Examples with TableDesc org.apache.hadoop.hive.ql.plan.TableDesc used on opensource projects

Example 11 with TableDesc

use of org.apache.hadoop.hive.ql.plan.TableDesc in project hive by apache.

the class VectorReduceSinkCommonOperator method initializeOp.

@Override
protected void initializeOp(Configuration hconf) throws HiveException {
    super.initializeOp(hconf);
    if (LOG.isDebugEnabled()) {
        // Determine the name of our map or reduce task for debug tracing.
        BaseWork work = Utilities.getMapWork(hconf);
        if (work == null) {
            work = Utilities.getReduceWork(hconf);
        }
        taskName = work.getName();
    }
    String context = hconf.get(Operator.CONTEXT_NAME_KEY, "");
    if (context != null && !context.isEmpty()) {
        context = "_" + context.replace(" ", "_");
    }
    statsMap.put(Counter.RECORDS_OUT_INTERMEDIATE + context, recordCounter);
    reduceSkipTag = conf.getSkipTag();
    reduceTagByte = (byte) conf.getTag();
    if (isLogInfoEnabled) {
        LOG.info("Using tag = " + (int) reduceTagByte);
    }
    TableDesc keyTableDesc = conf.getKeySerializeInfo();
    boolean[] columnSortOrder = getColumnSortOrder(keyTableDesc.getProperties(), reduceSinkKeyColumnMap.length);
    byte[] columnNullMarker = getColumnNullMarker(keyTableDesc.getProperties(), reduceSinkKeyColumnMap.length, columnSortOrder);
    byte[] columnNotNullMarker = getColumnNotNullMarker(keyTableDesc.getProperties(), reduceSinkKeyColumnMap.length, columnSortOrder);
    keyBinarySortableSerializeWrite = new BinarySortableSerializeWrite(columnSortOrder, columnNullMarker, columnNotNullMarker);
    // Create all nulls key.
    try {
        Output nullKeyOutput = new Output();
        keyBinarySortableSerializeWrite.set(nullKeyOutput);
        for (int i = 0; i < reduceSinkKeyColumnMap.length; i++) {
            keyBinarySortableSerializeWrite.writeNull();
        }
        int nullBytesLength = nullKeyOutput.getLength();
        nullBytes = new byte[nullBytesLength];
        System.arraycopy(nullKeyOutput.getData(), 0, nullBytes, 0, nullBytesLength);
        nullKeyHashCode = HashCodeUtil.calculateBytesHashCode(nullBytes, 0, nullBytesLength);
    } catch (Exception e) {
        throw new HiveException(e);
    }
    valueLazyBinarySerializeWrite = new LazyBinarySerializeWrite(reduceSinkValueColumnMap.length);
    valueVectorSerializeRow = new VectorSerializeRow<LazyBinarySerializeWrite>(valueLazyBinarySerializeWrite);
    valueVectorSerializeRow.init(reduceSinkValueTypeInfos, reduceSinkValueColumnMap);
    valueOutput = new Output();
    valueVectorSerializeRow.setOutput(valueOutput);
    keyWritable = new HiveKey();
    valueBytesWritable = new BytesWritable();
    batchCounter = 0;
}

Example 12 with TableDesc

use of org.apache.hadoop.hive.ql.plan.TableDesc in project hive by apache.

the class GenMapRedUtils method setMapWork.

