Search in sources :

Example 46 with Operator

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

the class SortMergeJoinTaskDispatcher method genSMBJoinWork.

// Convert the work in the SMB plan to a regular join
// Note that the operator tree is not fixed, only the path/alias mappings in the
// plan are fixed. The operator tree will still contain the SMBJoinOperator
private void genSMBJoinWork(MapWork currWork, SMBMapJoinOperator smbJoinOp) {
    // Remove the paths which are not part of aliasToPartitionInfo
    Map<String, PartitionDesc> aliasToPartitionInfo = currWork.getAliasToPartnInfo();
    List<Path> removePaths = new ArrayList<>();
    for (Map.Entry<Path, ArrayList<String>> entry : currWork.getPathToAliases().entrySet()) {
        boolean keepPath = false;
        for (String alias : entry.getValue()) {
            if (aliasToPartitionInfo.containsKey(alias)) {
                keepPath = true;
                break;
            }
        }
        // Remove if the path is not present
        if (!keepPath) {
            removePaths.add(entry.getKey());
        }
    }
    List<String> removeAliases = new ArrayList<String>();
    for (Path removePath : removePaths) {
        removeAliases.addAll(currWork.getPathToAliases().get(removePath));
        currWork.removePathToAlias(removePath);
        currWork.removePathToPartitionInfo(removePath);
    }
    for (String alias : removeAliases) {
        currWork.getAliasToPartnInfo().remove(alias);
        currWork.getAliasToWork().remove(alias);
    }
    // Get the MapredLocalWork
    MapredLocalWork localWork = smbJoinOp.getConf().getLocalWork();
    for (Map.Entry<String, Operator<? extends OperatorDesc>> entry : localWork.getAliasToWork().entrySet()) {
        String alias = entry.getKey();
        Operator<? extends OperatorDesc> op = entry.getValue();
        FetchWork fetchWork = localWork.getAliasToFetchWork().get(alias);
        // Add the entry in mapredwork
        currWork.getAliasToWork().put(alias, op);
        PartitionDesc partitionInfo = currWork.getAliasToPartnInfo().get(alias);
        if (fetchWork.getTblDir() != null) {
            currWork.mergeAliasedInput(alias, fetchWork.getTblDir(), partitionInfo);
        } else {
            for (Path pathDir : fetchWork.getPartDir()) {
                currWork.mergeAliasedInput(alias, pathDir, partitionInfo);
            }
        }
    }
    // Remove the dummy store operator from the tree
    for (Operator<? extends OperatorDesc> parentOp : smbJoinOp.getParentOperators()) {
        if (parentOp instanceof DummyStoreOperator) {
            Operator<? extends OperatorDesc> grandParentOp = parentOp.getParentOperators().get(0);
            smbJoinOp.replaceParent(parentOp, grandParentOp);
            grandParentOp.setChildOperators(parentOp.getChildOperators());
            parentOp.setParentOperators(null);
            parentOp.setParentOperators(null);
        }
    }
}
Also used : Path(org.apache.hadoop.fs.Path) ReduceSinkOperator(org.apache.hadoop.hive.ql.exec.ReduceSinkOperator) MapJoinOperator(org.apache.hadoop.hive.ql.exec.MapJoinOperator) FileSinkOperator(org.apache.hadoop.hive.ql.exec.FileSinkOperator) SMBMapJoinOperator(org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator) Operator(org.apache.hadoop.hive.ql.exec.Operator) DummyStoreOperator(org.apache.hadoop.hive.ql.exec.DummyStoreOperator) DummyStoreOperator(org.apache.hadoop.hive.ql.exec.DummyStoreOperator) ArrayList(java.util.ArrayList) MapredLocalWork(org.apache.hadoop.hive.ql.plan.MapredLocalWork) FetchWork(org.apache.hadoop.hive.ql.plan.FetchWork) PartitionDesc(org.apache.hadoop.hive.ql.plan.PartitionDesc) HashMap(java.util.HashMap) LinkedHashMap(java.util.LinkedHashMap) Map(java.util.Map) OperatorDesc(org.apache.hadoop.hive.ql.plan.OperatorDesc)

Example 47 with Operator

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

the class SortMergeJoinTaskDispatcher method getSMBMapJoinOp.

