Examples with TableStatisticAggregation io.trino.sql.planner.StatisticsAggregationPlanner.TableStatisticAggregation used on opensource projects

Example 1 with TableStatisticAggregation

use of io.trino.sql.planner.StatisticsAggregationPlanner.TableStatisticAggregation in project trino by trinodb.

the class LogicalPlanner method createAnalyzePlan.

private RelationPlan createAnalyzePlan(Analysis analysis, Analyze analyzeStatement) {
    TableHandle targetTable = analysis.getAnalyzeTarget().orElseThrow();
    // Plan table scan
    Map<String, ColumnHandle> columnHandles = metadata.getColumnHandles(session, targetTable);
    ImmutableList.Builder<Symbol> tableScanOutputs = ImmutableList.builder();
    ImmutableMap.Builder<Symbol, ColumnHandle> symbolToColumnHandle = ImmutableMap.builder();
    ImmutableMap.Builder<String, Symbol> columnNameToSymbol = ImmutableMap.builder();
    TableMetadata tableMetadata = metadata.getTableMetadata(session, targetTable);
    for (ColumnMetadata column : tableMetadata.getColumns()) {
        Symbol symbol = symbolAllocator.newSymbol(column.getName(), column.getType());
        tableScanOutputs.add(symbol);
        symbolToColumnHandle.put(symbol, columnHandles.get(column.getName()));
        columnNameToSymbol.put(column.getName(), symbol);
    }
    TableStatisticsMetadata tableStatisticsMetadata = metadata.getStatisticsCollectionMetadata(session, targetTable.getCatalogName().getCatalogName(), tableMetadata.getMetadata());
    TableStatisticAggregation tableStatisticAggregation = statisticsAggregationPlanner.createStatisticsAggregation(tableStatisticsMetadata, columnNameToSymbol.buildOrThrow());
    StatisticAggregations statisticAggregations = tableStatisticAggregation.getAggregations();
    List<Symbol> groupingSymbols = statisticAggregations.getGroupingSymbols();
    PlanNode planNode = new StatisticsWriterNode(idAllocator.getNextId(), new AggregationNode(idAllocator.getNextId(), TableScanNode.newInstance(idAllocator.getNextId(), targetTable, tableScanOutputs.build(), symbolToColumnHandle.buildOrThrow(), false, Optional.empty()), statisticAggregations.getAggregations(), singleGroupingSet(groupingSymbols), ImmutableList.of(), AggregationNode.Step.SINGLE, Optional.empty(), Optional.empty()), new StatisticsWriterNode.WriteStatisticsReference(targetTable), symbolAllocator.newSymbol("rows", BIGINT), tableStatisticsMetadata.getTableStatistics().contains(ROW_COUNT), tableStatisticAggregation.getDescriptor());
    return new RelationPlan(planNode, analysis.getScope(analyzeStatement), planNode.getOutputSymbols(), Optional.empty());
}

Example 2 with TableStatisticAggregation

use of io.trino.sql.planner.StatisticsAggregationPlanner.TableStatisticAggregation in project trino by trinodb.

the class LogicalPlanner method createTableWriterPlan.

private RelationPlan createTableWriterPlan(Analysis analysis, PlanNode source, List<Symbol> symbols, WriterTarget target, List<String> columnNames, List<ColumnMetadata> columnMetadataList, Optional<TableLayout> writeTableLayout, TableStatisticsMetadata statisticsMetadata) {
    Optional<PartitioningScheme> partitioningScheme = Optional.empty();
    Optional<PartitioningScheme> preferredPartitioningScheme = Optional.empty();
    if (writeTableLayout.isPresent()) {
        List<Symbol> partitionFunctionArguments = new ArrayList<>();
        writeTableLayout.get().getPartitionColumns().stream().mapToInt(columnNames::indexOf).mapToObj(symbols::get).forEach(partitionFunctionArguments::add);
        List<Symbol> outputLayout = new ArrayList<>(symbols);
        Optional<PartitioningHandle> partitioningHandle = writeTableLayout.get().getPartitioning();
        if (partitioningHandle.isPresent()) {
            partitioningScheme = Optional.of(new PartitioningScheme(Partitioning.create(partitioningHandle.get(), partitionFunctionArguments), outputLayout));
        } else {
            // empty connector partitioning handle means evenly partitioning on partitioning columns
            preferredPartitioningScheme = Optional.of(new PartitioningScheme(Partitioning.create(FIXED_HASH_DISTRIBUTION, partitionFunctionArguments), outputLayout));
        }
    }
    verify(columnNames.size() == symbols.size(), "columnNames.size() != symbols.size(): %s and %s", columnNames, symbols);
    Map<String, Symbol> columnToSymbolMap = zip(columnNames.stream(), symbols.stream(), SimpleImmutableEntry::new).collect(toImmutableMap(Entry::getKey, Entry::getValue));
    Set<Symbol> notNullColumnSymbols = columnMetadataList.stream().filter(column -> !column.isNullable()).map(ColumnMetadata::getName).map(columnToSymbolMap::get).collect(toImmutableSet());
    if (!statisticsMetadata.isEmpty()) {
        TableStatisticAggregation result = statisticsAggregationPlanner.createStatisticsAggregation(statisticsMetadata, columnToSymbolMap);
        StatisticAggregations.Parts aggregations = result.getAggregations().createPartialAggregations(symbolAllocator, plannerContext);
        // partial aggregation is run within the TableWriteOperator to calculate the statistics for
        // the data consumed by the TableWriteOperator
        // final aggregation is run within the TableFinishOperator to summarize collected statistics
        // by the partial aggregation from all of the writer nodes
        StatisticAggregations partialAggregation = aggregations.getPartialAggregation();
        TableFinishNode commitNode = new TableFinishNode(idAllocator.getNextId(), new TableWriterNode(idAllocator.getNextId(), source, target, symbolAllocator.newSymbol("partialrows", BIGINT), symbolAllocator.newSymbol("fragment", VARBINARY), symbols, columnNames, notNullColumnSymbols, partitioningScheme, preferredPartitioningScheme, Optional.of(partialAggregation), Optional.of(result.getDescriptor().map(aggregations.getMappings()::get))), target, symbolAllocator.newSymbol("rows", BIGINT), Optional.of(aggregations.getFinalAggregation()), Optional.of(result.getDescriptor()));
        return new RelationPlan(commitNode, analysis.getRootScope(), commitNode.getOutputSymbols(), Optional.empty());
    }
    TableFinishNode commitNode = new TableFinishNode(idAllocator.getNextId(), new TableWriterNode(idAllocator.getNextId(), source, target, symbolAllocator.newSymbol("partialrows", BIGINT), symbolAllocator.newSymbol("fragment", VARBINARY), symbols, columnNames, notNullColumnSymbols, partitioningScheme, preferredPartitioningScheme, Optional.empty(), Optional.empty()), target, symbolAllocator.newSymbol("rows", BIGINT), Optional.empty(), Optional.empty());
    return new RelationPlan(commitNode, analysis.getRootScope(), commitNode.getOutputSymbols(), Optional.empty());
}