/**
   * initialize MapWork
   *
   * @param alias_id
   *          current alias
   * @param topOp
   *          the top operator of the stack
   * @param plan
   *          map work to initialize
   * @param local
   *          whether you need to add to map-reduce or local work
   * @param pList
   *          pruned partition list. If it is null it will be computed on-the-fly.
   * @param inputs
   *          read entities for the map work
   * @param conf
   *          current instance of hive conf
   */
public static void setMapWork(MapWork plan, ParseContext parseCtx, Set<ReadEntity> inputs, PrunedPartitionList partsList, TableScanOperator tsOp, String alias_id, HiveConf conf, boolean local) throws SemanticException {
    ArrayList<Path> partDir = new ArrayList<Path>();
    ArrayList<PartitionDesc> partDesc = new ArrayList<PartitionDesc>();
    boolean isAcidTable = false;
    Path tblDir = null;
    plan.setNameToSplitSample(parseCtx.getNameToSplitSample());
    if (partsList == null) {
        try {
            partsList = PartitionPruner.prune(tsOp, parseCtx, alias_id);
            isAcidTable = tsOp.getConf().isAcidTable();
        } catch (SemanticException e) {
            throw e;
        }
    }
    // Generate the map work for this alias_id
    // pass both confirmed and unknown partitions through the map-reduce
    // framework
    Set<Partition> parts = partsList.getPartitions();
    PartitionDesc aliasPartnDesc = null;
    try {
        if (!parts.isEmpty()) {
            aliasPartnDesc = Utilities.getPartitionDesc(parts.iterator().next());
        }
    } catch (HiveException e) {
        LOG.error(org.apache.hadoop.util.StringUtils.stringifyException(e));
        throw new SemanticException(e.getMessage(), e);
    }
    // The table does not have any partitions
    if (aliasPartnDesc == null) {
        aliasPartnDesc = new PartitionDesc(Utilities.getTableDesc(tsOp.getConf().getTableMetadata()), null);
    }
    Map<String, String> props = tsOp.getConf().getOpProps();
    if (props != null) {
        Properties target = aliasPartnDesc.getProperties();
        target.putAll(props);
    }
    plan.getAliasToPartnInfo().put(alias_id, aliasPartnDesc);
    long sizeNeeded = Integer.MAX_VALUE;
    int fileLimit = -1;
    if (parseCtx.getGlobalLimitCtx().isEnable()) {
        if (isAcidTable) {
            LOG.info("Skip Global Limit optimization for ACID table");
            parseCtx.getGlobalLimitCtx().disableOpt();
        } else {
            long sizePerRow = HiveConf.getLongVar(parseCtx.getConf(), HiveConf.ConfVars.HIVELIMITMAXROWSIZE);
            sizeNeeded = (parseCtx.getGlobalLimitCtx().getGlobalOffset() + parseCtx.getGlobalLimitCtx().getGlobalLimit()) * sizePerRow;
            // for the optimization that reduce number of input file, we limit number
            // of files allowed. If more than specific number of files have to be
            // selected, we skip this optimization. Since having too many files as
            // inputs can cause unpredictable latency. It's not necessarily to be
            // cheaper.
            fileLimit = HiveConf.getIntVar(parseCtx.getConf(), HiveConf.ConfVars.HIVELIMITOPTLIMITFILE);
            if (sizePerRow <= 0 || fileLimit <= 0) {
                LOG.info("Skip optimization to reduce input size of 'limit'");
                parseCtx.getGlobalLimitCtx().disableOpt();
            } else if (parts.isEmpty()) {
                LOG.info("Empty input: skip limit optimization");
            } else {
                LOG.info("Try to reduce input size for 'limit' " + "sizeNeeded: " + sizeNeeded + "  file limit : " + fileLimit);
            }
        }
    }
    boolean isFirstPart = true;
    boolean emptyInput = true;
    boolean singlePartition = (parts.size() == 1);
    // Track the dependencies for the view. Consider a query like: select * from V;
    // where V is a view of the form: select * from T
    // The dependencies should include V at depth 0, and T at depth 1 (inferred).
    Map<String, ReadEntity> viewToInput = parseCtx.getViewAliasToInput();
    ReadEntity parentViewInfo = PlanUtils.getParentViewInfo(alias_id, viewToInput);
    // The table should also be considered a part of inputs, even if the table is a
    // partitioned table and whether any partition is selected or not
    //This read entity is a direct read entity and not an indirect read (that is when
    // this is being read because it is a dependency of a view).
    boolean isDirectRead = (parentViewInfo == null);
    TableDesc tblDesc = null;
    boolean initTableDesc = false;
    PlanUtils.addPartitionInputs(parts, inputs, parentViewInfo, isDirectRead);
    for (Partition part : parts) {
        // Later the properties have to come from the partition as opposed
        // to from the table in order to support versioning.
        Path[] paths = null;
        SampleDesc sampleDescr = parseCtx.getOpToSamplePruner().get(tsOp);
        // Lookup list bucketing pruner
        Map<String, ExprNodeDesc> partToPruner = parseCtx.getOpToPartToSkewedPruner().get(tsOp);
        ExprNodeDesc listBucketingPruner = (partToPruner != null) ? partToPruner.get(part.getName()) : null;
        if (sampleDescr != null) {
            assert (listBucketingPruner == null) : "Sampling and list bucketing can't coexit.";
            paths = SamplePruner.prune(part, sampleDescr);
            parseCtx.getGlobalLimitCtx().disableOpt();
