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

Example 16 with RowSchema

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

the class SemanticAnalyzer method genFilterPlan.

/**
 * create a filter plan. The condition and the inputs are specified.
 *
 * @param qb
 *          current query block
 * @param condn
 *          The condition to be resolved
 * @param input
 *          the input operator
 */
@SuppressWarnings("nls")
private Operator genFilterPlan(QB qb, ASTNode condn, Operator input, boolean useCaching) throws SemanticException {
    OpParseContext inputCtx = opParseCtx.get(input);
    RowResolver inputRR = inputCtx.getRowResolver();
    ExprNodeDesc filterCond = genExprNodeDesc(condn, inputRR, useCaching, isCBOExecuted());
    if (filterCond instanceof ExprNodeConstantDesc) {
        ExprNodeConstantDesc c = (ExprNodeConstantDesc) filterCond;
        if (Boolean.TRUE.equals(c.getValue())) {
            // If filter condition is TRUE, we ignore it
            return input;
        }
        if (ExprNodeDescUtils.isNullConstant(c)) {
            // If filter condition is NULL, transform to FALSE
            filterCond = new ExprNodeConstantDesc(TypeInfoFactory.booleanTypeInfo, false);
        }
    }
    Operator output = putOpInsertMap(OperatorFactory.getAndMakeChild(new FilterDesc(filterCond, false), new RowSchema(inputRR.getColumnInfos()), input), inputRR);
    if (LOG.isDebugEnabled()) {
        LOG.debug("Created Filter Plan for " + qb.getId() + " row schema: " + inputRR.toString());
    }
    return output;
}

