Examples with ColumnInfo org.apache.hadoop.hive.ql.exec.ColumnInfo used on opensource projects

Example 61 with ColumnInfo

use of org.apache.hadoop.hive.ql.exec.ColumnInfo in project hive by apache.

the class SemanticAnalyzer method genGroupByPlanReduceSinkOperator2MR.

/**
 * Generate the second ReduceSinkOperator for the Group By Plan
 * (parseInfo.getXXX(dest)). The new ReduceSinkOperator will be a child of
 * groupByOperatorInfo.
 *
 * The second ReduceSinkOperator will put the group by keys in the map-reduce
 * sort key, and put the partial aggregation results in the map-reduce value.
 *
 * @param numPartitionFields
 *          the number of fields in the map-reduce partition key. This should
 *          always be the same as the number of Group By keys. We should be
 *          able to remove this parameter since in this phase there is no
 *          distinct any more.
 * @return the new ReduceSinkOperator.
 * @throws SemanticException
 */
@SuppressWarnings("nls")
private Operator genGroupByPlanReduceSinkOperator2MR(QBParseInfo parseInfo, String dest, Operator groupByOperatorInfo, int numPartitionFields, int numReducers, boolean groupingSetsPresent) throws SemanticException {
    RowResolver reduceSinkInputRowResolver2 = opParseCtx.get(groupByOperatorInfo).getRowResolver();
    RowResolver reduceSinkOutputRowResolver2 = new RowResolver();
    reduceSinkOutputRowResolver2.setIsExprResolver(true);
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    List<ExprNodeDesc> reduceKeys = new ArrayList<ExprNodeDesc>();
    List<String> outputColumnNames = new ArrayList<String>();
    // Get group-by keys and store in reduceKeys
    List<ASTNode> grpByExprs = getGroupByForClause(parseInfo, dest);
    for (int i = 0; i < grpByExprs.size(); ++i) {
        ASTNode grpbyExpr = grpByExprs.get(i);
        String field = getColumnInternalName(i);
        outputColumnNames.add(field);
        TypeInfo typeInfo = reduceSinkInputRowResolver2.getExpression(grpbyExpr).getType();
        ExprNodeColumnDesc inputExpr = new ExprNodeColumnDesc(typeInfo, field, "", false);
        reduceKeys.add(inputExpr);
        ColumnInfo colInfo = new ColumnInfo(Utilities.ReduceField.KEY.toString() + "." + field, typeInfo, "", false);
        reduceSinkOutputRowResolver2.putExpression(grpbyExpr, colInfo);
        colExprMap.put(colInfo.getInternalName(), inputExpr);
    }
    // add a key for reduce sink
    if (groupingSetsPresent) {
        // Note that partitioning fields dont need to change, since it is either
        // partitioned randomly, or by all grouping keys + distinct keys
        processGroupingSetReduceSinkOperator(reduceSinkInputRowResolver2, reduceSinkOutputRowResolver2, reduceKeys, outputColumnNames, colExprMap);
    }
    // Get partial aggregation results and store in reduceValues
    List<ExprNodeDesc> reduceValues = new ArrayList<ExprNodeDesc>();
    int inputField = reduceKeys.size();
    Map<String, ASTNode> aggregationTrees = parseInfo.getAggregationExprsForClause(dest);
    for (Map.Entry<String, ASTNode> entry : aggregationTrees.entrySet()) {
        String field = getColumnInternalName(inputField);
        ASTNode t = entry.getValue();
        TypeInfo typeInfo = reduceSinkInputRowResolver2.getExpression(t).getType();
        ExprNodeColumnDesc exprDesc = new ExprNodeColumnDesc(typeInfo, field, "", false);
        reduceValues.add(exprDesc);
        inputField++;
        String col = getColumnInternalName(reduceValues.size() - 1);
        outputColumnNames.add(col);
        ColumnInfo colInfo = new ColumnInfo(Utilities.ReduceField.VALUE.toString() + "." + col, typeInfo, "", false);
        reduceSinkOutputRowResolver2.putExpression(t, colInfo);
        colExprMap.put(colInfo.getInternalName(), exprDesc);
    }
    ReduceSinkOperator rsOp = (ReduceSinkOperator) putOpInsertMap(OperatorFactory.getAndMakeChild(PlanUtils.getReduceSinkDesc(reduceKeys, reduceValues, outputColumnNames, true, -1, numPartitionFields, numReducers, AcidUtils.Operation.NOT_ACID, defaultNullOrder), new RowSchema(reduceSinkOutputRowResolver2.getColumnInfos()), groupByOperatorInfo), reduceSinkOutputRowResolver2);
    rsOp.setColumnExprMap(colExprMap);
    return rsOp;
}

Example 62 with ColumnInfo

use of org.apache.hadoop.hive.ql.exec.ColumnInfo in project hive by apache.

the class SemanticAnalyzer method processGroupingSetReduceSinkOperator.

// Process grouping set for the reduce sink operator
// For eg: consider: select key, value, count(1) from T group by key, value with rollup.
// Assuming map-side aggregation and no skew, the plan would look like:
// 
// TableScan --> Select --> GroupBy1 --> ReduceSink --> GroupBy2 --> Select --> FileSink
// 
// This function is called for ReduceSink to add the additional grouping keys introduced by
// GroupBy1 into the reduce keys.
private void processGroupingSetReduceSinkOperator(RowResolver reduceSinkInputRowResolver, RowResolver reduceSinkOutputRowResolver, List<ExprNodeDesc> reduceKeys, List<String> outputKeyColumnNames, Map<String, ExprNodeDesc> colExprMap) throws SemanticException {
    // add a key for reduce sink
    String groupingSetColumnName = reduceSinkInputRowResolver.get(null, VirtualColumn.GROUPINGID.getName()).getInternalName();
    ExprNodeDesc inputExpr = new ExprNodeColumnDesc(VirtualColumn.GROUPINGID.getTypeInfo(), groupingSetColumnName, null, false);
    reduceKeys.add(inputExpr);
    outputKeyColumnNames.add(getColumnInternalName(reduceKeys.size() - 1));
    String field = Utilities.ReduceField.KEY.toString() + "." + getColumnInternalName(reduceKeys.size() - 1);
    ColumnInfo colInfo = new ColumnInfo(field, reduceKeys.get(reduceKeys.size() - 1).getTypeInfo(), null, true);
    reduceSinkOutputRowResolver.put(null, VirtualColumn.GROUPINGID.getName(), colInfo);
    colExprMap.put(colInfo.getInternalName(), inputExpr);
}

