Examples with ComputedStatistics io.trino.spi.statistics.ComputedStatistics used on opensource projects

Example 1 with ComputedStatistics

use of io.trino.spi.statistics.ComputedStatistics in project trino by trinodb.

the class HiveMetadata method finishInsert.

@Override
public Optional<ConnectorOutputMetadata> finishInsert(ConnectorSession session, ConnectorInsertTableHandle insertHandle, Collection<Slice> fragments, Collection<ComputedStatistics> computedStatistics) {
    HiveInsertTableHandle handle = (HiveInsertTableHandle) insertHandle;
    List<PartitionUpdate> partitionUpdates = fragments.stream().map(Slice::getBytes).map(partitionUpdateCodec::fromJson).collect(toImmutableList());
    HiveStorageFormat tableStorageFormat = handle.getTableStorageFormat();
    partitionUpdates = PartitionUpdate.mergePartitionUpdates(partitionUpdates);
    Table table = metastore.getTable(handle.getSchemaName(), handle.getTableName()).orElseThrow(() -> new TableNotFoundException(handle.getSchemaTableName()));
    if (!table.getStorage().getStorageFormat().getInputFormat().equals(tableStorageFormat.getInputFormat()) && isRespectTableFormat(session)) {
        throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during insert");
    }
    if (handle.getBucketProperty().isPresent() && isCreateEmptyBucketFiles(session)) {
        List<PartitionUpdate> partitionUpdatesForMissingBuckets = computePartitionUpdatesForMissingBuckets(session, handle, table, false, partitionUpdates);
        // replace partitionUpdates before creating the empty files so that those files will be cleaned up if we end up rollback
        partitionUpdates = PartitionUpdate.mergePartitionUpdates(concat(partitionUpdates, partitionUpdatesForMissingBuckets));
        for (PartitionUpdate partitionUpdate : partitionUpdatesForMissingBuckets) {
            Optional<Partition> partition = table.getPartitionColumns().isEmpty() ? Optional.empty() : Optional.of(buildPartitionObject(session, table, partitionUpdate));
            if (handle.isTransactional() && partition.isPresent()) {
                PartitionStatistics statistics = PartitionStatistics.builder().setBasicStatistics(partitionUpdate.getStatistics()).build();
                metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition.get(), partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), statistics, handle.isRetriesEnabled());
            }
            createEmptyFiles(session, partitionUpdate.getWritePath(), table, partition, partitionUpdate.getFileNames());
        }
    }
    List<String> partitionedBy = table.getPartitionColumns().stream().map(Column::getName).collect(toImmutableList());
    Map<String, Type> columnTypes = handle.getInputColumns().stream().collect(toImmutableMap(HiveColumnHandle::getName, column -> column.getHiveType().getType(typeManager)));
    Map<List<String>, ComputedStatistics> partitionComputedStatistics = createComputedStatisticsToPartitionMap(computedStatistics, partitionedBy, columnTypes);
    for (PartitionUpdate partitionUpdate : partitionUpdates) {
        if (partitionUpdate.getName().isEmpty()) {
            // insert into unpartitioned table
            if (!table.getStorage().getStorageFormat().getInputFormat().equals(handle.getPartitionStorageFormat().getInputFormat()) && isRespectTableFormat(session)) {
                throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Table format changed during insert");
            }
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, ImmutableList.of()));
            if (partitionUpdate.getUpdateMode() == OVERWRITE) {
                // get privileges from existing table
                PrincipalPrivileges principalPrivileges = fromHivePrivilegeInfos(metastore.listTablePrivileges(handle.getSchemaName(), handle.getTableName(), Optional.empty()));
                // first drop it
                metastore.dropTable(session, handle.getSchemaName(), handle.getTableName());
                // create the table with the new location