Example 17 with RowSchema

use of org.apache.hadoop.hive.ql.exec.RowSchema 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 dest_type = qbm.getDestTypeForAlias(dest);
    // destination table if any
    Table dest_tab = null;
    // true for full ACID table and MM table
    boolean destTableIsTransactional;
    // should the destination table be written to using ACID
    boolean destTableIsFullAcid;
    boolean destTableIsTemporary = false;
    boolean destTableIsMaterialization = false;
    // destination partition if any
    Partition dest_part = null;
    // the intermediate destination directory
    Path queryTmpdir = null;
    // the final destination directory
    Path dest_path = null;
    TableDesc table_desc = 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, isMmCtas = false;
    Long writeId = null;
    HiveTxnManager txnMgr = SessionState.get().getTxnMgr();
    switch(dest_type.intValue()) {
        case QBMetaData.DEST_TABLE:
            {
                dest_tab = qbm.getDestTableForAlias(dest);
                destTableIsTransactional = AcidUtils.isTransactionalTable(dest_tab);
                destTableIsFullAcid = AcidUtils.isFullAcidTable(dest_tab);
                destTableIsTemporary = dest_tab.isTemporary();
                // Is the user trying to insert into a external tables
                checkExternalTable(dest_tab);
                partSpec = qbm.getPartSpecForAlias(dest);
                dest_path = dest_tab.getPath();
                checkImmutableTable(qb, dest_tab, dest_path, false);
                // check for partition
                List<FieldSchema> parts = dest_tab.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) {
                        dest_tab.validatePartColumnNames(partSpec, false);
                        dpCtx = new DynamicPartitionCtx(dest_tab, partSpec, conf.getVar(HiveConf.ConfVars.DEFAULTPARTITIONNAME), conf.getIntVar(HiveConf.ConfVars.DYNAMICPARTITIONMAXPARTSPERNODE));
                        qbm.setDPCtx(dest, dpCtx);
                    }
                }
                // Check for dynamic partitions.
                dpCtx = checkDynPart(qb, qbm, dest_tab, partSpec, dest);
                if (dpCtx != null && dpCtx.getSPPath() != null) {
                    dest_path = new Path(dest_tab.getPath(), dpCtx.getSPPath());
                }
                boolean isNonNativeTable = dest_tab.isNonNative();
                isMmTable = AcidUtils.isInsertOnlyTable(dest_tab.getParameters());
                if (isNonNativeTable || isMmTable) {
                    queryTmpdir = dest_path;
                } else {
                    queryTmpdir = ctx.getTempDirForFinalJobPath(dest_path);
                }
                if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                    Utilities.FILE_OP_LOGGER.trace("create filesink w/DEST_TABLE specifying " + queryTmpdir + " from " + dest_path);
                }
                if (dpCtx != null) {
                    // set the root of the temporary path where dynamic partition columns will populate
                    dpCtx.setRootPath(queryTmpdir);
                }
                // this table_desc does not contain the partitioning columns
                table_desc = Utilities.getTableDesc(dest_tab);
                // Add NOT NULL constraint check
                input = genConstraintsPlan(dest, qb, input);
                // Add sorting/bucketing if needed
                input = genBucketingSortingDest(dest, input, qb, table_desc, dest_tab, rsCtx);
                idToTableNameMap.put(String.valueOf(destTableId), dest_tab.getTableName());
                currentTableId = destTableId;
                destTableId++;
                lbCtx = constructListBucketingCtx(dest_tab.getSkewedColNames(), dest_tab.getSkewedColValues(), dest_tab.getSkewedColValueLocationMaps(), dest_tab.isStoredAsSubDirectories(), conf);
                // NOTE: specify Dynamic partitions in dest_tab for WriteEntity
                if (!isNonNativeTable) {
                    AcidUtils.Operation acidOp = AcidUtils.Operation.NOT_ACID;
                    if (destTableIsFullAcid) {
                        acidOp = getAcidType(table_desc.getOutputFileFormatClass(), dest);
                        // todo: should this be done for MM?  is it ok to use CombineHiveInputFormat with MM
                        checkAcidConstraints(qb, table_desc, dest_tab);
                    }
                    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 {
                            if (isMmTable) {
                                writeId = txnMgr.getTableWriteId(dest_tab.getDbName(), dest_tab.getTableName());
                            } else {
                                writeId = acidOp == Operation.NOT_ACID ? null : txnMgr.getTableWriteId(dest_tab.getDbName(), dest_tab.getTableName());
                            }
