruneClusterGroups(all, f)); + } + + /** Group whose clustering column is itself a virtual column (e.g. {@code lower(tenant)}). */ + private static TableClusterGroupSpec virtualClusteringGroup(String loweredTenant) + { + final RowSignature clustering = RowSignature.builder().add("tenant_lower", ColumnType.STRING).build(); + final ClusterGroupSchemaTestHelpers.Built built = ClusterGroupSchemaTestHelpers.buildClusterGroups( + clustering, + List.of(Arrays.asList(loweredTenant)) + ); + new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.create( + new ExpressionVirtualColumn( + "tenant_lower", + "lower(tenant)", + ColumnType.STRING, + TestExprMacroTable.INSTANCE + ) + ), + List.of("tenant_lower", ColumnHolder.TIME_COLUMN_NAME, "metric"), + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + List.of(OrderBy.ascending("tenant_lower"), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + clustering, + null, + built.dictionaries(), + built.specs() + ); + return built.specs().get(0); + } + + @Test + void testVirtualClusteringMatchesQueryVirtualColumnByEquivalence() + { + List all = groups( + virtualClusteringGroup("acme"), + virtualClusteringGroup("globex") + ); + // Query has its own virtual column with a different output name but equivalent expression. + final VirtualColumns queryVcs = VirtualColumns.create( + new ExpressionVirtualColumn( + "query_lower", + "lower(tenant)", + ColumnType.STRING, + TestExprMacroTable.INSTANCE + ) + ); + Filter f = new EqualityFilter("query_lower", ColumnType.STRING, "acme", null); + List kept = Projections.pruneClusterGroups(all, f, queryVcs); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + } + + @Test + void testVirtualClusteringRejectsNonEquivalentQueryVirtualColumn() + { + List all = groups(virtualClusteringGroup("acme")); + // Query VC is upper(tenant), not lower(tenant); not equivalent. + final VirtualColumns queryVcs = VirtualColumns.create( + new ExpressionVirtualColumn( + "query_upper", + "upper(tenant)", + ColumnType.STRING, + TestExprMacroTable.INSTANCE + ) + ); + Filter f = new EqualityFilter("query_upper", ColumnType.STRING, "ACME", null); + // No equivalence found → can't prune; keep group conservatively. + Assertions.assertEquals(all, Projections.pruneClusterGroups(all, f, queryVcs)); + } + + @Test + void testVirtualClusteringWithoutQueryVirtualColumnsIsConservative() + { + // If queryVirtualColumns is empty (or not provided), the pruner can't resolve the filter column to a + // clustering column via virtual-column equivalence → keep all groups. + List all = groups(virtualClusteringGroup("acme"), virtualClusteringGroup("globex")); + Filter f = new EqualityFilter("query_lower", ColumnType.STRING, "acme", null); + Assertions.assertEquals(all, Projections.pruneClusterGroups(all, f, VirtualColumns.EMPTY)); + // Same with the no-virtual-columns convenience overload. + Assertions.assertEquals(all, pruneClusterGroups(all, f)); + } + + @Test + void testDirectMatchWorksWithUnrelatedQueryVirtualColumns() + { + // Query VC by a name unrelated to any clustering column → query-VC check finds nothing for "tenant" → falls + // through to the direct-name path against the clustering signature. Unrelated VCs don't perturb the match. + List all = groups(stringGroup("acme"), stringGroup("globex")); + final VirtualColumns queryVcs = VirtualColumns.create( + new ExpressionVirtualColumn( + "ts_floor", + "timestamp_floor(__time, 'P1D', null, null)", + ColumnType.LONG, + TestExprMacroTable.INSTANCE + ) + ); + Filter f = new EqualityFilter("tenant", ColumnType.STRING, "acme", null); + List kept = Projections.pruneClusterGroups(all, f, queryVcs); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + } + + @Test + void testQueryVirtualColumnShadowingClusteringNameWithoutEquivalenceIsConservative() + { + // Query has a VC named "tenant" that is NOT equivalent to the clustering column "tenant" (different + // expression, totally different semantics). Once the query defines a VC by that name, the filter's "tenant" + // resolves to the VC's value, NOT the clustering column. The pruner must not mistake the VC reference for a + // clustering-column reference and prune on the clustering tuple — that would silently prune live rows whose + // VC value happens to equal "acme" even though the clustering tenant is "globex". Without the query-VC-first + // check this regressed and pruned the globex group. + List all = groups(stringGroup("acme"), stringGroup("globex")); + final VirtualColumns queryVcs = VirtualColumns.create( + new ExpressionVirtualColumn( + "tenant", // shadows the clustering column name + "lower(other)", // unrelated expression + ColumnType.STRING, + TestExprMacroTable.INSTANCE + ) + ); + Filter f = new EqualityFilter("tenant", ColumnType.STRING, "acme", null); + Assertions.assertEquals(all, Projections.pruneClusterGroups(all, f, queryVcs)); + } + + @Test + void testQueryVirtualColumnShadowingClusteringNameWithSameNameEquivalencePrunes() + { + // Group clusters on "tenant_lower" via a group VC of the same name (lower(tenant)). The query also defines + // a VC named "tenant_lower" with the identical expression — names collide AND the VCs are equivalent. The + // pruner should treat the filter as a clustering-column reference