private SMBMapJoinOperator getSMBMapJoinOp(Operator<? extends OperatorDesc> currOp, Operator<? extends OperatorDesc> reducer) {
    SMBMapJoinOperator ret = null;
    while (true) {
        if (currOp instanceof SMBMapJoinOperator) {
            if (ret != null) {
                return null;
            }
            ret = (SMBMapJoinOperator) currOp;
        }
        // Does any operator in the tree stop the task from being converted to a conditional task
        if (!currOp.opAllowedBeforeSortMergeJoin()) {
            return null;
        }
        List<Operator<? extends OperatorDesc>> childOps = currOp.getChildOperators();
        if ((childOps == null) || (childOps.isEmpty())) {
            return reducerAllowedSMBJoinOp(reducer) ? ret : null;
        }
        // multi-table inserts not supported
        if (childOps.size() > 1) {
            return null;
        }
        currOp = childOps.get(0);
    }
}
Also used : ReduceSinkOperator(org.apache.hadoop.hive.ql.exec.ReduceSinkOperator) MapJoinOperator(org.apache.hadoop.hive.ql.exec.MapJoinOperator) FileSinkOperator(org.apache.hadoop.hive.ql.exec.FileSinkOperator) SMBMapJoinOperator(org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator) Operator(org.apache.hadoop.hive.ql.exec.Operator) DummyStoreOperator(org.apache.hadoop.hive.ql.exec.DummyStoreOperator) SMBMapJoinOperator(org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator) OperatorDesc(org.apache.hadoop.hive.ql.plan.OperatorDesc)

Example 48 with Operator

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

the class NullScanTaskDispatcher method dispatch.

@Override
public Object dispatch(Node nd, Stack<Node> stack, Object... nodeOutputs) throws SemanticException {
    Task<? extends Serializable> task = (Task<? extends Serializable>) nd;
    // create a the context for walking operators
    ParseContext parseContext = physicalContext.getParseContext();
    WalkerCtx walkerCtx = new WalkerCtx();
    List<MapWork> mapWorks = new ArrayList<MapWork>(task.getMapWork());
    Collections.sort(mapWorks, new Comparator<MapWork>() {

        @Override
        public int compare(MapWork o1, MapWork o2) {
            return o1.getName().compareTo(o2.getName());
        }
    });
    for (MapWork mapWork : mapWorks) {
        LOG.debug("Looking at: " + mapWork.getName());
        Collection<Operator<? extends OperatorDesc>> topOperators = mapWork.getAliasToWork().values();
        if (topOperators.size() == 0) {
            LOG.debug("No top operators");
            return null;
        }
        LOG.debug("Looking for table scans where optimization is applicable");
        // The dispatcher fires the processor corresponding to the closest
        // matching rule and passes the context along
        Dispatcher disp = new DefaultRuleDispatcher(null, rules, walkerCtx);
        GraphWalker ogw = new PreOrderOnceWalker(disp);
        // Create a list of topOp nodes
        ArrayList<Node> topNodes = new ArrayList<Node>();
        // Get the top Nodes for this task
        for (Operator<? extends OperatorDesc> workOperator : topOperators) {
            if (parseContext.getTopOps().values().contains(workOperator)) {
                topNodes.add(workOperator);
            }
        }
        Operator<? extends OperatorDesc> reducer = task.getReducer(mapWork);
        if (reducer != null) {
            topNodes.add(reducer);
        }
        ogw.startWalking(topNodes, null);
        LOG.debug(String.format("Found %d null table scans", walkerCtx.getMetadataOnlyTableScans().size()));
        if (walkerCtx.getMetadataOnlyTableScans().size() > 0)
            processAlias(mapWork, walkerCtx.getMetadataOnlyTableScans());
    }
    return null;
}
Also used : TableScanOperator(org.apache.hadoop.hive.ql.exec.TableScanOperator) Operator(org.apache.hadoop.hive.ql.exec.Operator) Task(org.apache.hadoop.hive.ql.exec.Task) Serializable(java.io.Serializable) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) Node(org.apache.hadoop.hive.ql.lib.Node) ArrayList(java.util.ArrayList) Dispatcher(org.apache.hadoop.hive.ql.lib.Dispatcher) DefaultRuleDispatcher(org.apache.hadoop.hive.ql.lib.DefaultRuleDispatcher) MapWork(org.apache.hadoop.hive.ql.plan.MapWork) ParseContext(org.apache.hadoop.hive.ql.parse.ParseContext) PreOrderOnceWalker(org.apache.hadoop.hive.ql.lib.PreOrderOnceWalker) OperatorDesc(org.apache.hadoop.hive.ql.plan.OperatorDesc) WalkerCtx(org.apache.hadoop.hive.ql.optimizer.physical.MetadataOnlyOptimizer.WalkerCtx) GraphWalker(org.apache.hadoop.hive.ql.lib.GraphWalker)