                        }
                    } catch (LockException ex) {
                        throw new SemanticException("Failed to allocate write Id", ex);
                    }
                    boolean isReplace = !qb.getParseInfo().isInsertIntoTable(dest_tab.getDbName(), dest_tab.getTableName());
                    ltd = new LoadTableDesc(queryTmpdir, table_desc, dpCtx, acidOp, isReplace, writeId);
                    // 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(dest_tab.getDbName(), dest_tab.getTableName());
                    LoadFileType loadType = (!isInsertInto && !destTableIsTransactional) ? LoadFileType.REPLACE_ALL : LoadFileType.KEEP_EXISTING;
                    ltd.setLoadFileType(loadType);
                    ltd.setInsertOverwrite(!isInsertInto);
                    ltd.setLbCtx(lbCtx);
                    loadTableWork.add(ltd);
                } else {
                    // This is a non-native table.
                    // We need to set stats as inaccurate.
                    setStatsForNonNativeTable(dest_tab);
                    // true if it is insert overwrite.
                    boolean overwrite = !qb.getParseInfo().isInsertIntoTable(String.format("%s.%s", dest_tab.getDbName(), dest_tab.getTableName()));
                    createInsertDesc(dest_tab, overwrite);
                }
                if (dest_tab.isMaterializedView()) {
                    materializedViewUpdateDesc = new MaterializedViewDesc(dest_tab.getFullyQualifiedName(), false, false, true);
                }
                WriteEntity output = generateTableWriteEntity(dest, dest_tab, partSpec, ltd, dpCtx, isNonNativeTable);
                ctx.getLoadTableOutputMap().put(ltd, output);
                break;
            }
        case QBMetaData.DEST_PARTITION:
            {
                dest_part = qbm.getDestPartitionForAlias(dest);
                dest_tab = dest_part.getTable();
                destTableIsTransactional = AcidUtils.isTransactionalTable(dest_tab);
                destTableIsFullAcid = AcidUtils.isFullAcidTable(dest_tab);
                checkExternalTable(dest_tab);
                Path tabPath = dest_tab.getPath();
                Path partPath = dest_part.getDataLocation();
                checkImmutableTable(qb, dest_tab, partPath, true);
                // if the table is in a different dfs than the partition,
                // replace the partition's dfs with the table's dfs.
                dest_path = new Path(tabPath.toUri().getScheme(), tabPath.toUri().getAuthority(), partPath.toUri().getPath());
                isMmTable = AcidUtils.isInsertOnlyTable(dest_tab.getParameters());
                queryTmpdir = isMmTable ? dest_path : ctx.getTempDirForFinalJobPath(dest_path);
                if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                    Utilities.FILE_OP_LOGGER.trace("create filesink w/DEST_PARTITION specifying " + queryTmpdir + " from " + dest_path);
                }
                table_desc = Utilities.getTableDesc(dest_tab);
                // Add NOT NULL constraint check
                input = genConstraintsPlan(dest, qb, input);
                // Add sorting/bucketing if needed
                input = genBucketingSortingDest(dest, input, qb, table_desc, dest_tab, rsCtx);
                idToTableNameMap.put(String.valueOf(destTableId), dest_tab.getTableName());
                currentTableId = destTableId;
                destTableId++;
                lbCtx = constructListBucketingCtx(dest_part.getSkewedColNames(), dest_part.getSkewedColValues(), dest_part.getSkewedColValueLocationMaps(), dest_part.isStoredAsSubDirectories(), conf);
                AcidUtils.Operation acidOp = AcidUtils.Operation.NOT_ACID;
                if (destTableIsFullAcid) {
                    acidOp = getAcidType(table_desc.getOutputFileFormatClass(), dest);
                    // todo: should this be done for MM?  is it ok to use CombineHiveInputFormat with MM?
                    checkAcidConstraints(qb, table_desc, dest_tab);
                }
                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 {
                        if (isMmTable) {
                            writeId = txnMgr.getTableWriteId(dest_tab.getDbName(), dest_tab.getTableName());
                        } else {
                            writeId = (acidOp == Operation.NOT_ACID) ? null : txnMgr.getTableWriteId(dest_tab.getDbName(), dest_tab.getTableName());
                        }
                    }
                } catch (LockException ex) {
                    throw new SemanticException("Failed to allocate write Id", ex);
                }
                ltd = new LoadTableDesc(queryTmpdir, table_desc, dest_part.getSpec(), acidOp, writeId);
                // 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(dest_tab.getDbName(), dest_tab.getTableName());