Example 63 with ColumnInfo

use of org.apache.hadoop.hive.ql.exec.ColumnInfo in project hive by apache.

the class SemanticAnalyzer method genFileSinkPlan.

@SuppressWarnings("nls")
protected Operator genFileSinkPlan(String dest, QB qb, Operator input) throws SemanticException {
    RowResolver inputRR = opParseCtx.get(input).getRowResolver();
    QBMetaData qbm = qb.getMetaData();
    Integer destType = qbm.getDestTypeForAlias(dest);
    // destination table if any
    Table destinationTable = null;
    // true for full ACID table and MM table
    boolean destTableIsTransactional;
    // should the destination table be written to using ACID
    boolean destTableIsFullAcid;
    // should we add files directly to the final path
    boolean isDirectInsert = false;
    AcidUtils.Operation acidOperation = null;
    boolean destTableIsTemporary = false;
    boolean destTableIsMaterialization = false;
    // destination partition if any
    Partition destinationPartition = null;
    // the intermediate destination directory
    Path queryTmpdir = null;
    String moveTaskId = null;
    // the final destination directory
    Path destinationPath = null;
    TableDesc tableDescriptor = null;
    StructObjectInspector specificRowObjectInspector = null;
    int currentTableId = 0;
    boolean isLocal = false;
    SortBucketRSCtx rsCtx = new SortBucketRSCtx();
    DynamicPartitionCtx dpCtx = null;
    LoadTableDesc ltd = null;
    ListBucketingCtx lbCtx = null;
    Map<String, String> partSpec = null;
    boolean isMmTable = false, isMmCreate = false, isNonNativeTable = false;
    Long writeId = null;
    HiveTxnManager txnMgr = getTxnMgr();
    switch(destType.intValue()) {
        case QBMetaData.DEST_TABLE:
            {
                destinationTable = qbm.getDestTableForAlias(dest);
                destTableIsTransactional = AcidUtils.isTransactionalTable(destinationTable);
                destTableIsFullAcid = AcidUtils.isFullAcidTable(destinationTable);
                destTableIsTemporary = destinationTable.isTemporary();
                // Is the user trying to insert into a external tables
                checkExternalTable(destinationTable);
                partSpec = qbm.getPartSpecForAlias(dest);
                destinationPath = destinationTable.getPath();
                checkImmutableTable(qb, destinationTable, destinationPath, false);
                // check for partition
                List<FieldSchema> parts = destinationTable.getPartitionKeys();
                if (parts != null && parts.size() > 0) {
                    // table is partitioned
                    if (partSpec == null || partSpec.size() == 0) {
                        // user did NOT specify partition
                        throw new SemanticException(generateErrorMessage(qb.getParseInfo().getDestForClause(dest), ErrorMsg.NEED_PARTITION_ERROR.getMsg()));
                    }
                    dpCtx = qbm.getDPCtx(dest);
                    if (dpCtx == null) {
                        destinationTable.validatePartColumnNames(partSpec, false);
                        dpCtx = new DynamicPartitionCtx(partSpec, conf.getVar(HiveConf.ConfVars.DEFAULTPARTITIONNAME), conf.getIntVar(HiveConf.ConfVars.DYNAMICPARTITIONMAXPARTSPERNODE));
                        qbm.setDPCtx(dest, dpCtx);
                    }
                }
                // Check for dynamic partitions.
                dpCtx = checkDynPart(qb, qbm, destinationTable, partSpec, dest);
                if (dpCtx != null && dpCtx.getSPPath() != null) {
                    destinationPath = new Path(destinationTable.getPath(), dpCtx.getSPPath());
                }
                isNonNativeTable = destinationTable.isNonNative();
                isMmTable = AcidUtils.isInsertOnlyTable(destinationTable.getParameters());
                AcidUtils.Operation acidOp = AcidUtils.Operation.NOT_ACID;
                // this table_desc does not contain the partitioning columns
                tableDescriptor = Utilities.getTableDesc(destinationTable);
                if (!isNonNativeTable) {
                    if (destTableIsTransactional) {
                        acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                    }
                }
                isDirectInsert = isDirectInsert(destTableIsFullAcid, acidOp);
                acidOperation = acidOp;
                queryTmpdir = getTmpDir(isNonNativeTable, isMmTable, isDirectInsert, destinationPath);
                moveTaskId = getMoveTaskId();
                if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                    Utilities.FILE_OP_LOGGER.trace("create filesink w/DEST_TABLE specifying " + queryTmpdir + " from " + destinationPath);
                }
                if (dpCtx != null) {
                    // set the root of the temporary path where dynamic partition columns will populate
                    dpCtx.setRootPath(queryTmpdir);
                }
                // Add NOT NULL constraint check
                input = genConstraintsPlan(dest, qb, input);
                if (!qb.getIsQuery()) {
                    input = genConversionSelectOperator(dest, qb, input, destinationTable.getDeserializer(), dpCtx, parts);
                }
                if (destinationTable.isMaterializedView() && mvRebuildMode == MaterializationRebuildMode.INSERT_OVERWRITE_REBUILD) {
                    // Data organization (DISTRIBUTED, SORTED, CLUSTERED) for materialized view
                    // TODO: We only do this for a full rebuild
                    String sortColsStr = destinationTable.getProperty(Constants.MATERIALIZED_VIEW_SORT_COLUMNS);
                    String distributeColsStr = destinationTable.getProperty(Constants.MATERIALIZED_VIEW_DISTRIBUTE_COLUMNS);
                    if (sortColsStr != null || distributeColsStr != null) {
                        input = genMaterializedViewDataOrgPlan(destinationTable, sortColsStr, distributeColsStr, inputRR, input);
                    }
                } else {
                    // Add sorting/bucketing if needed
                    input = genBucketingSortingDest(dest, input, qb, tableDescriptor, destinationTable, rsCtx);
                }
                idToTableNameMap.put(String.valueOf(destTableId), destinationTable.getTableName());
                currentTableId = destTableId;