and prune normally (same-name target → + // no remap entry needed; the rewritten filter is identical to the original). + List all = groups( + virtualClusteringGroup("acme"), + virtualClusteringGroup("globex") + ); + final VirtualColumns queryVcs = VirtualColumns.create( + new ExpressionVirtualColumn( + "tenant_lower", // SAME name as the group's clustering column / group VC + "lower(tenant)", + ColumnType.STRING, + TestExprMacroTable.INSTANCE + ) + ); + Filter f = new EqualityFilter("tenant_lower", ColumnType.STRING, "acme", null); + List kept = Projections.pruneClusterGroups(all, f, queryVcs); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + } + + @Test + void testQueryVirtualColumnShadowsOneOfMultipleClusteringColumnsStillPrunesOthers() + { + // Two-column clustering (tenant, region). Query VC named "tenant" shadows that clustering column with a + // non-equivalent expression, but "region" is not shadowed. The AND filter on both columns should still let + // the region leaf prune groups where region doesn't match (Kleene: UNKNOWN AND FALSE = FALSE), while groups + // whose region matches stay (UNKNOWN AND TRUE = UNKNOWN → keep conservatively). + List all = groups( + multiGroup("acme", "us-east-1"), + multiGroup("acme", "us-west-2"), + multiGroup("globex", "us-east-1") + ); + final VirtualColumns queryVcs = VirtualColumns.create( + new ExpressionVirtualColumn( + "tenant", + "lower(other)", + ColumnType.STRING, + TestExprMacroTable.INSTANCE + ) + ); + Filter f = new AndFilter(filters( + new EqualityFilter("tenant", ColumnType.STRING, "acme", null), + new EqualityFilter("region", ColumnType.STRING, "us-east-1", null) + )); + List kept = Projections.pruneClusterGroups(all, f, queryVcs); + Assertions.assertEquals(2, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); // (acme, us-east-1) + Assertions.assertSame(all.get(2), kept.get(1)); // (globex, us-east-1) + } + + // --- Null clustering value handling --- + // + // Null clustering values are allowed; these tests pin the pruner's behavior across the supported filter types: + // - NullFilter: matches iff the column value is null. + // - EqualityFilter: does NOT match nulls by design — returns MUST_NOT_MATCH for a null clustering value. + // - TypedInFilter: matches nulls iff null is in the values list. + + @Test + void testNullFilterMatchesNullClusteringValue() + { + final TableClusterGroupSpec nullGroup = stringGroup(null); + final TableClusterGroupSpec acmeGroup = stringGroup("acme"); + List all = groups(nullGroup, acmeGroup); + Filter f = NullFilter.forColumn("tenant"); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(nullGroup, kept.get(0)); + } + + @Test + void testNullFilterPrunesNonNullClusteringValue() + { + List all = groups(stringGroup("acme")); + Filter f = NullFilter.forColumn("tenant"); + Assertions.assertTrue(pruneClusterGroups(all, f).isEmpty()); + } + + @Test + void testEqualityFilterPrunesNullClusteringValue() + { + // EqualityFilter does not match nulls (filter constructor rejects null match values; null group's rows never + // match a non-null literal). A group whose clustering value is null is pruned regardless of the literal. + List all = groups(stringGroup(null), stringGroup("acme")); + Filter f = new EqualityFilter("tenant", ColumnType.STRING, "acme", null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(all.get(1), kept.get(0)); + } + + @Test + void testTypedInFilterIncludingNullMatchesNullClusteringValue() + { + final TableClusterGroupSpec nullGroup = stringGroup(null); + final TableClusterGroupSpec acmeGroup = stringGroup("acme"); + final TableClusterGroupSpec globexGroup = stringGroup("globex"); + List all = groups(nullGroup, acmeGroup, globexGroup); + Filter f = new TypedInFilter("tenant", ColumnType.STRING, Arrays.asList(null, "globex"), null, null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(2, kept.size()); + Assertions.assertSame(nullGroup, kept.get(0)); + Assertions.assertSame(globexGroup, kept.get(1)); + } + + @Test + void testTypedInFilterWithoutNullPrunesNullClusteringValue() + { + final TableClusterGroupSpec nullGroup = stringGroup(null); + final TableClusterGroupSpec acmeGroup = stringGroup("acme"); + List all = groups(nullGroup, acmeGroup); + Filter f = new TypedInFilter("tenant", ColumnType.STRING, List.of("acme", "globex"), null, null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(acmeGroup, kept.get(0)); + } + + @Test + void testNotNullFilterOverClusteringColumnIsTriState() + { + // NOT IS NULL: must keep groups with non-null clustering values, prune groups with null clustering values. + final TableClusterGroupSpec nullGroup = stringGroup(null); + final TableClusterGroupSpec acmeGroup = stringGroup("acme"); + List all = groups(nullGroup, acmeGroup); + Filter f = new NotFilter(NullFilter.forColumn("tenant")); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(acmeGroup, kept.get(0)); + } + + // --- Non-tenant domains: IoT-style mixed-type and generic-partition fixtures. --- + + @Test + void testIoTEqualityOnLongClusteringColumn() + { + final TableClusterGroupSpec d101East = deviceRegionGroup(101L, "us-east-1"); + final TableClusterGroupSpec d202East = deviceRegionGroup(202L, "us-east-1"); + final TableClusterGroupSpec d101West = deviceRegionGroup(101L, "us-west-2"); + List all = groups(d101East, d202East, d101West); + Filter f = new EqualityFilter("device_id", ColumnType.LONG, 101L, null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(2, kept.size()); + Assertions.assertSame(d101East, kept.get(0)); + Assertions.assertSame(d101West, kept.get(1)); + } + + @Test + void testIoTAndAcrossMixedTypeClusteringColumns() + { + // AND across a LONG and a STRING clustering column; exercises per-type dictionary routing. + final TableClusterGroupSpec d101East = deviceRegionGroup(101L, "us-east-1"); + final TableClusterGroupSpec d202East = deviceRegionGroup(202L, "us-east-1"); + final TableClusterGroupSpec d101West = deviceRegionGroup(101L, "us-west-2"); + List all = groups(d101East, d202East, d101West); + Filter f = new AndFilter(filters( + new EqualityFilter("device_id", ColumnType.LONG, 101L, null), + new EqualityFilter("region", ColumnType.STRING, "us-east-1", null) + )); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(d101East, kept.get(0)); + } + + @Test + void testIoTTypedInFilterOnLongDeviceId() + { + // Typed IN on the LONG column; pruner must route via the column's type, not the STRING dict. + List all = groups( + deviceRegionGroup(101L, "us-east-1"), + deviceRegionGroup(202L, "us-east-1"), + deviceRegionGroup(303L, "us-east-1") + ); + Filter f = new TypedInFilter("device_id", ColumnType.LONG, List.of(101L, 303L), null, null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(2, kept.size()); + Assertions.assertSame(all.get(0), kept.get(0)); + Assertions.assertSame(all.get(2), kept.get(1)); + } + + @Test + void testGenericPartitionClusteringPrunesOnEqualityFilter() + { + final TableClusterGroupSpec pA = partitionGroup("A"); + final TableClusterGroupSpec pB = partitionGroup("B"); + final TableClusterGroupSpec pC = partitionGroup("C"); + List all = groups(pA, pB, pC); + Filter f = new EqualityFilter("partition", ColumnType.STRING, "B", null); + List kept = pruneClusterGroups(all, f); + Assertions.assertEquals(1, kept.size()); + Assertions.assertSame(pB, kept.get(0)); + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/projections/ProjectionsTest.java b/processing/src/test/java/org/apache/druid/segment/projections/ProjectionsTest.java index 46fb8912c078..01b391001b77 100644 --- a/processing/src/test/java/org/apache/druid/segment/projections/ProjectionsTest.java +++ b/processing/src/test/java/org/apache/druid/segment/projections/ProjectionsTest.java @@ -547,6 +547,44 @@ public void testSchemaMatchIntervalProjectionGranularity() Assertions.assertEquals(expected, projectionMatch); } + private static RowSignature sig(String name, ColumnType type) + { + return RowSignature.builder().add(name, type).build(); + } + + @Test + void testGetClusterGroupSegmentInternalFileName() + { + // Smoosh layout: __base$_.../. The IDs encode the group's clustering identity via the + // summary's per-column dictionaries. + Assertions.assertEquals( + "__base$0/tenant", + Projections.getClusterGroupSegmentInternalFileName(List.of(0), "tenant") + ); + Assertions.assertEquals( + "__base$5/__time", + Projections.getClusterGroupSegmentInternalFileName(List.of(5), "__time") + ); + Assertions.assertEquals( + "__base$0_1_3/__time", + Projections.getClusterGroupSegmentInternalFileName(List.of(0, 1, 3), "__time") + ); + Assertions.assertEquals("__base$42/", Projections.getClusterGroupSegmentInternalFilePrefix(List.of(42))); + Assertions.assertEquals("__base$1_2/", Projections.getClusterGroupSegmentInternalFilePrefix(List.of(1, 2))); + } + + @Test + void testIsAllowedClusteringType() + { + Assertions.assertTrue(Projections.isAllowedClusteringType(ColumnType.STRING)); + Assertions.assertTrue(Projections.isAllowedClusteringType(ColumnType.LONG)); + Assertions.assertTrue(Projections.isAllowedClusteringType(ColumnType.DOUBLE)); + Assertions.assertTrue(Projections.isAllowedClusteringType(ColumnType.FLOAT)); + Assertions.assertFalse(Projections.isAllowedClusteringType(null)); + Assertions.assertFalse(Projections.isAllowedClusteringType(ColumnType.STRING_ARRAY)); + Assertions.assertFalse(Projections.isAllowedClusteringType(ColumnType.UNKNOWN_COMPLEX)); + } + private static class RowSignatureChecker implements Projections.PhysicalColumnChecker { private final RowSignature rowSignature; diff --git a/processing/src/test/java/org/apache/druid/segment/projections/TableClusterGroupSpecTest.java b/processing/src/test/java/org/apache/druid/segment/projections/TableClusterGroupSpecTest.java new file mode 100644 index 000000000000..7026276108d3 --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/projections/TableClusterGroupSpecTest.java @@ -0,0 +1,324 @@ +/* + * 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.druid.segment.projections; + +import com.fasterxml.jackson.core.JsonProcessingException; +import com.fasterxml.jackson.databind.ObjectMapper; +import nl.jqno.equalsverifier.EqualsVerifier; +import nl.jqno.equalsverifier.Warning; +import org.apache.druid.error.DruidException; +import org.apache.druid.query.OrderBy; +import org.apache.druid.query.aggregation.AggregatorFactory; +import org.apache.druid.query.aggregation.CountAggregatorFactory; +import org.apache.druid.segment.TestHelper; +import