                LoadFileType loadType = (!isInsertInto && !destTableIsTransactional) ? LoadFileType.REPLACE_ALL : LoadFileType.KEEP_EXISTING;
                ltd.setLoadFileType(loadType);
                ltd.setInsertOverwrite(!isInsertInto);
                ltd.setLbCtx(lbCtx);
                loadTableWork.add(ltd);
                if (!outputs.add(new WriteEntity(dest_part, determineWriteType(ltd, dest_tab.isNonNative(), dest)))) {
                    throw new SemanticException(ErrorMsg.OUTPUT_SPECIFIED_MULTIPLE_TIMES.getMsg(dest_tab.getTableName() + "@" + dest_part.getName()));
                }
                break;
            }
        case QBMetaData.DEST_LOCAL_FILE:
            isLocal = true;
        // fall through
        case QBMetaData.DEST_DFS_FILE:
            {
                dest_path = new Path(qbm.getDestFileForAlias(dest));
                ArrayList<ColumnInfo> colInfos = inputRR.getColumnInfos();
                // CTAS case: the file output format and serde are defined by the create
                // table command rather than taking the default value
                List<FieldSchema> field_schemas = null;
                CreateTableDesc tblDesc = qb.getTableDesc();
                CreateViewDesc viewDesc = qb.getViewDesc();
                boolean isCtas = false;
                if (tblDesc != null) {
                    field_schemas = new ArrayList<FieldSchema>();
                    destTableIsTemporary = tblDesc.isTemporary();
                    destTableIsMaterialization = tblDesc.isMaterialization();
                    if (AcidUtils.isInsertOnlyTable(tblDesc.getTblProps(), true)) {
                        isMmTable = isMmCtas = true;
                        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(tblDesc.getDatabaseName(), tblDesc.getTableName());
                            }
                        } catch (LockException ex) {
                            throw new SemanticException("Failed to allocate write Id", ex);
                        }
                        tblDesc.setInitialMmWriteId(writeId);
                    }
                } else if (viewDesc != null) {
                    field_schemas = new ArrayList<FieldSchema>();
                    destTableIsTemporary = false;
                }
                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(dest_path, conf);
                        queryTmpdir = isMmTable ? qPath : ctx.getTempDirForFinalJobPath(qPath);
                        if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                            Utilities.FILE_OP_LOGGER.trace("Setting query directory " + queryTmpdir + " from " + dest_path + " (" + isMmTable + ")");
                        }
                    } catch (Exception e) {
                        throw new SemanticException("Error creating temporary folder on: " + dest_path, e);
                    }
                }
                ColsAndTypes ct = deriveFileSinkColTypes(inputRR, field_schemas);
                String cols = ct.cols, colTypes = ct.colTypes;
                // update the create table descriptor with the resulting schema.
                if (tblDesc != null) {
                    tblDesc.setCols(new ArrayList<FieldSchema>(field_schemas));
                } else if (viewDesc != null) {
                    viewDesc.setSchema(new ArrayList<FieldSchema>(field_schemas));
                }
                destTableIsTransactional = tblDesc != null && AcidUtils.isTransactionalTable(tblDesc);
                destTableIsFullAcid = tblDesc != null && AcidUtils.isFullAcidTable(tblDesc);
                boolean isDestTempFile = true;
                if (!ctx.isMRTmpFileURI(dest_path.toUri().toString())) {
                    idToTableNameMap.put(String.valueOf(destTableId), dest_path.toUri().toString());
                    currentTableId = destTableId;
                    destTableId++;
                    isDestTempFile = false;
                }
                boolean isDfsDir = (dest_type.intValue() == QBMetaData.DEST_DFS_FILE);
                // Create LFD even for MM CTAS - it's a no-op move, but it still seems to be used for stats.
                loadFileWork.add(new LoadFileDesc(tblDesc, viewDesc, queryTmpdir, dest_path, isDfsDir, cols, colTypes, // there is a change here - prev version had 'transadtional', one beofre' acid'
                destTableIsFullAcid ? Operation.INSERT : Operation.NOT_ACID, isMmCtas));
                if (tblDesc == null) {
                    if (viewDesc != null) {
                        table_desc = PlanUtils.getTableDesc(viewDesc, cols, colTypes);
                    } else if (qb.getIsQuery()) {
                        String fileFormat;
                        if (SessionState.get().getIsUsingThriftJDBCBinarySerDe()) {
                            fileFormat = "SequenceFile";