Example 49 with Operator

use of org.apache.hadoop.hive.ql.exec.Operator 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;
    // should the destination table be written to using ACID
    boolean destTableIsAcid = false;
    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;
    switch(dest_type.intValue()) {
        case QBMetaData.DEST_TABLE:
            {
                dest_tab = qbm.getDestTableForAlias(dest);
                destTableIsAcid = AcidUtils.isAcidTable(dest_tab);
                destTableIsTemporary = dest_tab.isTemporary();
                // Is the user trying to insert into a external tables
                if ((!conf.getBoolVar(HiveConf.ConfVars.HIVE_INSERT_INTO_EXTERNAL_TABLES)) && (dest_tab.getTableType().equals(TableType.EXTERNAL_TABLE))) {
                    throw new SemanticException(ErrorMsg.INSERT_EXTERNAL_TABLE.getMsg(dest_tab.getTableName()));
                }
                partSpec = qbm.getPartSpecForAlias(dest);
                dest_path = dest_tab.getPath();
                // verify that our destination is empty before proceeding
                if (dest_tab.isImmutable() && qb.getParseInfo().isInsertIntoTable(dest_tab.getDbName(), dest_tab.getTableName())) {
                    try {
                        FileSystem fs = dest_path.getFileSystem(conf);
                        if (!MetaStoreUtils.isDirEmpty(fs, dest_path)) {
                            LOG.warn("Attempted write into an immutable table : " + dest_tab.getTableName() + " : " + dest_path);
                            throw new SemanticException(ErrorMsg.INSERT_INTO_IMMUTABLE_TABLE.getMsg(dest_tab.getTableName()));
                        }
                    } catch (IOException ioe) {
                        LOG.warn("Error while trying to determine if immutable table has any data : " + dest_tab.getTableName() + " : " + dest_path);
                        throw new SemanticException(ErrorMsg.INSERT_INTO_IMMUTABLE_TABLE.getMsg(ioe.getMessage()));
                    }
                }
                // 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);
                    }
                    if (!HiveConf.getBoolVar(conf, HiveConf.ConfVars.DYNAMICPARTITIONING)) {
                        // allow DP
                        throw new SemanticException(generateErrorMessage(qb.getParseInfo().getDestForClause(dest), ErrorMsg.DYNAMIC_PARTITION_DISABLED.getMsg()));
                    }
                    if (dpCtx.getSPPath() != null) {
                        dest_path = new Path(dest_tab.getPath(), dpCtx.getSPPath());
                    }
                    if ((dest_tab.getNumBuckets() > 0)) {
                        dpCtx.setNumBuckets(dest_tab.getNumBuckets());
                    }
                }
                boolean isNonNativeTable = dest_tab.isNonNative();
                if (isNonNativeTable) {
                    queryTmpdir = dest_path;
                } else {
                    queryTmpdir = ctx.getTempDirForPath(dest_path, true);
                }
                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 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 (destTableIsAcid) {
                        acidOp = getAcidType(table_desc.getOutputFileFormatClass(), dest);
                        checkAcidConstraints(qb, table_desc, dest_tab);
                    }
                    ltd = new LoadTableDesc(queryTmpdir, table_desc, dpCtx, acidOp);
                    ltd.setReplace(!qb.getParseInfo().isInsertIntoTable(dest_tab.getDbName(), dest_tab.getTableName()));
                    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);
                }
                WriteEntity output = null;
                // list of dynamically created partitions are known.
                if ((dpCtx == null || dpCtx.getNumDPCols() == 0)) {
                    output = new WriteEntity(dest_tab, determineWriteType(ltd, isNonNativeTable, dest));
                    if (!outputs.add(output)) {
                        throw new SemanticException(ErrorMsg.OUTPUT_SPECIFIED_MULTIPLE_TIMES.getMsg(dest_tab.getTableName()));
                    }
                }
                if ((dpCtx != null) && (dpCtx.getNumDPCols() >= 0)) {
                    // No static partition specified
                    if (dpCtx.getNumSPCols() == 0) {
                        output = new WriteEntity(dest_tab, determineWriteType(ltd, isNonNativeTable, dest), false);
                        outputs.add(output);
                        output.setDynamicPartitionWrite(true);
                    } else // part of the partition specified
                    // Create a DummyPartition in this case. Since, the metastore does not store partial
                    // partitions currently, we need to store dummy partitions