                metastore.createTable(session, table, principalPrivileges, Optional.of(partitionUpdate.getWritePath()), Optional.of(partitionUpdate.getFileNames()), false, partitionStatistics, handle.isRetriesEnabled());
            } else if (partitionUpdate.getUpdateMode() == NEW || partitionUpdate.getUpdateMode() == APPEND) {
                // insert into unpartitioned table
                metastore.finishInsertIntoExistingTable(session, handle.getSchemaName(), handle.getTableName(), partitionUpdate.getWritePath(), partitionUpdate.getFileNames(), partitionStatistics, handle.isRetriesEnabled());
            } else {
                throw new IllegalArgumentException("Unsupported update mode: " + partitionUpdate.getUpdateMode());
            }
        } else if (partitionUpdate.getUpdateMode() == APPEND) {
            // insert into existing partition
            List<String> partitionValues = toPartitionValues(partitionUpdate.getName());
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partitionValues));
            metastore.finishInsertIntoExistingPartition(session, handle.getSchemaName(), handle.getTableName(), partitionValues, partitionUpdate.getWritePath(), partitionUpdate.getFileNames(), partitionStatistics, handle.isRetriesEnabled());
        } else if (partitionUpdate.getUpdateMode() == NEW || partitionUpdate.getUpdateMode() == OVERWRITE) {
            // insert into new partition or overwrite existing partition
            Partition partition = buildPartitionObject(session, table, partitionUpdate);
            if (!partition.getStorage().getStorageFormat().getInputFormat().equals(handle.getPartitionStorageFormat().getInputFormat()) && isRespectTableFormat(session)) {
                throw new TrinoException(HIVE_CONCURRENT_MODIFICATION_DETECTED, "Partition format changed during insert");
            }
            PartitionStatistics partitionStatistics = createPartitionStatistics(partitionUpdate.getStatistics(), columnTypes, getColumnStatistics(partitionComputedStatistics, partition.getValues()));
            if (partitionUpdate.getUpdateMode() == OVERWRITE) {
                if (handle.getLocationHandle().getWriteMode() == DIRECT_TO_TARGET_EXISTING_DIRECTORY) {
                    removeNonCurrentQueryFiles(session, partitionUpdate.getTargetPath());
                    if (handle.isRetriesEnabled()) {
                        HdfsContext hdfsContext = new HdfsContext(session);
                        cleanExtraOutputFiles(hdfsEnvironment, hdfsContext, session.getQueryId(), partitionUpdate.getTargetPath(), ImmutableSet.copyOf(partitionUpdate.getFileNames()));
                    }
                } else {
                    metastore.dropPartition(session, handle.getSchemaName(), handle.getTableName(), partition.getValues(), true);
                    metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition, partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), partitionStatistics, handle.isRetriesEnabled());
                }
            } else {
                metastore.addPartition(session, handle.getSchemaName(), handle.getTableName(), partition, partitionUpdate.getWritePath(), Optional.of(partitionUpdate.getFileNames()), partitionStatistics, handle.isRetriesEnabled());
            }
        } else {
            throw new IllegalArgumentException(format("Unsupported update mode: %s", partitionUpdate.getUpdateMode()));
        }
    }
    if (isFullAcidTable(table.getParameters())) {
        HdfsContext context = new HdfsContext(session);
        for (PartitionUpdate update : partitionUpdates) {
            long writeId = handle.getTransaction().getWriteId();
            Path deltaDirectory = new Path(format("%s/%s/%s", table.getStorage().getLocation(), update.getName(), deltaSubdir(writeId, writeId, 0)));
            createOrcAcidVersionFile(context, deltaDirectory);
        }
    }
    return Optional.of(new HiveWrittenPartitions(partitionUpdates.stream().map(PartitionUpdate::getName).collect(toImmutableList())));
}