                            HiveConf.setVar(conf, HiveConf.ConfVars.HIVEQUERYRESULTFILEFORMAT, fileFormat);
                            table_desc = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, fileFormat, ThriftJDBCBinarySerDe.class);
                            // 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 {
                            fileFormat = HiveConf.getVar(conf, HiveConf.ConfVars.HIVEQUERYRESULTFILEFORMAT);
                            Class<? extends Deserializer> serdeClass = LazySimpleSerDe.class;
                            if (fileFormat.equals(PlanUtils.LLAP_OUTPUT_FORMAT_KEY)) {
                                serdeClass = LazyBinarySerDe2.class;
                            }
                            table_desc = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, fileFormat, serdeClass);
                        }
                    } else {
                        table_desc = PlanUtils.getDefaultTableDesc(qb.getDirectoryDesc(), cols, colTypes);
                    }
                } else {
                    table_desc = PlanUtils.getTableDesc(tblDesc, cols, colTypes);
                }
                if (!outputs.add(new WriteEntity(dest_path, !isDfsDir, isDestTempFile))) {
                    throw new SemanticException(ErrorMsg.OUTPUT_SPECIFIED_MULTIPLE_TIMES.getMsg(dest_path.toUri().toString()));
                }
                break;
            }
        default:
            throw new SemanticException("Unknown destination type: " + dest_type);
    }
    if (!(dest_type.intValue() == QBMetaData.DEST_DFS_FILE && qb.getIsQuery())) {
        input = genConversionSelectOperator(dest, qb, input, table_desc, dpCtx);
    }
    inputRR = opParseCtx.get(input).getRowResolver();
    ArrayList<ColumnInfo> vecCol = new ArrayList<ColumnInfo>();
    if (updating(dest) || deleting(dest)) {
        vecCol.add(new ColumnInfo(VirtualColumn.ROWID.getName(), VirtualColumn.ROWID.getTypeInfo(), "", true));
    } else {
        try {
            StructObjectInspector rowObjectInspector = (StructObjectInspector) table_desc.getDeserializer(conf).getObjectInspector();
            List<? extends StructField> fields = rowObjectInspector.getAllStructFieldRefs();
            for (int i = 0; i < fields.size(); i++) {
                vecCol.add(new ColumnInfo(fields.get(i).getFieldName(), TypeInfoUtils.getTypeInfoFromObjectInspector(fields.get(i).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 = (dest_tab == null || !((dest_tab.getNumBuckets() > 0) || (dest_tab.getSortCols() != null && dest_tab.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 (dest_type.intValue() == QBMetaData.DEST_TABLE || dest_type.intValue() == QBMetaData.DEST_PARTITION) {
        genPartnCols(dest, input, qb, table_desc, dest_tab, rsCtx);
    }
    FileSinkDesc fileSinkDesc = createFileSinkDesc(dest, table_desc, dest_part, // this was 1/4 acid
    dest_path, // 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, dest_tab, writeId, isMmCtas, dest_type, qb);
    if (isMmCtas) {
        // Add FSD so that the LoadTask compilation could fix up its path to avoid the move.
        tableDesc.setWriter(fileSinkDesc);
    }
    if (fileSinkDesc.getInsertOverwrite()) {
        if (ltd != null) {
            ltd.setInsertOverwrite(true);
        }
    }
    if (SessionState.get().isHiveServerQuery() && null != table_desc && table_desc.getSerdeClassName().equalsIgnoreCase(ThriftJDBCBinarySerDe.class.getName()) && HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_SERVER2_THRIFT_RESULTSET_SERIALIZE_IN_TASKS)) {
        fileSinkDesc.setIsUsingThriftJDBCBinarySerDe(true);
    } else {
        fileSinkDesc.setIsUsingThriftJDBCBinarySerDe(false);
    }
    Operator output = putOpInsertMap(OperatorFactory.getAndMakeChild(fileSinkDesc, fsRS, input), inputRR);
    handleLineage(ltd, output);
    if (LOG.isDebugEnabled()) {
        LOG.debug("Created FileSink Plan for clause: " + dest + "dest_path: " + dest_path + " row schema: " + inputRR.toString());
    }
    FileSinkOperator fso = (FileSinkOperator) output;
    fso.getConf().setTable(dest_tab);
    // and it is an insert overwrite or insert into table
    if (dest_tab != null && !dest_tab.isNonNative() && conf.getBoolVar(ConfVars.HIVESTATSAUTOGATHER) && conf.getBoolVar(ConfVars.HIVESTATSCOLAUTOGATHER) && ColumnStatsAutoGatherContext.canRunAutogatherStats(fso)) {
        if (dest_type.intValue() == QBMetaData.DEST_TABLE) {
            genAutoColumnStatsGatheringPipeline(qb, table_desc, partSpec, input, qb.getParseInfo().isInsertIntoTable(dest_tab.getDbName(), dest_tab.getTableName()));
        } else if (dest_type.intValue() == QBMetaData.DEST_PARTITION) {
            genAutoColumnStatsGatheringPipeline(qb, table_desc, dest_part.getSpec(), input, qb.getParseInfo().isInsertIntoTable(dest_tab.getDbName(), dest_tab.getTableName()));
        }
    }
    return output;
}