                destTableId++;
                // NOTE: specify Dynamic partitions in dest_tab for WriteEntity
                if (!isNonNativeTable || destinationTable.getStorageHandler().commitInMoveTask()) {
                    if (destTableIsTransactional) {
                        acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                        checkAcidConstraints();
                    } else {
                        lbCtx = constructListBucketingCtx(destinationTable.getSkewedColNames(), destinationTable.getSkewedColValues(), destinationTable.getSkewedColValueLocationMaps(), destinationTable.isStoredAsSubDirectories());
                    }
                    try {
                        if (ctx.getExplainConfig() != null) {
                            // For explain plan, txn won't be opened and doesn't make sense to allocate write id
                            writeId = null;
                        } else {
                            if (isMmTable) {
                                writeId = txnMgr.getTableWriteId(destinationTable.getDbName(), destinationTable.getTableName());
                            } else {
                                writeId = acidOp == Operation.NOT_ACID ? null : txnMgr.getTableWriteId(destinationTable.getDbName(), destinationTable.getTableName());
                            }
                        }
                    } catch (LockException ex) {
                        throw new SemanticException("Failed to allocate write Id", ex);
                    }
                    boolean isReplace = !qb.getParseInfo().isInsertIntoTable(destinationTable.getDbName(), destinationTable.getTableName());
                    ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, dpCtx, acidOp, isReplace, writeId);
                    if (writeId != null) {
                        ltd.setStmtId(txnMgr.getCurrentStmtId());
                    }
                    ltd.setMoveTaskId(moveTaskId);
                    // For Acid table, Insert Overwrite shouldn't replace the table content. We keep the old
                    // deltas and base and leave them up to the cleaner to clean up
                    boolean isInsertInto = qb.getParseInfo().isInsertIntoTable(destinationTable.getDbName(), destinationTable.getTableName());
                    LoadFileType loadType;
                    if (isDirectInsert) {
                        loadType = LoadFileType.IGNORE;
                    } else if (!isInsertInto && !destTableIsTransactional) {
                        loadType = LoadFileType.REPLACE_ALL;
                    } else {
                        loadType = LoadFileType.KEEP_EXISTING;
                    }
                    ltd.setLoadFileType(loadType);
                    ltd.setInsertOverwrite(!isInsertInto);
                    ltd.setIsDirectInsert(isDirectInsert);
                    ltd.setLbCtx(lbCtx);
                    loadTableWork.add(ltd);
                } else {
                    // This is a non-native table.
                    // We need to set stats as inaccurate.
                    setStatsForNonNativeTable(destinationTable.getDbName(), destinationTable.getTableName());
                    // true if it is insert overwrite.
                    boolean overwrite = !qb.getParseInfo().isInsertIntoTable(String.format("%s.%s", destinationTable.getDbName(), destinationTable.getTableName()));
                    createPreInsertDesc(destinationTable, overwrite);
                    ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, partSpec == null ? ImmutableMap.of() : partSpec);
                    ltd.setInsertOverwrite(overwrite);
                    ltd.setLoadFileType(overwrite ? LoadFileType.REPLACE_ALL : LoadFileType.KEEP_EXISTING);
                }
                if (destinationTable.isMaterializedView()) {
                    materializedViewUpdateDesc = new MaterializedViewUpdateDesc(destinationTable.getFullyQualifiedName(), false, false, true);
                }
                WriteEntity output = generateTableWriteEntity(dest, destinationTable, partSpec, ltd, dpCtx);
                ctx.getLoadTableOutputMap().put(ltd, output);
                break;
            }
        case QBMetaData.DEST_PARTITION:
            {
                destinationPartition = qbm.getDestPartitionForAlias(dest);
                destinationTable = destinationPartition.getTable();
                destTableIsTransactional = AcidUtils.isTransactionalTable(destinationTable);
                destTableIsFullAcid = AcidUtils.isFullAcidTable(destinationTable);
                checkExternalTable(destinationTable);
                Path partPath = destinationPartition.getDataLocation();
                checkImmutableTable(qb, destinationTable, partPath, true);
                // Previous behavior (HIVE-1707) used to replace the partition's dfs with the table's dfs.
                // The changes in HIVE-19891 appears to no longer support that behavior.
                destinationPath = partPath;
                if (MetaStoreUtils.isArchived(destinationPartition.getTPartition())) {
                    try {
                        String conflictingArchive = ArchiveUtils.conflictingArchiveNameOrNull(db, destinationTable, destinationPartition.getSpec());
                        String message = String.format("Insert conflict with existing archive: %s", conflictingArchive);
                        throw new SemanticException(message);
                    } catch (SemanticException err) {
                        throw err;
                    } catch (HiveException err) {
                        throw new SemanticException(err);
                    }
                }
                isNonNativeTable = destinationTable.isNonNative();
                isMmTable = AcidUtils.isInsertOnlyTable(destinationTable.getParameters());
                AcidUtils.Operation acidOp = AcidUtils.Operation.NOT_ACID;
                // this table_desc does not contain the partitioning columns
                tableDescriptor = Utilities.getTableDesc(destinationTable);
                if (!isNonNativeTable) {
                    if (destTableIsTransactional) {
                        acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                    }
                }