Example 2 with ComputedStatistics

use of io.trino.spi.statistics.ComputedStatistics in project trino by trinodb.

the class StatisticsWriterOperator method getOutput.

@Override
public Page getOutput() {
    if (state != State.FINISHING) {
        return null;
    }
    state = State.FINISHED;
    Collection<ComputedStatistics> computedStatistics = computedStatisticsBuilder.build();
    statisticsWriter.writeStatistics(computedStatistics);
    // output page will only be constructed once,
    // so a new PageBuilder is constructed (instead of using PageBuilder.reset)
    PageBuilder page = new PageBuilder(1, TYPES);
    page.declarePosition();
    BlockBuilder rowsBuilder = page.getBlockBuilder(0);
    if (rowCountEnabled) {
        BIGINT.writeLong(rowsBuilder, getRowCount(computedStatistics));
    } else {
        rowsBuilder.appendNull();
    }
    return page.build();
}

Example 3 with ComputedStatistics

use of io.trino.spi.statistics.ComputedStatistics in project trino by trinodb.

the class TestHiveGlueMetastore method testUpdatePartitionedStatisticsOnCreate.

@Test
public void testUpdatePartitionedStatisticsOnCreate() {
    SchemaTableName tableName = temporaryTable("update_partitioned_statistics_create");
    try (Transaction transaction = newTransaction()) {
        ConnectorSession session = newSession();
        ConnectorMetadata metadata = transaction.getMetadata();
        List<ColumnMetadata> columns = ImmutableList.of(new ColumnMetadata("a_column", BigintType.BIGINT), new ColumnMetadata("part_column", BigintType.BIGINT));
        ConnectorTableMetadata tableMetadata = new ConnectorTableMetadata(tableName, columns, createTableProperties(TEXTFILE, ImmutableList.of("part_column")));
        ConnectorOutputTableHandle createTableHandle = metadata.beginCreateTable(session, tableMetadata, Optional.empty(), NO_RETRIES);
        // write data
        ConnectorPageSink sink = pageSinkProvider.createPageSink(transaction.getTransactionHandle(), session, createTableHandle);
        MaterializedResult data = MaterializedResult.resultBuilder(session, BigintType.BIGINT, BigintType.BIGINT).row(1L, 1L).row(2L, 1L).row(3L, 1L).row(4L, 2L).row(5L, 2L).build();
        sink.appendPage(data.toPage());
        Collection<Slice> fragments = getFutureValue(sink.finish());
        // prepare statistics
        ComputedStatistics statistics1 = ComputedStatistics.builder(ImmutableList.of("part_column"), ImmutableList.of(singleValueBlock(1))).addTableStatistic(TableStatisticType.ROW_COUNT, singleValueBlock(3)).addColumnStatistic(new ColumnStatisticMetadata("a_column", MIN_VALUE), singleValueBlock(1)).addColumnStatistic(new ColumnStatisticMetadata("a_column", MAX_VALUE), singleValueBlock(3)).addColumnStatistic(new ColumnStatisticMetadata("a_column", NUMBER_OF_DISTINCT_VALUES), singleValueBlock(3)).addColumnStatistic(new ColumnStatisticMetadata("a_column", NUMBER_OF_NON_NULL_VALUES), singleValueBlock(3)).build();
        ComputedStatistics statistics2 = ComputedStatistics.builder(ImmutableList.of("part_column"), ImmutableList.of(singleValueBlock(2))).addTableStatistic(TableStatisticType.ROW_COUNT, singleValueBlock(2)).addColumnStatistic(new ColumnStatisticMetadata("a_column", MIN_VALUE), singleValueBlock(4)).addColumnStatistic(new ColumnStatisticMetadata("a_column", MAX_VALUE), singleValueBlock(5)).addColumnStatistic(new ColumnStatisticMetadata("a_column", NUMBER_OF_DISTINCT_VALUES), singleValueBlock(2)).addColumnStatistic(new ColumnStatisticMetadata("a_column", NUMBER_OF_NON_NULL_VALUES), singleValueBlock(2)).build();
        // finish CTAS
        metadata.finishCreateTable(session, createTableHandle, fragments, ImmutableList.of(statistics1, statistics2));
        transaction.commit();
    } finally {
        dropTable(tableName);
    }
}

Example 4 with ComputedStatistics

use of io.trino.spi.statistics.ComputedStatistics in project trino by trinodb.

the class TestHiveGlueMetastore method testUpdateStatisticsOnCreate.

@Test
public void testUpdateStatisticsOnCreate() {
    SchemaTableName tableName = temporaryTable("update_statistics_create");
    try (Transaction transaction = newTransaction()) {
        ConnectorSession session = newSession();
        ConnectorMetadata metadata = transaction.getMetadata();
        List<ColumnMetadata> columns = ImmutableList.of(new ColumnMetadata("a_column", BigintType.BIGINT));
        ConnectorTableMetadata tableMetadata = new ConnectorTableMetadata(tableName, columns, createTableProperties(TEXTFILE));
        ConnectorOutputTableHandle createTableHandle = metadata.beginCreateTable(session, tableMetadata, Optional.empty(), NO_RETRIES);
        // write data
        ConnectorPageSink sink = pageSinkProvider.createPageSink(transaction.getTransactionHandle(), session, createTableHandle);
        MaterializedResult data = MaterializedResult.resultBuilder(session, BigintType.BIGINT).row(1L).row(2L).row(3L).row(4L).row(5L).build();
        sink.appendPage(data.toPage());
        Collection<Slice> fragments = getFutureValue(sink.finish());
        // prepare statistics
        ComputedStatistics statistics = ComputedStatistics.builder(ImmutableList.of(), ImmutableList.of()).addTableStatistic(TableStatisticType.ROW_COUNT, singleValueBlock(5)).addColumnStatistic(new ColumnStatisticMetadata("a_column", MIN_VALUE), singleValueBlock(1)).addColumnStatistic(new ColumnStatisticMetadata("a_column", MAX_VALUE), singleValueBlock(5)).addColumnStatistic(new ColumnStatisticMetadata("a_column", NUMBER_OF_DISTINCT_VALUES), singleValueBlock(5)).addColumnStatistic(new ColumnStatisticMetadata("a_column", NUMBER_OF_NON_NULL_VALUES), singleValueBlock(5)).build();
        // finish CTAS
        metadata.finishCreateTable(session, createTableHandle, fragments, ImmutableList.of(statistics));
        transaction.commit();
    } finally {
        dropTable(tableName);
    }
}