        } else if (listBucketingPruner != null) {
            assert (sampleDescr == null) : "Sampling and list bucketing can't coexist.";
            /* Use list bucketing prunner's path. */
            paths = ListBucketingPruner.prune(parseCtx, part, listBucketingPruner);
        } else {
            // contain enough size, we change to normal mode.
            if (parseCtx.getGlobalLimitCtx().isEnable()) {
                if (isFirstPart) {
                    long sizeLeft = sizeNeeded;
                    ArrayList<Path> retPathList = new ArrayList<Path>();
                    SamplePruner.LimitPruneRetStatus status = SamplePruner.limitPrune(part, sizeLeft, fileLimit, retPathList);
                    if (status.equals(SamplePruner.LimitPruneRetStatus.NoFile)) {
                        continue;
                    } else if (status.equals(SamplePruner.LimitPruneRetStatus.NotQualify)) {
                        LOG.info("Use full input -- first " + fileLimit + " files are more than " + sizeNeeded + " bytes");
                        parseCtx.getGlobalLimitCtx().disableOpt();
                    } else {
                        emptyInput = false;
                        paths = new Path[retPathList.size()];
                        int index = 0;
                        for (Path path : retPathList) {
                            paths[index++] = path;
                        }
                        if (status.equals(SamplePruner.LimitPruneRetStatus.NeedAllFiles) && singlePartition) {
                            // if all files are needed to meet the size limit, we disable
                            // optimization. It usually happens for empty table/partition or
                            // table/partition with only one file. By disabling this
                            // optimization, we can avoid retrying the query if there is
                            // not sufficient rows.
                            parseCtx.getGlobalLimitCtx().disableOpt();
                        }
                    }
                    isFirstPart = false;
                } else {
                    paths = new Path[0];
                }
            }
            if (!parseCtx.getGlobalLimitCtx().isEnable()) {
                paths = part.getPath();
            }
        }
        // is it a partitioned table ?
        if (!part.getTable().isPartitioned()) {
            assert (tblDir == null);
            tblDir = paths[0];
            if (!initTableDesc) {
                tblDesc = Utilities.getTableDesc(part.getTable());
                initTableDesc = true;
            }
        } else if (tblDesc == null) {
            if (!initTableDesc) {
                tblDesc = Utilities.getTableDesc(part.getTable());
                initTableDesc = true;
            }
        }
        if (props != null) {
            Properties target = tblDesc.getProperties();
            target.putAll(props);
        }
        for (Path p : paths) {
            if (p == null) {
                continue;
            }
            String path = p.toString();
            if (LOG.isDebugEnabled()) {
                LOG.debug("Adding " + path + " of table" + alias_id);
            }
            partDir.add(p);
            try {
                if (part.getTable().isPartitioned()) {
                    partDesc.add(Utilities.getPartitionDesc(part));
                } else {
                    partDesc.add(Utilities.getPartitionDescFromTableDesc(tblDesc, part, false));
                }
            } catch (HiveException e) {
                LOG.error(org.apache.hadoop.util.StringUtils.stringifyException(e));
                throw new SemanticException(e.getMessage(), e);
            }
        }
    }
    if (emptyInput) {
        parseCtx.getGlobalLimitCtx().disableOpt();
    }
    Utilities.addSchemaEvolutionToTableScanOperator(partsList.getSourceTable(), tsOp);
    Iterator<Path> iterPath = partDir.iterator();
    Iterator<PartitionDesc> iterPartnDesc = partDesc.iterator();
    if (!local) {
        while (iterPath.hasNext()) {
            assert iterPartnDesc.hasNext();
            Path path = iterPath.next();
            PartitionDesc prtDesc = iterPartnDesc.next();
            // Add the path to alias mapping
            plan.addPathToAlias(path, alias_id);
            plan.addPathToPartitionInfo(path, prtDesc);
            if (LOG.isDebugEnabled()) {
                LOG.debug("Information added for path " + path);
            }
        }
        assert plan.getAliasToWork().get(alias_id) == null;
        plan.getAliasToWork().put(alias_id, tsOp);
    } else {
        // populate local work if needed
        MapredLocalWork localPlan = plan.getMapRedLocalWork();
        if (localPlan == null) {
            localPlan = new MapredLocalWork(new LinkedHashMap<String, Operator<? extends OperatorDesc>>(), new LinkedHashMap<String, FetchWork>());
        }
        assert localPlan.getAliasToWork().get(alias_id) == null;
        assert localPlan.getAliasToFetchWork().get(alias_id) == null;
        localPlan.getAliasToWork().put(alias_id, tsOp);
        if (tblDir == null) {
            tblDesc = Utilities.getTableDesc(partsList.getSourceTable());
            localPlan.getAliasToFetchWork().put(alias_id, new FetchWork(partDir, partDesc, tblDesc));
        } else {
            localPlan.getAliasToFetchWork().put(alias_id, new FetchWork(tblDir, tblDesc));
        }
        plan.setMapRedLocalWork(localPlan);
    }
}