                isDirectInsert = isDirectInsert(destTableIsFullAcid, acidOp);
                acidOperation = acidOp;
                queryTmpdir = getTmpDir(isNonNativeTable, isMmTable, isDirectInsert, destinationPath);
                moveTaskId = getMoveTaskId();
                if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                    Utilities.FILE_OP_LOGGER.trace("create filesink w/DEST_PARTITION specifying " + queryTmpdir + " from " + destinationPath);
                }
                // Add NOT NULL constraint check
                input = genConstraintsPlan(dest, qb, input);
                if (!qb.getIsQuery()) {
                    input = genConversionSelectOperator(dest, qb, input, destinationTable.getDeserializer(), dpCtx, null);
                }
                if (destinationTable.isMaterializedView() && mvRebuildMode == MaterializationRebuildMode.INSERT_OVERWRITE_REBUILD) {
                    // Data organization (DISTRIBUTED, SORTED, CLUSTERED) for materialized view
                    // TODO: We only do this for a full rebuild
                    String sortColsStr = destinationTable.getProperty(Constants.MATERIALIZED_VIEW_SORT_COLUMNS);
                    String distributeColsStr = destinationTable.getProperty(Constants.MATERIALIZED_VIEW_DISTRIBUTE_COLUMNS);
                    if (sortColsStr != null || distributeColsStr != null) {
                        input = genMaterializedViewDataOrgPlan(destinationTable, sortColsStr, distributeColsStr, inputRR, input);
                    }
                } else {
                    // Add sorting/bucketing if needed
                    input = genBucketingSortingDest(dest, input, qb, tableDescriptor, destinationTable, rsCtx);
                }
                idToTableNameMap.put(String.valueOf(destTableId), destinationTable.getTableName());
                currentTableId = destTableId;
                destTableId++;
                if (destTableIsTransactional) {
                    acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                    checkAcidConstraints();
                } else {
                    // Transactional tables can't be list bucketed or have skewed cols
                    lbCtx = constructListBucketingCtx(destinationPartition.getSkewedColNames(), destinationPartition.getSkewedColValues(), destinationPartition.getSkewedColValueLocationMaps(), destinationPartition.isStoredAsSubDirectories());
                }
                try {
                    if (ctx.getExplainConfig() != null) {
                        // For explain plan, txn won't be opened and doesn't make sense to allocate write id
                        writeId = null;
                    } else {
                        if (isMmTable) {
                            writeId = txnMgr.getTableWriteId(destinationTable.getDbName(), destinationTable.getTableName());
                        } else {
                            writeId = (acidOp == Operation.NOT_ACID) ? null : txnMgr.getTableWriteId(destinationTable.getDbName(), destinationTable.getTableName());
                        }
                    }
                } catch (LockException ex) {
                    throw new SemanticException("Failed to allocate write Id", ex);
                }
                ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, destinationPartition.getSpec(), acidOp, writeId);
                if (writeId != null) {
                    ltd.setStmtId(txnMgr.getCurrentStmtId());
                }
                // For the current context for generating File Sink Operator, it is either INSERT INTO or INSERT OVERWRITE.
                // So the next line works.
                boolean isInsertInto = !qb.getParseInfo().isDestToOpTypeInsertOverwrite(dest);
                // For Acid table, Insert Overwrite shouldn't replace the table content. We keep the old
                // deltas and base and leave them up to the cleaner to clean up
                LoadFileType loadType;
                if (isDirectInsert) {
                    loadType = LoadFileType.IGNORE;
                } else if (!isInsertInto && !destTableIsTransactional) {
                    loadType = LoadFileType.REPLACE_ALL;
                } else {
                    loadType = LoadFileType.KEEP_EXISTING;
                }
                ltd.setLoadFileType(loadType);
                ltd.setInsertOverwrite(!isInsertInto);
                ltd.setIsDirectInsert(isDirectInsert);
                ltd.setLbCtx(lbCtx);
                ltd.setMoveTaskId(moveTaskId);
                loadTableWork.add(ltd);
                if (!outputs.add(new WriteEntity(destinationPartition, determineWriteType(ltd, dest)))) {
                    throw new SemanticException(ErrorMsg.OUTPUT_SPECIFIED_MULTIPLE_TIMES.getMsg(destinationTable.getTableName() + "@" + destinationPartition.getName()));
                }
                break;
            }
        case QBMetaData.DEST_LOCAL_FILE:
            isLocal = true;
        // fall through
        case QBMetaData.DEST_DFS_FILE:
            {
                destinationPath = getDestinationFilePath(qbm.getDestFileForAlias(dest), isMmTable);
                // CTAS case: the file output format and serde are defined by the create
                // table command rather than taking the default value
                List<FieldSchema> fieldSchemas = null;
                List<FieldSchema> partitionColumns = null;
                List<String> partitionColumnNames = null;
                List<FieldSchema> sortColumns = null;
                List<String> sortColumnNames = null;
                List<FieldSchema> distributeColumns = null;
                List<String> distributeColumnNames = null;
                List<ColumnInfo> fileSinkColInfos = null;
                List<ColumnInfo> sortColInfos = null;
                List<ColumnInfo> distributeColInfos = null;
                TableName tableName = null;
                Map<String, String> tblProps = null;
                CreateTableDesc tblDesc = qb.getTableDesc();
                CreateMaterializedViewDesc viewDesc = qb.getViewDesc();
                if (tblDesc != null) {
                    fieldSchemas = new ArrayList<>();
                    partitionColumns = new ArrayList<>();
                    partitionColumnNames = tblDesc.getPartColNames();