                    {
                        try {
                            String ppath = dpCtx.getSPPath();
                            ppath = ppath.substring(0, ppath.length() - 1);
                            DummyPartition p = new DummyPartition(dest_tab, dest_tab.getDbName() + "@" + dest_tab.getTableName() + "@" + ppath, partSpec);
                            output = new WriteEntity(p, getWriteType(dest), false);
                            output.setDynamicPartitionWrite(true);
                            outputs.add(output);
                        } catch (HiveException e) {
                            throw new SemanticException(e.getMessage(), e);
                        }
                    }
                }
                ctx.getLoadTableOutputMap().put(ltd, output);
                break;
            }
        case QBMetaData.DEST_PARTITION:
            {
                dest_part = qbm.getDestPartitionForAlias(dest);
                dest_tab = dest_part.getTable();
                destTableIsAcid = AcidUtils.isAcidTable(dest_tab);
                if ((!conf.getBoolVar(HiveConf.ConfVars.HIVE_INSERT_INTO_EXTERNAL_TABLES)) && dest_tab.getTableType().equals(TableType.EXTERNAL_TABLE)) {
                    throw new SemanticException(ErrorMsg.INSERT_EXTERNAL_TABLE.getMsg(dest_tab.getTableName()));
                }
                Path tabPath = dest_tab.getPath();
                Path partPath = dest_part.getDataLocation();
                // verify that our destination is empty before proceeding
                if (dest_tab.isImmutable() && qb.getParseInfo().isInsertIntoTable(dest_tab.getDbName(), dest_tab.getTableName())) {
                    try {
                        FileSystem fs = partPath.getFileSystem(conf);
                        if (!MetaStoreUtils.isDirEmpty(fs, partPath)) {
                            LOG.warn("Attempted write into an immutable table partition : " + dest_tab.getTableName() + " : " + partPath);
                            throw new SemanticException(ErrorMsg.INSERT_INTO_IMMUTABLE_TABLE.getMsg(dest_tab.getTableName()));
                        }
                    } catch (IOException ioe) {
                        LOG.warn("Error while trying to determine if immutable table partition has any data : " + dest_tab.getTableName() + " : " + partPath);
                        throw new SemanticException(ErrorMsg.INSERT_INTO_IMMUTABLE_TABLE.getMsg(ioe.getMessage()));
                    }
                }
                // 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());
                queryTmpdir = ctx.getTempDirForPath(dest_path, true);
                table_desc = Utilities.getTableDesc(dest_tab);
                // 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 (destTableIsAcid) {
                    acidOp = getAcidType(table_desc.getOutputFileFormatClass(), dest);
                    checkAcidConstraints(qb, table_desc, dest_tab);
                }
                ltd = new LoadTableDesc(queryTmpdir, table_desc, dest_part.getSpec(), acidOp);
                ltd.setReplace(!qb.getParseInfo().isInsertIntoTable(dest_tab.getDbName(), dest_tab.getTableName()));
                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));
                if (isLocal) {
                    // for local directory - we always write to map-red intermediate
                    // store and then copy to local fs
                    queryTmpdir = ctx.getMRTmpPath();
                } else {
                    try {
                        Path qPath = FileUtils.makeQualified(dest_path, conf);
                        queryTmpdir = ctx.getTempDirForPath(qPath, true);
                    } catch (Exception e) {
                        throw new SemanticException("Error creating temporary folder on: " + dest_path, e);
                    }
                }
                String cols = "";
                String colTypes = "";
                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();
                if (tblDesc != null) {
                    field_schemas = new ArrayList<FieldSchema>();
                    destTableIsTemporary = tblDesc.isTemporary();
                    destTableIsMaterialization = tblDesc.isMaterialization();
                } else if (viewDesc != null) {
                    field_schemas = new ArrayList<FieldSchema>();
                    destTableIsTemporary = false;
                }
                boolean first = true;
                for (ColumnInfo colInfo : colInfos) {
                    String[] nm = inputRR.reverseLookup(colInfo.getInternalName());
                    if (nm[1] != null) {
                        // non-null column alias
                        colInfo.setAlias(nm[1]);
                    }