Example 18 with RowSchema

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

the class SemanticAnalyzer method genJoinReduceSinkChild.

@SuppressWarnings("nls")
private Operator genJoinReduceSinkChild(QB qb, ExprNodeDesc[] joinKeys, Operator<?> child, String[] srcs, int tag) throws SemanticException {
    // dummy for backtracking
    Operator dummy = Operator.createDummy();
    dummy.setParentOperators(Arrays.asList(child));
    RowResolver inputRR = opParseCtx.get(child).getRowResolver();
    RowResolver outputRR = new RowResolver();
    ArrayList<String> outputColumns = new ArrayList<String>();
    ArrayList<ExprNodeDesc> reduceKeys = new ArrayList<ExprNodeDesc>();
    ArrayList<ExprNodeDesc> reduceKeysBack = new ArrayList<ExprNodeDesc>();
    // Compute join keys and store in reduceKeys
    for (ExprNodeDesc joinKey : joinKeys) {
        reduceKeys.add(joinKey);
        reduceKeysBack.add(ExprNodeDescUtils.backtrack(joinKey, dummy, child));
    }
    // Walk over the input row resolver and copy in the output
    ArrayList<ExprNodeDesc> reduceValues = new ArrayList<ExprNodeDesc>();
    ArrayList<ExprNodeDesc> reduceValuesBack = new ArrayList<ExprNodeDesc>();
    Map<String, ExprNodeDesc> colExprMap = new HashMap<String, ExprNodeDesc>();
    List<ColumnInfo> columns = inputRR.getColumnInfos();
    int[] index = new int[columns.size()];
    for (int i = 0; i < columns.size(); i++) {
        ColumnInfo colInfo = columns.get(i);
        String[] nm = inputRR.reverseLookup(colInfo.getInternalName());
        String[] nm2 = inputRR.getAlternateMappings(colInfo.getInternalName());
        ExprNodeDesc expr = new ExprNodeColumnDesc(colInfo);
        // backtrack can be null when input is script operator
        ExprNodeDesc exprBack = ExprNodeDescUtils.backtrack(expr, dummy, child);
        int kindex;
        if (exprBack == null) {
            kindex = -1;
        } else if (ExprNodeDescUtils.isConstant(exprBack)) {
            kindex = reduceKeysBack.indexOf(exprBack);
        } else {
            kindex = ExprNodeDescUtils.indexOf(exprBack, reduceKeysBack);
        }
        if (kindex >= 0) {
            ColumnInfo newColInfo = new ColumnInfo(colInfo);
            newColInfo.setInternalName(Utilities.ReduceField.KEY + ".reducesinkkey" + kindex);
            newColInfo.setTabAlias(nm[0]);
            outputRR.put(nm[0], nm[1], newColInfo);
            if (nm2 != null) {
                outputRR.addMappingOnly(nm2[0], nm2[1], newColInfo);
            }
            index[i] = kindex;
            continue;
        }
        index[i] = -reduceValues.size() - 1;
        String outputColName = getColumnInternalName(reduceValues.size());
        reduceValues.add(expr);
        reduceValuesBack.add(exprBack);
        ColumnInfo newColInfo = new ColumnInfo(colInfo);
        newColInfo.setInternalName(Utilities.ReduceField.VALUE + "." + outputColName);
        newColInfo.setTabAlias(nm[0]);
        outputRR.put(nm[0], nm[1], newColInfo);
        if (nm2 != null) {
            outputRR.addMappingOnly(nm2[0], nm2[1], newColInfo);
        }
        outputColumns.add(outputColName);
    }
    dummy.setParentOperators(null);
    int numReds = -1;
    // Use only 1 reducer in case of cartesian product
    if (reduceKeys.size() == 0) {
        numReds = 1;
        String error = StrictChecks.checkCartesian(conf);
        if (error != null) {
            throw new SemanticException(error);
        }
    }
    ReduceSinkDesc rsDesc = PlanUtils.getReduceSinkDesc(reduceKeys, reduceValues, outputColumns, false, tag, reduceKeys.size(), numReds, AcidUtils.Operation.NOT_ACID);
    ReduceSinkOperator rsOp = (ReduceSinkOperator) putOpInsertMap(OperatorFactory.getAndMakeChild(rsDesc, new RowSchema(outputRR.getColumnInfos()), child), outputRR);
    List<String> keyColNames = rsDesc.getOutputKeyColumnNames();
    for (int i = 0; i < keyColNames.size(); i++) {
        colExprMap.put(Utilities.ReduceField.KEY + "." + keyColNames.get(i), reduceKeys.get(i));
    }
    List<String> valColNames = rsDesc.getOutputValueColumnNames();
    for (int i = 0; i < valColNames.size(); i++) {
        colExprMap.put(Utilities.ReduceField.VALUE + "." + valColNames.get(i), reduceValues.get(i));
    }
    rsOp.setValueIndex(index);
    rsOp.setColumnExprMap(colExprMap);
    rsOp.setInputAliases(srcs);
    return rsOp;
}