                    fileSinkColInfos = new ArrayList<>();
                    destTableIsTemporary = tblDesc.isTemporary();
                    destTableIsMaterialization = tblDesc.isMaterialization();
                    tableName = TableName.fromString(tblDesc.getDbTableName(), null, tblDesc.getDatabaseName());
                    tblProps = tblDesc.getTblProps();
                } else if (viewDesc != null) {
                    fieldSchemas = new ArrayList<>();
                    partitionColumns = new ArrayList<>();
                    partitionColumnNames = viewDesc.getPartColNames();
                    sortColumns = new ArrayList<>();
                    sortColumnNames = viewDesc.getSortColNames();
                    distributeColumns = new ArrayList<>();
                    distributeColumnNames = viewDesc.getDistributeColNames();
                    fileSinkColInfos = new ArrayList<>();
                    sortColInfos = new ArrayList<>();
                    distributeColInfos = new ArrayList<>();
                    destTableIsTemporary = false;
                    destTableIsMaterialization = false;
                    tableName = HiveTableName.ofNullableWithNoDefault(viewDesc.getViewName());
                    tblProps = viewDesc.getTblProps();
                }
                destTableIsTransactional = tblProps != null && AcidUtils.isTablePropertyTransactional(tblProps);
                if (destTableIsTransactional) {
                    try {
                        if (ctx.getExplainConfig() != null) {
                            // For explain plan, txn won't be opened and doesn't make sense to allocate write id
                            writeId = 0L;
                        } else {
                            writeId = txnMgr.getTableWriteId(tableName.getDb(), tableName.getTable());
                        }
                    } catch (LockException ex) {
                        throw new SemanticException("Failed to allocate write Id", ex);
                    }
                    if (AcidUtils.isInsertOnlyTable(tblProps, true)) {
                        isMmTable = isMmCreate = true;
                        if (tblDesc != null) {
                            tblDesc.setInitialMmWriteId(writeId);
                        } else {
                            viewDesc.setInitialMmWriteId(writeId);
                        }
                    }
                }
                if (isLocal) {
                    assert !isMmTable;
                    // for local directory - we always write to map-red intermediate
                    // store and then copy to local fs
                    queryTmpdir = ctx.getMRTmpPath();
                } else {
                    // no copy is required. we may want to revisit this policy in future
                    try {
                        Path qPath = FileUtils.makeQualified(destinationPath, conf);
                        queryTmpdir = isMmTable ? qPath : ctx.getTempDirForFinalJobPath(qPath);
                        if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                            Utilities.FILE_OP_LOGGER.trace("Setting query directory " + queryTmpdir + " from " + destinationPath + " (" + isMmTable + ")");
                        }
                    } catch (Exception e) {
                        throw new SemanticException("Error creating temporary folder on: " + destinationPath, e);
                    }
                }
                // Check for dynamic partitions.
                final String cols, colTypes;
                final boolean isPartitioned;
                if (dpCtx != null) {
                    throw new SemanticException("Dynamic partition context has already been created, this should not happen");
                }
                if (!CollectionUtils.isEmpty(partitionColumnNames)) {
                    ColsAndTypes ct = deriveFileSinkColTypes(inputRR, partitionColumnNames, sortColumnNames, distributeColumnNames, fieldSchemas, partitionColumns, sortColumns, distributeColumns, fileSinkColInfos, sortColInfos, distributeColInfos);
                    cols = ct.cols;
                    colTypes = ct.colTypes;
                    dpCtx = new DynamicPartitionCtx(partitionColumnNames, conf.getVar(HiveConf.ConfVars.DEFAULTPARTITIONNAME), conf.getIntVar(HiveConf.ConfVars.DYNAMICPARTITIONMAXPARTSPERNODE));
                    qbm.setDPCtx(dest, dpCtx);
                    // set the root of the temporary path where dynamic partition columns will populate
                    dpCtx.setRootPath(queryTmpdir);
                    isPartitioned = true;
                } else {
                    ColsAndTypes ct = deriveFileSinkColTypes(inputRR, sortColumnNames, distributeColumnNames, fieldSchemas, sortColumns, distributeColumns, sortColInfos, distributeColInfos);
                    cols = ct.cols;
                    colTypes = ct.colTypes;
                    isPartitioned = false;
                }
                // update the create table descriptor with the resulting schema.
                if (tblDesc != null) {
                    tblDesc.setCols(new ArrayList<>(fieldSchemas));
                    tblDesc.setPartCols(new ArrayList<>(partitionColumns));
                } else if (viewDesc != null) {
                    viewDesc.setSchema(new ArrayList<>(fieldSchemas));
                    viewDesc.setPartCols(new ArrayList<>(partitionColumns));
                    if (viewDesc.isOrganized()) {
                        viewDesc.setSortCols(new ArrayList<>(sortColumns));
                        viewDesc.setDistributeCols(new ArrayList<>(distributeColumns));
                    }
                }
                boolean isDestTempFile = true;
                if (ctx.isMRTmpFileURI(destinationPath.toUri().toString()) == false && ctx.isResultCacheDir(destinationPath) == false) {
                    // not a temp dir and not a result cache dir
                    idToTableNameMap.put(String.valueOf(destTableId), destinationPath.toUri().toString());
                    currentTableId = destTableId;
                    destTableId++;
                    isDestTempFile = false;
                }
                if (tblDesc == null) {
                    if (viewDesc != null) {
                        tableDescriptor = PlanUtils.getTableDesc(viewDesc, cols, colTypes);