                    //default column name
                    String colName = colInfo.getInternalName();
                    if (field_schemas != null) {
                        FieldSchema col = new FieldSchema();
                        if (!("".equals(nm[0])) && nm[1] != null) {
                            // remove ``
                            colName = unescapeIdentifier(colInfo.getAlias()).toLowerCase();
                        }
                        colName = fixCtasColumnName(colName);
                        col.setName(colName);
                        String typeName = colInfo.getType().getTypeName();
                        // CTAS should NOT create a VOID type
                        if (typeName.equals(serdeConstants.VOID_TYPE_NAME)) {
                            throw new SemanticException(ErrorMsg.CTAS_CREATES_VOID_TYPE.getMsg(colName));
                        }
                        col.setType(typeName);
                        field_schemas.add(col);
                    }
                    if (!first) {
                        cols = cols.concat(",");
                        colTypes = colTypes.concat(":");
                    }
                    first = false;
                    cols = cols.concat(colName);
                    // Replace VOID type with string when the output is a temp table or
                    // local files.
                    // A VOID type can be generated under the query:
                    //
                    // select NULL from tt;
                    // or
                    // insert overwrite local directory "abc" select NULL from tt;
                    //
                    // where there is no column type to which the NULL value should be
                    // converted.
                    //
                    String tName = colInfo.getType().getTypeName();
                    if (tName.equals(serdeConstants.VOID_TYPE_NAME)) {
                        colTypes = colTypes.concat(serdeConstants.STRING_TYPE_NAME);
                    } else {
                        colTypes = colTypes.concat(tName);
                    }
                }
                // 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));
                }
                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);
                loadFileWork.add(new LoadFileDesc(tblDesc, viewDesc, queryTmpdir, dest_path, isDfsDir, cols, colTypes));
                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);
                            table_desc = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, fileFormat, LazySimpleSerDe.class);
                        }
                    } 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);
    }
    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 &= !destTableIsAcid;
    // 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 = new FileSinkDesc(queryTmpdir, table_desc, conf.getBoolVar(HiveConf.ConfVars.COMPRESSRESULT), currentTableId, rsCtx.isMultiFileSpray(), canBeMerged, rsCtx.getNumFiles(), rsCtx.getTotalFiles(), rsCtx.getPartnCols(), dpCtx, dest_path);
    boolean isHiveServerQuery = SessionState.get().isHiveServerQuery();
    fileSinkDesc.setHiveServerQuery(isHiveServerQuery);
    // FileSinkOperator knows how to properly write to it.
    if (destTableIsAcid) {
        AcidUtils.Operation wt = updating(dest) ? AcidUtils.Operation.UPDATE : (deleting(dest) ? AcidUtils.Operation.DELETE : AcidUtils.Operation.INSERT);
        fileSinkDesc.setWriteType(wt);
        acidFileSinks.add(fileSinkDesc);
    }
    fileSinkDesc.setTemporary(destTableIsTemporary);
    fileSinkDesc.setMaterialization(destTableIsMaterialization);
    /* Set List Bucketing context. */
    if (lbCtx != null) {
        lbCtx.processRowSkewedIndex(fsRS);
        lbCtx.calculateSkewedValueSubDirList();
    }
    fileSinkDesc.setLbCtx(lbCtx);
    // set the stats publishing/aggregating key prefix
    // the same as directory name. The directory name
    // can be changed in the optimizer but the key should not be changed
    // it should be the same as the MoveWork's sourceDir.
    fileSinkDesc.setStatsAggPrefix(fileSinkDesc.getDirName().toString());
    if (!destTableIsMaterialization && HiveConf.getVar(conf, HIVESTATSDBCLASS).equalsIgnoreCase(StatDB.fs.name())) {
        String statsTmpLoc = ctx.getTempDirForPath(dest_path).toString();
        fileSinkDesc.setStatsTmpDir(statsTmpLoc);
        LOG.debug("Set stats collection dir : " + statsTmpLoc);
    }
    if (dest_part != null) {
        try {
            String staticSpec = Warehouse.makePartPath(dest_part.getSpec());
            fileSinkDesc.setStaticSpec(staticSpec);
        } catch (MetaException e) {
            throw new SemanticException(e);
        }
    } else if (dpCtx != null) {