Example 5 with ComputedStatistics

use of io.trino.spi.statistics.ComputedStatistics in project trino by trinodb.

the class DeltaLakeMetadata method finishStatisticsCollection.

@Override
public void finishStatisticsCollection(ConnectorSession session, ConnectorTableHandle table, Collection<ComputedStatistics> computedStatistics) {
    DeltaLakeTableHandle tableHandle = (DeltaLakeTableHandle) table;
    AnalyzeHandle analyzeHandle = tableHandle.getAnalyzeHandle().orElseThrow(() -> new IllegalArgumentException("analyzeHandle not set"));
    String location = metastore.getTableLocation(tableHandle.getSchemaTableName(), session);
    Optional<DeltaLakeStatistics> oldStatistics = statisticsAccess.readDeltaLakeStatistics(session, location);
    // more elaborate logic for handling statistics model evaluation may need to be introduced in the future
    // for now let's have a simple check rejecting update
    oldStatistics.ifPresent(statistics -> checkArgument(statistics.getModelVersion() == DeltaLakeStatistics.CURRENT_MODEL_VERSION, "Existing table statistics are incompatible, run the drop statistics procedure on this table before re-analyzing"));
    Map<String, DeltaLakeColumnStatistics> oldColumnStatistics = oldStatistics.map(DeltaLakeStatistics::getColumnStatistics).orElseGet(ImmutableMap::of);
    Map<String, DeltaLakeColumnStatistics> newColumnStatistics = toDeltaLakeColumnStatistics(computedStatistics);
    Map<String, DeltaLakeColumnStatistics> mergedColumnStatistics = new HashMap<>();
    // only keep stats for existing columns
    Set<String> newColumns = newColumnStatistics.keySet();
    oldColumnStatistics.entrySet().stream().filter(entry -> newColumns.contains(entry.getKey())).forEach(entry -> mergedColumnStatistics.put(entry.getKey(), entry.getValue()));
    newColumnStatistics.forEach((columnName, columnStatistics) -> {
        mergedColumnStatistics.merge(columnName, columnStatistics, DeltaLakeColumnStatistics::update);
    });
    Optional<Instant> maxFileModificationTime = getMaxFileModificationTime(computedStatistics);
    // We do not want to hinder our future calls to ANALYZE if one of the files we analyzed have modification time far in the future.
    // Therefore we cap the value we store in extended_stats.json to current_time as observed on Trino coordinator.
    Instant finalAlreadyAnalyzedModifiedTimeMax = Instant.now();
    if (maxFileModificationTime.isPresent()) {
        finalAlreadyAnalyzedModifiedTimeMax = Comparators.min(maxFileModificationTime.get(), finalAlreadyAnalyzedModifiedTimeMax);
    }
    // also ensure that we are not traveling back in time
    if (oldStatistics.isPresent()) {
        finalAlreadyAnalyzedModifiedTimeMax = Comparators.max(oldStatistics.get().getAlreadyAnalyzedModifiedTimeMax(), finalAlreadyAnalyzedModifiedTimeMax);
    }
    if (analyzeHandle.getColumns().isPresent() && !mergedColumnStatistics.keySet().equals(analyzeHandle.getColumns().get())) {
        // sanity validation
        throw new IllegalStateException(format("Unexpected columns in in mergedColumnStatistics %s; expected %s", mergedColumnStatistics.keySet(), analyzeHandle.getColumns().get()));
    }
    DeltaLakeStatistics mergedDeltaLakeStatistics = new DeltaLakeStatistics(finalAlreadyAnalyzedModifiedTimeMax, mergedColumnStatistics, analyzeHandle.getColumns());
    statisticsAccess.updateDeltaLakeStatistics(session, location, mergedDeltaLakeStatistics);
}