org.apache.druid.segment.VirtualColumns; +import org.apache.druid.segment.column.ColumnHolder; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.List; + +class TableClusterGroupSpecTest +{ + private static final ObjectMapper JSON_MAPPER = TestHelper.makeJsonMapper(); + + // Column names that appear in more than one test fixture. + private static final String COL_TENANT = "tenant"; + private static final String COL_REGION = "region"; + private static final String COL_PRIORITY = "priority"; + + // Clustering-value literals used by more than one test. + private static final String VAL_ACME = "acme"; + private static final String VAL_GLOBEX = "globex"; + private static final String VAL_US_EAST_1 = "us-east-1"; + private static final String VAL_US_WEST_2 = "us-west-2"; + + // Common clustering signatures reused across tests. + private static final RowSignature TENANT_CLUSTER_SIGNATURE = + RowSignature.builder().add(COL_TENANT, ColumnType.STRING).build(); + private static final RowSignature TENANT_REGION_CLUSTER_SIGNATURE = + RowSignature.builder() + .add(COL_TENANT, ColumnType.STRING) + .add(COL_REGION, ColumnType.STRING) + .build(); + + /** Schema + the specs it wraps, returned together so tests can assert on both. */ + private record Built(ClusteredValueGroupsBaseTableSchema schema, List specs) + { + } + + /** Build a schema with one cluster group per supplied typed tuple. */ + private static Built buildSummary(RowSignature clusteringColumns, List> tuples) + { + final ClusterGroupSchemaTestHelpers.Built built = + ClusterGroupSchemaTestHelpers.buildClusterGroups(clusteringColumns, tuples); + final List dataColumns = List.of(ColumnHolder.TIME_COLUMN_NAME, COL_REGION, "metric"); + final ArrayList allColumns = new ArrayList<>(clusteringColumns.getColumnNames()); + allColumns.addAll(dataColumns); + final ArrayList ordering = new ArrayList<>(); + for (String col : clusteringColumns.getColumnNames()) { + ordering.add(OrderBy.ascending(col)); + } + ordering.add(OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)); + ordering.add(OrderBy.ascending(COL_REGION)); + final ClusteredValueGroupsBaseTableSchema schema = new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + allColumns, + new AggregatorFactory[]{new CountAggregatorFactory("count")}, + ordering, + clusteringColumns, + null, + built.dictionaries(), + built.specs() + ); + return new Built(schema, built.specs()); + } + + /** Typed-tuple convenience: {@link Arrays#asList} allows null entries (unlike {@link List#of}). */ + private static List tuple(Object... values) + { + return Arrays.asList(values); + } + + /** Round-trip via the enclosing summary (specs aren't ProjectionSchema subtypes); returns specs[0]. */ + private static TableClusterGroupSpec roundTrip(ClusteredValueGroupsBaseTableSchema sum) throws JsonProcessingException + { + final String json = JSON_MAPPER.writeValueAsString(sum); + final ClusteredValueGroupsBaseTableSchema deserialized = + (ClusteredValueGroupsBaseTableSchema) JSON_MAPPER.readValue(json, ProjectionSchema.class); + return deserialized.getClusterGroups().get(0); + } + + @Test + void testDictionariesAndIdsResolveFromTuples() + { + // Both columns are STRING, so their values share one merged dict. + final Built b = buildSummary( + TENANT_REGION_CLUSTER_SIGNATURE, + List.of( + tuple(VAL_GLOBEX, VAL_US_EAST_1), + tuple(VAL_ACME, VAL_US_WEST_2), + tuple(VAL_ACME, VAL_US_EAST_1) + ) + ); + Assertions.assertEquals( + List.of(VAL_ACME, VAL_GLOBEX, VAL_US_EAST_1, VAL_US_WEST_2), + b.schema().getClusteringDictionaries().getStringDictionary() + ); + Assertions.assertEquals(List.of(), b.schema().getClusteringDictionaries().getLongDictionary()); + Assertions.assertEquals(List.of(1, 2), b.specs().get(0).getClusteringValueIds()); + Assertions.assertEquals(List.of(0, 3), b.specs().get(1).getClusteringValueIds()); + Assertions.assertEquals(List.of(0, 2), b.specs().get(2).getClusteringValueIds()); + Assertions.assertArrayEquals(new Object[]{VAL_GLOBEX, VAL_US_EAST_1}, b.specs().get(0).lookupClusteringValues()); + Assertions.assertArrayEquals(new Object[]{VAL_ACME, VAL_US_WEST_2}, b.specs().get(1).lookupClusteringValues()); + Assertions.assertArrayEquals(new Object[]{VAL_ACME, VAL_US_EAST_1}, b.specs().get(2).lookupClusteringValues()); + } + + @Test + void testDictionarySortsNullsFirst() + { + // Null is placed at position 0 in the STRING dictionary, with non-nulls following in ascending order. + final Built b = buildSummary(TENANT_CLUSTER_SIGNATURE, List.of(tuple((Object) null), tuple(VAL_ACME))); + final List stringDict = b.schema().getClusteringDictionaries().getStringDictionary(); + Assertions.assertEquals(2, stringDict.size()); + Assertions.assertNull(stringDict.get(0)); + Assertions.assertEquals(VAL_ACME, stringDict.get(1)); + Assertions.assertEquals(List.of(0), b.specs().get(0).getClusteringValueIds()); + Assertions.assertEquals(List.of(1), b.specs().get(1).getClusteringValueIds()); + } + + @Test + void testSerdeStringValue() throws JsonProcessingException + { + final Built b = buildSummary(TENANT_CLUSTER_SIGNATURE, List.of(tuple(VAL_ACME))); + Assertions.assertEquals(b.specs().get(0), roundTrip(b.schema())); + } + + @Test + void testSerdeLongValueCoercedFromJsonNumber() throws JsonProcessingException + { + final Built b = buildSummary( + RowSignature.builder().add(COL_PRIORITY, ColumnType.LONG).build(), + List.of(tuple(5L)) + ); + final TableClusterGroupSpec roundTripped = roundTrip(b.schema()); + Assertions.assertEquals(b.specs().get(0), roundTripped); + Assertions.assertEquals(Long.class, roundTripped.lookupClusteringValues()[0].getClass()); + Assertions.assertEquals(5L, roundTripped.lookupClusteringValues()[0]); + } + + @Test + void testSerdeFloatValue() throws JsonProcessingException + { + final Built b = buildSummary( + RowSignature.builder().add("ratio", ColumnType.FLOAT).build(), + List.of(tuple(0.5f)) + ); + final TableClusterGroupSpec roundTripped = roundTrip(b.schema()); + Assertions.assertEquals(b.specs().get(0), roundTripped); + Assertions.assertEquals(Float.class, roundTripped.lookupClusteringValues()[0].getClass()); + Assertions.assertEquals(0.5f, roundTripped.lookupClusteringValues()[0]); + } + + @Test + void testSerdeMultiColumnWithSpecialCharsAndMixedTypes() throws JsonProcessingException + { + final Built b = buildSummary( + RowSignature.builder() + .add(COL_TENANT, ColumnType.STRING) + .add(COL_PRIORITY, ColumnType.LONG) + .build(), + List.of(tuple("A/B", 5L)) + ); + Assertions.assertEquals(b.specs().get(0), roundTrip(b.schema())); + } + + @Test + void testMixedTypeMultiGroupDictionaryRouting() + { + final Built b = buildSummary( + RowSignature.builder() + .add("device_id", ColumnType.LONG) + .add("region", ColumnType.STRING) + .build(), + List.of( + tuple(202L, "us-east-1"), + tuple(101L, "us-west-2"), + tuple(101L, "us-east-1") + ) + ); + Assertions.assertEquals(List.of(101L, 202L), b.schema().getClusteringDictionaries().getLongDictionary()); + Assertions.assertEquals( + List.of("us-east-1", "us-west-2"), + b.schema().getClusteringDictionaries().getStringDictionary() + ); + // IDs are positions in each column's typed dict; LONG for position 0, STRING for position 1. + Assertions.assertEquals(List.of(1, 0), b.specs().get(0).getClusteringValueIds()); + Assertions.assertEquals(List.of(0, 1), b.specs().get(1).getClusteringValueIds()); + Assertions.assertEquals(List.of(0, 0), b.specs().get(2).getClusteringValueIds()); + Assertions.assertArrayEquals(new Object[]{202L, "us-east-1"}, b.specs().get(0).lookupClusteringValues()); + Assertions.assertArrayEquals(new Object[]{101L, "us-west-2"}, b.specs().get(1).lookupClusteringValues()); + Assertions.assertArrayEquals(new Object[]{101L, "us-east-1"}, b.specs().get(2).lookupClusteringValues()); + } + + @Test + void testJsonShapeIsOnlyClusteringValueIds() throws JsonProcessingException + { + // Spec carries only clusteringValueIds (+ type tag); summary-owned fields must not leak in. + final Built b = buildSummary(TENANT_CLUSTER_SIGNATURE, List.of(tuple(VAL_ACME))); + final String json = JSON_MAPPER.writeValueAsString(b.schema()); + Assertions.assertTrue(json.contains("\"clusterGroups\""), "summary serializes clusterGroups"); + Assertions.assertTrue(json.contains("\"clusteringDictionaries\""), "summary serializes clusteringDictionaries"); + Assertions.assertTrue(json.contains("\"clusteringValueIds\":[0]"), "spec carries clusteringValueIds"); + // Any of the summary-only fields appearing inside the spec object would mean we duplicated them. + final int specStart = json.indexOf("\"clusterGroups\":["); + final int specEnd = json.indexOf("]", specStart); + final String specJson = json.substring(specStart, specEnd + 1); + Assertions.assertFalse(specJson.contains("\"columns\""), "spec must not carry columns"); + Assertions.assertFalse(specJson.contains("\"ordering\""), "spec must not carry ordering"); + Assertions.assertFalse(specJson.contains("\"clusteringColumns\""), "spec must not carry clusteringColumns"); + Assertions.assertFalse(specJson.contains("\"aggregators\""), "spec must not carry aggregators"); + Assertions.assertFalse(specJson.contains("\"virtualColumns\""), "spec must not carry virtualColumns"); + // serialized form is dictionary IDs, not typed values: spec must not contain the literal value. + Assertions.assertFalse(specJson.contains("\"" + VAL_ACME + "\""), "spec must not carry typed values inline"); + } + + @Test + void testNullClusteringValueRoundTrips() throws JsonProcessingException + { + final Built b = buildSummary(TENANT_CLUSTER_SIGNATURE, List.of(tuple((Object) null))); + final TableClusterGroupSpec roundTripped = roundTrip(b.schema()); + Assertions.assertEquals(b.specs().get(0), roundTripped); + Assertions.assertNull(roundTripped.lookupClusteringValues()[0]); + } + + @Test + void testMixedNullAndValueClusteringRoundTrips() throws JsonProcessingException + { + final Built b = buildSummary(TENANT_REGION_CLUSTER_SIGNATURE, List.of(tuple(VAL_ACME, null))); + Assertions.assertEquals(b.specs().get(0), roundTrip(b.schema())); + } + + @Test + void testMismatchedTupleSizeRejectedByHelper() + { + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> buildSummary(TENANT_REGION_CLUSTER_SIGNATURE, List.of(tuple(VAL_ACME))) + ); + Assertions.assertTrue(t.getMessage().contains("must match clusteringColumns size")); + } + + @Test + void testGettersBeforeSetSummaryThrow() + { + // Spec without a summary back-reference: summary-dependent getters throw. + final TableClusterGroupSpec spec = new TableClusterGroupSpec(List.of(0), null); + Assertions.assertThrows(DruidException.class, spec::lookupClusteringValues); + Assertions.assertThrows(DruidException.class, spec::getSummary); + } + + @Test + void testSetSummaryTwiceRejected() + { + final Built b = buildSummary(TENANT_CLUSTER_SIGNATURE, List.of(tuple(VAL_ACME))); + final Built other = buildSummary(TENANT_CLUSTER_SIGNATURE, List.of(tuple(VAL_GLOBEX))); + Throwable t = Assertions.assertThrows( + DruidException.class, + () -> b.specs().get(0).setSummary(other.schema()) + ); + Assertions.assertTrue(t.getMessage().contains("summary already set")); + } + + @Test + void testEqualsAndHashcode() + { + final ClusteredValueGroupsBaseTableSchema redSummary = new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of(COL_TENANT, ColumnHolder.TIME_COLUMN_NAME), + null, + List.of(OrderBy.ascending(COL_TENANT), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + TENANT_CLUSTER_SIGNATURE, + null, + ClusteringDictionaries.EMPTY, + null + ); + final ClusteredValueGroupsBaseTableSchema blueSummary = new ClusteredValueGroupsBaseTableSchema( + VirtualColumns.EMPTY, + List.of(COL_PRIORITY, ColumnHolder.TIME_COLUMN_NAME), + null, + List.of(OrderBy.ascending(COL_PRIORITY), OrderBy.ascending(ColumnHolder.TIME_COLUMN_NAME)), + RowSignature.builder().add(COL_PRIORITY, ColumnType.LONG).build(), + null, + ClusteringDictionaries.EMPTY, + null + ); + EqualsVerifier.forClass(TableClusterGroupSpec.class) + .withIgnoredFields("summary", "numRows") + .withPrefabValues(ClusteredValueGroupsBaseTableSchema.class, redSummary, blueSummary) + .suppress(Warning.NONFINAL_FIELDS) + .usingGetClass() + .verify(); + } +} diff --git a/processing/src/test/java/org/apache/druid/segment/vector/ConcatenatingVectorCursorTest.java b/processing/src/test/java/org/apache/druid/segment/vector/ConcatenatingVectorCursorTest.java new file mode 100644 index 000000000000..5a9ebf1ba2c2 --- /dev/null +++ b/processing/src/test/java/org/apache/druid/segment/vector/ConcatenatingVectorCursorTest.java @@ -0,0 +1,540 @@ +/* + * 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.druid.segment.vector; + +import com.google.common.base.Supplier; +import com.google.common.base.Suppliers; +import org.apache.druid.java.util.common.io.Closer; +import org.apache.druid.query.OrderBy; +import org.apache.druid.query.dimension.DimensionSpec; +import org.apache.druid.segment.Cursor; +import org.apache.druid.segment.CursorHolder; +import org.apache.druid.segment.column.ColumnCapabilities; +import org.apache.druid.segment.column.ColumnType; +import org.apache.druid.segment.column.RowSignature; +import org.apache.druid.segment.projections.ClusteringVectorColumnSelectorFactory; +import org.junit.jupiter.api.Assertions; +import org.junit.jupiter.api.Test; + +import javax.annotation.Nullable; +import java.util.ArrayList; +import java.util.List; + +class ConcatenatingVectorCursorTest +{ + private static final RowSignature CLUSTER_SIGNATURE = RowSignature.builder().add("tenant", ColumnType.STRING).build(); + + private final Closer closer = Closer.create(); + + @Test + void testWalksTwoNonEmptyGroupsBackToBack() + { + FakeVectorCursorHolder a = new FakeVectorCursorHolder(List.of("a1", "a2"), 4); + FakeVectorCursorHolder b = new FakeVectorCursorHolder(List.of("b1"), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"acme"} + ); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + List.of(holderSupplier(a), holderSupplier(b)), + List.of(new Object[]{"acme"}, new Object[]{"globex"}), + wrapper + ); + + VectorObjectSelector tenant = c.getColumnSelectorFactory().makeObjectSelector("tenant"); + VectorObjectSelector metric = c.getColumnSelectorFactory().makeObjectSelector("metric"); + + // Group 1 vector: ["a1", "a2"] + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals(2, c.getCurrentVectorSize()); + Object[] tenantVec1 = tenant.getObjectVector(); + Object[] metricVec1 = metric.getObjectVector(); + Assertions.assertEquals("acme", tenantVec1[0]); + Assertions.assertEquals("acme", tenantVec1[1]); + Assertions.assertEquals("a1", metricVec1[0]); + Assertions.assertEquals("a2", metricVec1[1]); + c.advance(); + + // Group 2 vector: ["b1"] + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals(1, c.getCurrentVectorSize()); + Object[] tenantVec2 = tenant.getObjectVector(); + Object[] metricVec2 = metric.getObjectVector(); + Assertions.assertEquals("globex", tenantVec2[0]); + Assertions.assertEquals("b1", metricVec2[0]); + c.advance(); + + Assertions.assertTrue(c.isDone()); + Assertions.assertDoesNotThrow(closer::close); + Assertions.assertTrue(a.closed); + Assertions.assertTrue(b.closed); + } + + @Test + void testSkipsLeadingEmptyGroup() + { + FakeVectorCursorHolder empty = new FakeVectorCursorHolder(List.of(), 4); + FakeVectorCursorHolder full = new FakeVectorCursorHolder(List.of("x"), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"placeholder"} + ); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + List.of(holderSupplier(empty), holderSupplier(full)), + List.of(new Object[]{"a"}, new Object[]{"b"}), + wrapper + ); + + VectorObjectSelector tenant = c.getColumnSelectorFactory().makeObjectSelector("tenant"); + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals("b", tenant.getObjectVector()[0]); + c.advance(); + Assertions.assertTrue(c.isDone()); + Assertions.assertDoesNotThrow(closer::close); + Assertions.assertTrue(empty.closed); + Assertions.assertTrue(full.closed); + } + + @Test + void testSkipsTrailingEmptyGroup() + { + FakeVectorCursorHolder full = new FakeVectorCursorHolder(List.of("x"), 4); + FakeVectorCursorHolder empty = new FakeVectorCursorHolder(List.of(), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"a"} + ); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + List.of(holderSupplier(full), holderSupplier(empty)), + List.of(new Object[]{"a"}, new Object[]{"b"}), + wrapper + ); + + VectorObjectSelector tenant = c.getColumnSelectorFactory().makeObjectSelector("tenant"); + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals("a", tenant.getObjectVector()[0]); + c.advance(); + Assertions.assertTrue(c.isDone()); + } + + @Test + void testAllEmptyGroups() + { + FakeVectorCursorHolder e1 = new FakeVectorCursorHolder(List.of(), 4); + FakeVectorCursorHolder e2 = new FakeVectorCursorHolder(List.of(), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"placeholder"} + ); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + List.of(holderSupplier(e1), holderSupplier(e2)), + List.of(new Object[]{"a"}, new Object[]{"b"}), + wrapper + ); + + Assertions.assertTrue(c.isDone()); + Assertions.assertEquals(0, c.getCurrentVectorSize()); + } + + @Test + void testSingleGroupDegenerateCase() + { + FakeVectorCursorHolder only = new FakeVectorCursorHolder(List.of("x", "y"), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"a"} + ); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + List.of(holderSupplier(only)), + List.of(new Object[]{"a"}), + wrapper + ); + + VectorObjectSelector tenant = c.getColumnSelectorFactory().makeObjectSelector("tenant"); + VectorObjectSelector metric = c.getColumnSelectorFactory().makeObjectSelector("metric"); + + Assertions.assertEquals(2, c.getCurrentVectorSize()); + Assertions.assertEquals("a", tenant.getObjectVector()[0]); + Assertions.assertEquals("a", tenant.getObjectVector()[1]); + Assertions.assertEquals("x", metric.getObjectVector()[0]); + Assertions.assertEquals("y", metric.getObjectVector()[1]); + c.advance(); + Assertions.assertTrue(c.isDone()); + } + + @Test + void testPartialVectorAtGroupBoundary() + { + // Group A has 3 rows but max vector size is 4; its vector is partial (size = 3, not 4). + FakeVectorCursorHolder a = new FakeVectorCursorHolder(List.of("a1", "a2", "a3"), 4); + FakeVectorCursorHolder b = new FakeVectorCursorHolder(List.of("b1", "b2"), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"acme"} + ); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + List.of(holderSupplier(a), holderSupplier(b)), + List.of(new Object[]{"acme"}, new Object[]{"globex"}), + wrapper + ); + + VectorObjectSelector tenant = c.getColumnSelectorFactory().makeObjectSelector("tenant"); + + // Group A's vector is partial; 3 of max 4. + Assertions.assertEquals(3, c.getCurrentVectorSize()); + Assertions.assertEquals("acme", tenant.getObjectVector()[0]); + c.advance(); + + // Group B's vector starts fresh; 2 rows. + Assertions.assertFalse(c.isDone()); + Assertions.assertEquals(2, c.getCurrentVectorSize()); + Assertions.assertEquals("globex", tenant.getObjectVector()[0]); + Assertions.assertEquals("globex", tenant.getObjectVector()[1]); + c.advance(); + + Assertions.assertTrue(c.isDone()); + } + + @Test + void testCloserClosesAllOpenedHolders() + { + FakeVectorCursorHolder a = new FakeVectorCursorHolder(List.of("a1"), 4); + FakeVectorCursorHolder b = new FakeVectorCursorHolder(List.of("b1"), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"x"} + ); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + List.of(holderSupplier(a), holderSupplier(b)), + List.of(new Object[]{"x"}, new Object[]{"y"}), + wrapper + ); + + c.getColumnSelectorFactory(); + c.advance(); + c.advance(); // exhausts + // Outer holder owns the closer; ConcatenatingVectorCursor itself is not Closeable. + Assertions.assertDoesNotThrow(closer::close); + Assertions.assertTrue(a.closed); + Assertions.assertTrue(b.closed); + } + + @Test + void testGroupsAreOpenedLazilyOnTransitionNotEagerly() + { + final boolean[] secondOpened = {false}; + FakeVectorCursorHolder first = new FakeVectorCursorHolder(List.of("x"), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"a"} + ); + + List> suppliers = new ArrayList<>(); + suppliers.add(holderSupplier(first)); + suppliers.add(() -> { + secondOpened[0] = true; + return new FakeVectorCursorHolder(List.of("y"), 4); + }); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + suppliers, + List.of(new Object[]{"a"}, new Object[]{"b"}), + wrapper + ); + + c.getColumnSelectorFactory(); + Assertions.assertFalse(secondOpened[0]); + + c.advance(); // exhausts first → opens second + Assertions.assertTrue(secondOpened[0]); + } + + @Test + void testMaxVectorSizeIsConfiguredValueAcrossAllStates() + { + // The configured maxVectorSize is a query-level constant; it must not be derived from the current group's + // cursor (which may not exist before init or after exhaustion). + final int configuredMaxVectorSize = 17; + FakeVectorCursorHolder a = new FakeVectorCursorHolder(List.of("a1"), 