        fileSinkDesc.setStaticSpec(dpCtx.getSPPath());
    }
    if (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);
    if (ltd != null && SessionState.get() != null) {
        SessionState.get().getLineageState().mapDirToOp(ltd.getSourcePath(), (FileSinkOperator) output);
    } else if (queryState.getCommandType().equals(HiveOperation.CREATETABLE_AS_SELECT.getOperationName())) {
        Path tlocation = null;
        String tName = Utilities.getDbTableName(tableDesc.getTableName())[1];
        try {
            Warehouse wh = new Warehouse(conf);
            tlocation = wh.getTablePath(db.getDatabase(tableDesc.getDatabaseName()), tName);
        } catch (MetaException | HiveException e) {
            throw new SemanticException(e);
        }
        SessionState.get().getLineageState().mapDirToOp(tlocation, (FileSinkOperator) 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);
    fsopToTable.put(fso, dest_tab);
    // and it is an insert overwrite or insert into table
    if (dest_tab != null && 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;
}
Also used : AbstractMapJoinOperator(org.apache.hadoop.hive.ql.exec.AbstractMapJoinOperator) SelectOperator(org.apache.hadoop.hive.ql.exec.SelectOperator) JoinOperator(org.apache.hadoop.hive.ql.exec.JoinOperator) Operator(org.apache.hadoop.hive.ql.exec.Operator) GroupByOperator(org.apache.hadoop.hive.ql.exec.GroupByOperator) FileSinkOperator(org.apache.hadoop.hive.ql.exec.FileSinkOperator) FilterOperator(org.apache.hadoop.hive.ql.exec.FilterOperator) ReduceSinkOperator(org.apache.hadoop.hive.ql.exec.ReduceSinkOperator) TableScanOperator(org.apache.hadoop.hive.ql.exec.TableScanOperator) UnionOperator(org.apache.hadoop.hive.ql.exec.UnionOperator) SMBMapJoinOperator(org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator) Warehouse(org.apache.hadoop.hive.metastore.Warehouse) HiveException(org.apache.hadoop.hive.ql.metadata.HiveException) FileSinkDesc(org.apache.hadoop.hive.ql.plan.FileSinkDesc) LazySimpleSerDe(org.apache.hadoop.hive.serde2.lazy.LazySimpleSerDe) FieldSchema(org.apache.hadoop.hive.metastore.api.FieldSchema) DynamicPartitionCtx(org.apache.hadoop.hive.ql.plan.DynamicPartitionCtx) ArrayList(java.util.ArrayList) ColumnInfo(org.apache.hadoop.hive.ql.exec.ColumnInfo) Operation(org.apache.hadoop.hive.ql.io.AcidUtils.Operation) HiveOperation(org.apache.hadoop.hive.ql.plan.HiveOperation) CreateViewDesc(org.apache.hadoop.hive.ql.plan.CreateViewDesc) ThriftJDBCBinarySerDe(org.apache.hadoop.hive.serde2.thrift.ThriftJDBCBinarySerDe) FileSystem(org.apache.hadoop.fs.FileSystem) NoOpFetchFormatter(org.apache.hadoop.hive.serde2.NoOpFetchFormatter) ListBucketingCtx(org.apache.hadoop.hive.ql.plan.ListBucketingCtx) LinkedList(java.util.LinkedList) ArrayList(java.util.ArrayList) List(java.util.List) WriteEntity(org.apache.hadoop.hive.ql.hooks.WriteEntity) CalciteSemanticException(org.apache.hadoop.hive.ql.optimizer.calcite.CalciteSemanticException) MetaException(org.apache.hadoop.hive.metastore.api.MetaException) Path(org.apache.hadoop.fs.Path) Partition(org.apache.hadoop.hive.ql.metadata.Partition) DummyPartition(org.apache.hadoop.hive.ql.metadata.DummyPartition) LoadFileDesc(org.apache.hadoop.hive.ql.plan.LoadFileDesc) RowSchema(org.apache.hadoop.hive.ql.exec.RowSchema) Table(org.apache.hadoop.hive.ql.metadata.Table) FileSinkOperator(org.apache.hadoop.hive.ql.exec.FileSinkOperator) IOException(java.io.IOException) IOException(java.io.IOException) CalciteSemanticException(org.apache.hadoop.hive.ql.optimizer.calcite.CalciteSemanticException) MetaException(org.apache.hadoop.hive.metastore.api.MetaException) HiveException(org.apache.hadoop.hive.ql.metadata.HiveException) SerDeException(org.apache.hadoop.hive.serde2.SerDeException) PatternSyntaxException(java.util.regex.PatternSyntaxException) FileNotFoundException(java.io.FileNotFoundException) AccessControlException(java.security.AccessControlException) InvalidTableException(org.apache.hadoop.hive.ql.metadata.InvalidTableException) LoadTableDesc(org.apache.hadoop.hive.ql.plan.LoadTableDesc) DummyPartition(org.apache.hadoop.hive.ql.metadata.DummyPartition) CreateTableDesc(org.apache.hadoop.hive.ql.plan.CreateTableDesc) Operation(org.apache.hadoop.hive.ql.io.AcidUtils.Operation) CreateTableDesc(org.apache.hadoop.hive.ql.plan.CreateTableDesc) InsertTableDesc(org.apache.hadoop.hive.ql.plan.InsertTableDesc) LoadTableDesc(org.apache.hadoop.hive.ql.plan.LoadTableDesc) AlterTableDesc(org.apache.hadoop.hive.ql.plan.AlterTableDesc) TableDesc(org.apache.hadoop.hive.ql.plan.TableDesc) StandardStructObjectInspector(org.apache.hadoop.hive.serde2.objectinspector.StandardStructObjectInspector) StructObjectInspector(org.apache.hadoop.hive.serde2.objectinspector.StructObjectInspector) AcidUtils(org.apache.hadoop.hive.ql.io.AcidUtils)