                    } else if (qb.getIsQuery()) {
                        Class<? extends Deserializer> serdeClass = LazySimpleSerDe.class;
                        String fileFormat = conf.getResultFileFormat().toString();
                        if (SessionState.get().getIsUsingThriftJDBCBinarySerDe()) {
                            serdeClass = ThriftJDBCBinarySerDe.class;
                            fileFormat = ResultFileFormat.SEQUENCEFILE.toString();
                            // Set the fetch formatter to be a no-op for the ListSinkOperator, since we'll
                            // write out formatted thrift objects to SequenceFile
                            conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, NoOpFetchFormatter.class.getName());
                        } else if (fileFormat.equals(PlanUtils.LLAP_OUTPUT_FORMAT_KEY)) {
                            // If this output format is Llap, check to see if Arrow is requested
                            boolean useArrow = HiveConf.getBoolVar(conf, HiveConf.ConfVars.LLAP_OUTPUT_FORMAT_ARROW);
                            serdeClass = useArrow ? ArrowColumnarBatchSerDe.class : LazyBinarySerDe2.class;
                        }
                        tableDescriptor = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, fileFormat, serdeClass);
                    } else {
                        tableDescriptor = PlanUtils.getDefaultTableDesc(qb.getDirectoryDesc(), cols, colTypes);
                    }
                } else {
                    tableDescriptor = PlanUtils.getTableDesc(tblDesc, cols, colTypes);
                }
                // if available, set location in table desc properties
                if (tblDesc != null && tblDesc.getLocation() != null && tableDescriptor != null && !tableDescriptor.getProperties().containsKey(hive_metastoreConstants.META_TABLE_LOCATION)) {
                    tableDescriptor.getProperties().setProperty(hive_metastoreConstants.META_TABLE_LOCATION, tblDesc.getLocation());
                }
                // We need a specific rowObjectInspector in this case
                try {
                    specificRowObjectInspector = (StructObjectInspector) tableDescriptor.getDeserializer(conf).getObjectInspector();
                } catch (Exception e) {
                    throw new SemanticException(e.getMessage(), e);
                }
                boolean isDfsDir = (destType == QBMetaData.DEST_DFS_FILE);
                try {
                    destinationTable = tblDesc != null ? tblDesc.toTable(conf) : viewDesc != null ? viewDesc.toTable(conf) : null;
                } catch (HiveException e) {
                    throw new SemanticException(e);
                }
                destTableIsFullAcid = AcidUtils.isFullAcidTable(destinationTable);
                // Data organization (DISTRIBUTED, SORTED, CLUSTERED) for materialized view
                if (viewDesc != null && viewDesc.isOrganized()) {
                    input = genMaterializedViewDataOrgPlan(sortColInfos, distributeColInfos, inputRR, input);
                }
                moveTaskId = getMoveTaskId();
                if (isPartitioned) {
                    // Create a SELECT that may reorder the columns if needed
                    RowResolver rowResolver = new RowResolver();
                    List<ExprNodeDesc> columnExprs = new ArrayList<>();
                    List<String> colNames = new ArrayList<>();
                    Map<String, ExprNodeDesc> colExprMap = new HashMap<>();
                    for (int i = 0; i < fileSinkColInfos.size(); i++) {
                        ColumnInfo ci = fileSinkColInfos.get(i);
                        ExprNodeDesc columnExpr = new ExprNodeColumnDesc(ci);
                        String name = getColumnInternalName(i);
                        rowResolver.put("", name, new ColumnInfo(name, columnExpr.getTypeInfo(), "", false));
                        columnExprs.add(columnExpr);
                        colNames.add(name);
                        colExprMap.put(name, columnExpr);
                    }
                    input = putOpInsertMap(OperatorFactory.getAndMakeChild(new SelectDesc(columnExprs, colNames), new RowSchema(rowResolver.getColumnInfos()), input), rowResolver);
                    input.setColumnExprMap(colExprMap);
                    // If this is a partitioned CTAS or MV statement, we are going to create a LoadTableDesc
                    // object. Although the table does not exist in metastore, we will swap the CreateTableTask
                    // and MoveTask resulting from this LoadTable so in this specific case, first we create
                    // the metastore table, then we move and commit the partitions. At least for the time being,
                    // this order needs to be enforced because metastore expects a table to exist before we can
                    // add any partitions to it.
                    isNonNativeTable = tableDescriptor.isNonNative();
                    if (!isNonNativeTable || destinationTable.getStorageHandler().commitInMoveTask()) {
                        AcidUtils.Operation acidOp = AcidUtils.Operation.NOT_ACID;
                        if (destTableIsTransactional) {
                            acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                            checkAcidConstraints();
                        }
                        // isReplace = false in case concurrent operation is executed
                        ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, dpCtx, acidOp, false, writeId);
                        if (writeId != null) {
                            ltd.setStmtId(txnMgr.getCurrentStmtId());
                        }
                        ltd.setLoadFileType(LoadFileType.KEEP_EXISTING);
                        ltd.setInsertOverwrite(false);
                        loadTableWork.add(ltd);
                    } else {
                        // This is a non-native table.
                        // We need to set stats as inaccurate.
                        setStatsForNonNativeTable(tableDescriptor.getDbName(), tableDescriptor.getTableName());
                        ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, dpCtx.getPartSpec());
                        ltd.setInsertOverwrite(false);