4); + + ClusteringVectorColumnSelectorFactory wrapper = new ClusteringVectorColumnSelectorFactory( + new FakeVectorFactory(List.of(), new int[]{0}, 4), + CLUSTER_SIGNATURE, + new Object[]{"acme"}, + configuredMaxVectorSize + ); + + ConcatenatingVectorCursor c = new ConcatenatingVectorCursor( + List.of(holderSupplier(a)), + List.of(new Object[]{"acme"}), + wrapper + ); + + // Pre-init. + Assertions.assertEquals(configuredMaxVectorSize, c.getMaxVectorSize()); + // Wrapper factory reports the same. + Assertions.assertEquals(configuredMaxVectorSize, c.getColumnSelectorFactory().getMaxVectorSize()); + // After init. + Assertions.assertEquals(configuredMaxVectorSize, c.getMaxVectorSize()); + c.advance(); + // Post-exhaustion. + Assertions.assertTrue(c.isDone()); + Assertions.assertEquals(configuredMaxVectorSize, c.getMaxVectorSize()); + } + + private Supplier holderSupplier(FakeVectorCursorHolder h) + { + return Suppliers.memoize(() -> closer.register(h))::get; + } + + private static class FakeVectorObjectSelector implements VectorObjectSelector + { + private final List rows; + private final int[] offset; + private final int maxVectorSize; + private final Object[] buffer; + + FakeVectorObjectSelector(List rows, int[] offset, int maxVectorSize) + { + this.rows = rows; + this.offset = offset; + this.maxVectorSize = maxVectorSize; + this.buffer = new Object[maxVectorSize]; + } + + @Override + public Object[] getObjectVector() + { + final int size = Math.min(maxVectorSize, rows.size() - offset[0]); + for (int i = 0; i < size; i++) { + buffer[i] = rows.get(offset[0] + i); + } + return buffer; + } + + @Override + public int getMaxVectorSize() + { + return maxVectorSize; + } + + @Override + public int getCurrentVectorSize() + { + return Math.min(maxVectorSize, rows.size() - offset[0]); + } + } + + private static class FakeVectorFactory implements VectorColumnSelectorFactory + { + private final List metricRows; + private final int[] offset; + private final int maxVectorSize; + + FakeVectorFactory(List metricRows, int[] offset, int maxVectorSize) + { + this.metricRows = metricRows; + this.offset = offset; + this.maxVectorSize = maxVectorSize; + } + + @Override + public ReadableVectorInspector getReadableVectorInspector() + { + return new ReadableVectorInspector() + { + @Override + public int getMaxVectorSize() + { + return maxVectorSize; + } + + @Override + public int getCurrentVectorSize() + { + return Math.min(maxVectorSize, metricRows.size() - offset[0]); + } + + @Override + public int getId() + { + return offset[0]; + } + }; + } + + @Override + public int getMaxVectorSize() + { + return maxVectorSize; + } + + @Override + public SingleValueDimensionVectorSelector makeSingleValueDimensionSelector(DimensionSpec dimensionSpec) + { + throw new UnsupportedOperationException("not used in these tests"); + } + + @Override + public MultiValueDimensionVectorSelector makeMultiValueDimensionSelector(DimensionSpec dimensionSpec) + { + throw new UnsupportedOperationException("not used in these tests"); + } + + @Override + public VectorValueSelector makeValueSelector(String column) + { + throw new UnsupportedOperationException("not used in these tests"); + } + + @Override + public VectorObjectSelector makeObjectSelector(String column) + { + if ("metric".equals(column)) { + return new FakeVectorObjectSelector(metricRows, offset, maxVectorSize); + } + throw new UnsupportedOperationException("unknown column [" + column + "]"); + } + + @Nullable + @Override + public ColumnCapabilities getColumnCapabilities(String column) + { + return null; + } + } + + private static class FakeVectorCursor implements VectorCursor + { + private final List rows; + private final int maxVectorSize; + private final int[] offset = new int[]{0}; + private final FakeVectorFactory factory; + + FakeVectorCursor(List rows, int maxVectorSize) + { + this.rows = rows; + this.maxVectorSize = maxVectorSize; + this.factory = new FakeVectorFactory(rows, offset, maxVectorSize); + } + + @Override + public VectorColumnSelectorFactory getColumnSelectorFactory() + { + return factory; + } + + @Override + public void advance() + { + offset[0] += getCurrentVectorSize(); + } + + @Override + public boolean isDone() + { + return offset[0] >= rows.size(); + } + + @Override + public void reset() + { + offset[0] = 0; + } + + @Override + public int getMaxVectorSize() + { + return maxVectorSize; + } + + @Override + public int getCurrentVectorSize() + { + return Math.min(maxVectorSize, rows.size() - offset[0]); + } + } + + private static class FakeVectorCursorHolder implements CursorHolder + { + private final List rows; + private final int maxVectorSize; + boolean closed; + + FakeVectorCursorHolder(List rows, int maxVectorSize) + { + this.rows = rows; + this.maxVectorSize = maxVectorSize; + } + + @Override + public Cursor asCursor() + { + throw new UnsupportedOperationException("vector-only holder for these tests"); + } + + @Override + public VectorCursor asVectorCursor() + { + return new FakeVectorCursor(rows, maxVectorSize); + } + + @Override + public boolean canVectorize() + { + return true; + } + + @Override + public List getOrdering() + { + return List.of(); + } + + @Override + public void close() + { + closed = true; + } + } +}