Example 50 with Operator

use of org.apache.hadoop.hive.ql.exec.Operator 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;
}
Also used : AbstractMapJoinOperator(org.apache.hadoop.hive.ql.exec.AbstractMapJoinOperator) SelectOperator(org.apache.hadoop.hive.ql.exec.SelectOperator) JoinOperator(org.apache.hadoop.hive.ql.exec.JoinOperator) Operator(org.apache.hadoop.hive.ql.exec.Operator) GroupByOperator(org.apache.hadoop.hive.ql.exec.GroupByOperator) FileSinkOperator(org.apache.hadoop.hive.ql.exec.FileSinkOperator) FilterOperator(org.apache.hadoop.hive.ql.exec.FilterOperator) ReduceSinkOperator(org.apache.hadoop.hive.ql.exec.ReduceSinkOperator) TableScanOperator(org.apache.hadoop.hive.ql.exec.TableScanOperator) UnionOperator(org.apache.hadoop.hive.ql.exec.UnionOperator) SMBMapJoinOperator(org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator) FilterDesc(org.apache.hadoop.hive.ql.plan.FilterDesc) ExprNodeConstantDesc(org.apache.hadoop.hive.ql.plan.ExprNodeConstantDesc) RowSchema(org.apache.hadoop.hive.ql.exec.RowSchema) ExprNodeDesc(org.apache.hadoop.hive.ql.plan.ExprNodeDesc)

Aggregations

Operator (org.apache.hadoop.hive.ql.exec.Operator)130 ReduceSinkOperator (org.apache.hadoop.hive.ql.exec.ReduceSinkOperator)98 TableScanOperator (org.apache.hadoop.hive.ql.exec.TableScanOperator)91 JoinOperator (org.apache.hadoop.hive.ql.exec.JoinOperator)77 ArrayList (java.util.ArrayList)76 FileSinkOperator (org.apache.hadoop.hive.ql.exec.FileSinkOperator)75 OperatorDesc (org.apache.hadoop.hive.ql.plan.OperatorDesc)65 UnionOperator (org.apache.hadoop.hive.ql.exec.UnionOperator)62 SMBMapJoinOperator (org.apache.hadoop.hive.ql.exec.SMBMapJoinOperator)61 GroupByOperator (org.apache.hadoop.hive.ql.exec.GroupByOperator)57 SelectOperator (org.apache.hadoop.hive.ql.exec.SelectOperator)56 FilterOperator (org.apache.hadoop.hive.ql.exec.FilterOperator)54 AbstractMapJoinOperator (org.apache.hadoop.hive.ql.exec.AbstractMapJoinOperator)45 MapJoinOperator (org.apache.hadoop.hive.ql.exec.MapJoinOperator)40 ExprNodeDesc (org.apache.hadoop.hive.ql.plan.ExprNodeDesc)39 HashMap (java.util.HashMap)36 RowSchema (org.apache.hadoop.hive.ql.exec.RowSchema)36 LinkedHashMap (java.util.LinkedHashMap)35 ColumnInfo (org.apache.hadoop.hive.ql.exec.ColumnInfo)28 List (java.util.List)22