Example 19 with RowSchema

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

the class SemanticAnalyzer method genLimitPlan.

@SuppressWarnings("nls")
private Operator genLimitPlan(String dest, QB qb, Operator input, int offset, int limit) throws SemanticException {
    // A map-only job can be optimized - instead of converting it to a
    // map-reduce job, we can have another map
    // job to do the same to avoid the cost of sorting in the map-reduce phase.
    // A better approach would be to
    // write into a local file and then have a map-only job.
    // Add the limit operator to get the value fields
    RowResolver inputRR = opParseCtx.get(input).getRowResolver();
    LimitDesc limitDesc = new LimitDesc(offset, limit);
    globalLimitCtx.setLastReduceLimitDesc(limitDesc);
    Operator limitMap = putOpInsertMap(OperatorFactory.getAndMakeChild(limitDesc, new RowSchema(inputRR.getColumnInfos()), input), inputRR);
    if (LOG.isDebugEnabled()) {
        LOG.debug("Created LimitOperator Plan for clause: " + dest + " row schema: " + inputRR.toString());
    }
    return limitMap;
}

Example 20 with RowSchema

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

the class SemanticAnalyzer method genConstraintsPlan.

private Operator genConstraintsPlan(String dest, QB qb, Operator input) throws SemanticException {
    if (deleting(dest)) {
        // for DELETE statements NOT NULL constraint need not be checked
        return input;
    }
    // MERGE statements could have inserted a cardinality violation branch, we need to avoid that
    if (mergeCardinalityViolationBranch(input)) {
        return input;
    }
    // if this is an insert into statement we might need to add constraint check
    Table targetTable = null;
    Integer dest_type = qb.getMetaData().getDestTypeForAlias(dest);
    if (dest_type == QBMetaData.DEST_TABLE) {
        targetTable = qb.getMetaData().getDestTableForAlias(dest);
    } else if (dest_type == QBMetaData.DEST_PARTITION) {
        Partition dest_part = qb.getMetaData().getDestPartitionForAlias(dest);
        targetTable = dest_part.getTable();
    } else {
        throw new SemanticException("Generating constraint check plan: Invalid target type: " + dest);
    }
    RowResolver inputRR = opParseCtx.get(input).getRowResolver();
    ExprNodeDesc nullConstraintExpr = getNotNullConstraintExpr(targetTable, input, dest);
    ExprNodeDesc checkConstraintExpr = getCheckConstraintExpr(targetTable, input, inputRR, dest);
    ExprNodeDesc combinedConstraintExpr = null;
    if (nullConstraintExpr != null && checkConstraintExpr != null) {
        assert (input.getParentOperators().size() == 1);
        combinedConstraintExpr = TypeCheckProcFactory.DefaultExprProcessor.getFuncExprNodeDesc("and", nullConstraintExpr, checkConstraintExpr);
    } else if (nullConstraintExpr != null) {
        combinedConstraintExpr = nullConstraintExpr;
    } else if (checkConstraintExpr != null) {
        combinedConstraintExpr = checkConstraintExpr;
    }
    if (combinedConstraintExpr != null) {
        ExprNodeDesc constraintUDF = TypeCheckProcFactory.DefaultExprProcessor.getFuncExprNodeDesc("enforce_constraint", combinedConstraintExpr);
        Operator newConstraintFilter = putOpInsertMap(OperatorFactory.getAndMakeChild(new FilterDesc(constraintUDF, false), new RowSchema(inputRR.getColumnInfos()), input), inputRR);
        return newConstraintFilter;
    }
    return input;
}