                        ltd.setLoadFileType(LoadFileType.KEEP_EXISTING);
                    }
                    ltd.setMoveTaskId(moveTaskId);
                    ltd.setMdTable(destinationTable);
                    WriteEntity output = generateTableWriteEntity(dest, destinationTable, dpCtx.getPartSpec(), ltd, dpCtx);
                    ctx.getLoadTableOutputMap().put(ltd, output);
                } else {
                    // Create LFD even for MM CTAS - it's a no-op move, but it still seems to be used for stats.
                    LoadFileDesc loadFileDesc = new LoadFileDesc(tblDesc, viewDesc, queryTmpdir, destinationPath, isDfsDir, cols, colTypes, // there is a change here - prev version had 'transactional', one before 'acid'
                    destTableIsFullAcid ? Operation.INSERT : Operation.NOT_ACID, isMmCreate);
                    loadFileDesc.setMoveTaskId(moveTaskId);
                    loadFileWork.add(loadFileDesc);
                    try {
                        Path qualifiedPath = destinationPath.getFileSystem(conf).makeQualified(destinationPath);
                        if (!outputs.add(new WriteEntity(qualifiedPath, !isDfsDir, isDestTempFile))) {
                            throw new SemanticException(ErrorMsg.OUTPUT_SPECIFIED_MULTIPLE_TIMES.getMsg(destinationPath.toUri().toString()));
                        }
                    } catch (IOException ex) {
                        throw new SemanticException("Error while getting the full qualified path for the given directory: " + ex.getMessage());
                    }
                }
                break;
            }
        default:
            throw new SemanticException("Unknown destination type: " + destType);
    }
    inputRR = opParseCtx.get(input).getRowResolver();
    List<ColumnInfo> vecCol = new ArrayList<ColumnInfo>();
    if (updating(dest) || deleting(dest)) {
        vecCol.add(new ColumnInfo(VirtualColumn.ROWID.getName(), VirtualColumn.ROWID.getTypeInfo(), "", true));
    } else {
        try {
            // If we already have a specific inspector (view or directory as a target) use that
            // Otherwise use the table deserializer to get the inspector
            StructObjectInspector rowObjectInspector = specificRowObjectInspector != null ? specificRowObjectInspector : (StructObjectInspector) destinationTable.getDeserializer().getObjectInspector();
            List<? extends StructField> fields = rowObjectInspector.getAllStructFieldRefs();
            for (StructField field : fields) {
                vecCol.add(new ColumnInfo(field.getFieldName(), TypeInfoUtils.getTypeInfoFromObjectInspector(field.getFieldObjectInspector()), "", false));
            }
        } catch (Exception e) {
            throw new SemanticException(e.getMessage(), e);
        }
    }
    RowSchema fsRS = new RowSchema(vecCol);
    // The output files of a FileSink can be merged if they are either not being written to a table
    // or are being written to a table which is not bucketed
    // and table the table is not sorted
    boolean canBeMerged = (destinationTable == null || !((destinationTable.getNumBuckets() > 0) || (destinationTable.getSortCols() != null && destinationTable.getSortCols().size() > 0)));
    // If this table is working with ACID semantics, turn off merging
    canBeMerged &= !destTableIsFullAcid;
    // Generate the partition columns from the parent input
    if (destType == QBMetaData.DEST_TABLE || destType == QBMetaData.DEST_PARTITION) {
        genPartnCols(dest, input, qb, tableDescriptor, destinationTable, rsCtx);
    }
    FileSinkDesc fileSinkDesc = createFileSinkDesc(dest, tableDescriptor, destinationPartition, // this was 1/4 acid
    destinationPath, // this was 1/4 acid
    currentTableId, // this was 1/4 acid
    destTableIsFullAcid, // this was 1/4 acid
    destTableIsTemporary, destTableIsMaterialization, queryTmpdir, rsCtx, dpCtx, lbCtx, fsRS, canBeMerged, destinationTable, writeId, isMmCreate, destType, qb, isDirectInsert, acidOperation, moveTaskId);
    if (isMmCreate) {
        // Add FSD so that the LoadTask compilation could fix up its path to avoid the move.
        if (tableDesc != null) {
            tableDesc.setWriter(fileSinkDesc);
        } else {
            createVwDesc.setWriter(fileSinkDesc);
        }
    }
    if (fileSinkDesc.getInsertOverwrite()) {
        if (ltd != null) {
            ltd.setInsertOverwrite(true);
        }
    }
    if (null != tableDescriptor && useBatchingSerializer(tableDescriptor.getSerdeClassName())) {
        fileSinkDesc.setIsUsingBatchingSerDe(true);
    } else {
        fileSinkDesc.setIsUsingBatchingSerDe(false);
    }
    Operator output = putOpInsertMap(OperatorFactory.getAndMakeChild(fileSinkDesc, fsRS, input), inputRR);
    // in case of a merge statement.
    if (!isDirectInsert || acidOperation == AcidUtils.Operation.INSERT) {
        handleLineage(ltd, output);
    }
    setWriteIdForSurrogateKeys(ltd, input);
    LOG.debug("Created FileSink Plan for clause: {}dest_path: {} row schema: {}", dest, destinationPath, inputRR);
    FileSinkOperator fso = (FileSinkOperator) output;
    fso.getConf().setTable(destinationTable);
    // and it is an insert overwrite or insert into table
    if (conf.getBoolVar(ConfVars.HIVESTATSAUTOGATHER) && conf.getBoolVar(ConfVars.HIVESTATSCOLAUTOGATHER) && destinationTable != null && (!destinationTable.isNonNative() || destinationTable.getStorageHandler().commitInMoveTask()) && !destTableIsTemporary && !destTableIsMaterialization && ColumnStatsAutoGatherContext.canRunAutogatherStats(fso)) {
        if (destType == QBMetaData.DEST_TABLE) {
            genAutoColumnStatsGatheringPipeline(destinationTable, partSpec, input, qb.getParseInfo().isInsertIntoTable(destinationTable.getDbName(), destinationTable.getTableName()), false);
        } else if (destType == QBMetaData.DEST_PARTITION) {
            genAutoColumnStatsGatheringPipeline(destinationTable, destinationPartition.getSpec(), input, qb.getParseInfo().isInsertIntoTable(destinationTable.getDbName(), destinationTable.getTableName()), false);
        } else if (destType == QBMetaData.DEST_LOCAL_FILE || destType == QBMetaData.DEST_DFS_FILE) {
            // CTAS or CMV statement
            genAutoColumnStatsGatheringPipeline(destinationTable, null, input, false, true);
        }
    }
    return output;
}

Example 64 with ColumnInfo

use of org.apache.hadoop.hive.ql.exec.ColumnInfo in project hive by apache.

the class SemanticAnalyzer method createNewGroupingKey.

/*
   * Create a new grouping key for grouping id.
   * A dummy grouping id. is added. At runtime, the group by operator
   * creates 'n' rows per input row, where 'n' is the number of grouping sets.
   */
private void createNewGroupingKey(List<ExprNodeDesc> groupByKeys, List<String> outputColumnNames, RowResolver groupByOutputRowResolver, Map<String, ExprNodeDesc> colExprMap) {
    // The value for the constant does not matter. It is replaced by the grouping set
    // value for the actual implementation
    ExprNodeConstantDesc constant = new ExprNodeConstantDesc(VirtualColumn.GROUPINGID.getTypeInfo(), 0L);
    groupByKeys.add(constant);
    String field = getColumnInternalName(groupByKeys.size() - 1);
    outputColumnNames.add(field);
    groupByOutputRowResolver.put(null, VirtualColumn.GROUPINGID.getName(), new ColumnInfo(field, VirtualColumn.GROUPINGID.getTypeInfo(), null, true));
    colExprMap.put(field, constant);
}

Example 65 with ColumnInfo

use of org.apache.hadoop.hive.ql.exec.ColumnInfo in project hive by apache.

the class SemanticAnalyzer method genGroupByPlanReduceSinkOperator.

/**
 * Generate the ReduceSinkOperator for the Group By Query Block
 * (qb.getPartInfo().getXXX(dest)). The new ReduceSinkOperator will be a child
 * of inputOperatorInfo.
 *
 * It will put all Group By keys and the distinct field (if any) in the
 * map-reduce sort key, and all other fields in the map-reduce value.
 *
 * @param numPartitionFields
 *          the number of fields for map-reduce partitioning. This is usually
 *          the number of fields in the Group By keys.
 * @return the new ReduceSinkOperator.
 * @throws SemanticException
 */
@SuppressWarnings("nls")
private ReduceSinkOperator genGroupByPlanReduceSinkOperator(QB qb, String dest, Operator inputOperatorInfo, List<ASTNode> grpByExprs, int numPartitionFields, boolean changeNumPartitionFields, int numReducers, boolean mapAggrDone, boolean groupingSetsPresent) throws SemanticException {
    RowResolver reduceSinkInputRowResolver = opParseCtx.get(inputOperatorInfo).getRowResolver();
    QBParseInfo parseInfo = qb.getParseInfo();
    RowResolver reduceSinkOutputRowResolver = new RowResolver();
    reduceSinkOutputRowResolver.setIsExprResolver(true);
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    // Pre-compute group-by keys and store in reduceKeys
    List<String> outputKeyColumnNames = new ArrayList<String>();
    List<String> outputValueColumnNames = new ArrayList<String>();
    List<ExprNodeDesc> reduceKeys = getReduceKeysForReduceSink(grpByExprs, reduceSinkInputRowResolver, reduceSinkOutputRowResolver, outputKeyColumnNames, colExprMap);
    int keyLength = reduceKeys.size();
    int numOfColsRmedFromkey = grpByExprs.size() - keyLength;
    // add a key for reduce sink
    if (groupingSetsPresent) {
        // Process grouping set for the reduce sink operator
        processGroupingSetReduceSinkOperator(reduceSinkInputRowResolver, reduceSinkOutputRowResolver, reduceKeys, outputKeyColumnNames, colExprMap);
        if (changeNumPartitionFields) {
            numPartitionFields++;
        }
    }
    List<List<Integer>> distinctColIndices = getDistinctColIndicesForReduceSink(parseInfo, dest, reduceKeys, reduceSinkInputRowResolver, reduceSinkOutputRowResolver, outputKeyColumnNames, colExprMap);
    List<ExprNodeDesc> reduceValues = new ArrayList<ExprNodeDesc>();
    Map<String, ASTNode> aggregationTrees = parseInfo.getAggregationExprsForClause(dest);
    if (!mapAggrDone) {
        getReduceValuesForReduceSinkNoMapAgg(parseInfo, dest, reduceSinkInputRowResolver, reduceSinkOutputRowResolver, outputValueColumnNames, reduceValues, colExprMap);
    } else {
        // Put partial aggregation results in reduceValues
        int inputField = reduceKeys.size() + numOfColsRmedFromkey;
        for (Map.Entry<String, ASTNode> entry : aggregationTrees.entrySet()) {
            TypeInfo type = reduceSinkInputRowResolver.getColumnInfos().get(inputField).getType();
            ExprNodeColumnDesc exprDesc = new ExprNodeColumnDesc(type, getColumnInternalName(inputField), "", false);
            reduceValues.add(exprDesc);
            inputField++;
            String outputColName = getColumnInternalName(reduceValues.size() - 1);
            outputValueColumnNames.add(outputColName);
            String internalName = Utilities.ReduceField.VALUE.toString() + "." + outputColName;
            reduceSinkOutputRowResolver.putExpression(entry.getValue(), new ColumnInfo(internalName, type, null, false));
            colExprMap.put(internalName, exprDesc);
        }
    }
    ReduceSinkOperator rsOp = (ReduceSinkOperator) putOpInsertMap(OperatorFactory.getAndMakeChild(PlanUtils.getReduceSinkDesc(reduceKeys, groupingSetsPresent ? keyLength + 1 : keyLength, reduceValues, distinctColIndices, outputKeyColumnNames, outputValueColumnNames, true, -1, numPartitionFields, numReducers, AcidUtils.Operation.NOT_ACID, defaultNullOrder), new RowSchema(reduceSinkOutputRowResolver.getColumnInfos()), inputOperatorInfo), reduceSinkOutputRowResolver);
    rsOp.setColumnExprMap(colExprMap);
    return rsOp;
}