Example 13 with TableDesc

use of org.apache.hadoop.hive.ql.plan.TableDesc in project hive by apache.

the class GenMapRedUtils method splitTasks.

@SuppressWarnings("nls")
private static /**
   * Split two tasks by creating a temporary file between them.
   *
   * @param op reduce sink operator being processed
   * @param parentTask the parent task
   * @param childTask the child task
   * @param opProcCtx context
   **/
void splitTasks(ReduceSinkOperator op, Task<? extends Serializable> parentTask, Task<? extends Serializable> childTask, GenMRProcContext opProcCtx) throws SemanticException {
    if (op.getNumParent() != 1) {
        throw new IllegalStateException("Expecting operator " + op + " to have one parent. " + "But found multiple parents : " + op.getParentOperators());
    }
    ParseContext parseCtx = opProcCtx.getParseCtx();
    parentTask.addDependentTask(childTask);
    // Root Task cannot depend on any other task, therefore childTask cannot be
    // a root Task
    List<Task<? extends Serializable>> rootTasks = opProcCtx.getRootTasks();
    if (rootTasks.contains(childTask)) {
        rootTasks.remove(childTask);
    }
    // Generate the temporary file name
    Context baseCtx = parseCtx.getContext();
    Path taskTmpDir = baseCtx.getMRTmpPath();
    Operator<? extends OperatorDesc> parent = op.getParentOperators().get(0);
    TableDesc tt_desc = PlanUtils.getIntermediateFileTableDesc(PlanUtils.getFieldSchemasFromRowSchema(parent.getSchema(), "temporarycol"));
    // Create the temporary file, its corresponding FileSinkOperaotr, and
    // its corresponding TableScanOperator.
    TableScanOperator tableScanOp = createTemporaryFile(parent, op, taskTmpDir, tt_desc, parseCtx);
    Map<Operator<? extends OperatorDesc>, GenMapRedCtx> mapCurrCtx = opProcCtx.getMapCurrCtx();
    mapCurrCtx.put(tableScanOp, new GenMapRedCtx(childTask, null));
    String streamDesc = taskTmpDir.toUri().toString();
    MapredWork cplan = (MapredWork) childTask.getWork();
    if (needsTagging(cplan.getReduceWork())) {
        Operator<? extends OperatorDesc> reducerOp = cplan.getReduceWork().getReducer();
        String id = null;
        if (reducerOp instanceof JoinOperator) {
            if (parseCtx.getJoinOps().contains(reducerOp)) {
                id = ((JoinOperator) reducerOp).getConf().getId();
            }
        } else if (reducerOp instanceof MapJoinOperator) {
            if (parseCtx.getMapJoinOps().contains(reducerOp)) {
                id = ((MapJoinOperator) reducerOp).getConf().getId();
            }
        } else if (reducerOp instanceof SMBMapJoinOperator) {
            if (parseCtx.getSmbMapJoinOps().contains(reducerOp)) {
                id = ((SMBMapJoinOperator) reducerOp).getConf().getId();
            }
        }
        if (id != null) {
            streamDesc = id + ":$INTNAME";
        } else {
            streamDesc = "$INTNAME";
        }
        String origStreamDesc = streamDesc;
        int pos = 0;
        while (cplan.getMapWork().getAliasToWork().get(streamDesc) != null) {
            streamDesc = origStreamDesc.concat(String.valueOf(++pos));
        }
        // TODO: Allocate work to remove the temporary files and make that
        // dependent on the redTask
        cplan.getReduceWork().setNeedsTagging(true);
    }
    // Add the path to alias mapping
    setTaskPlan(taskTmpDir, streamDesc, tableScanOp, cplan.getMapWork(), false, tt_desc);
    opProcCtx.setCurrTopOp(null);
    opProcCtx.setCurrAliasId(null);
    opProcCtx.setCurrTask(childTask);
    opProcCtx.addRootIfPossible(parentTask);
}

Example 14 with TableDesc

use of org.apache.hadoop.hive.ql.plan.TableDesc in project hive by apache.

the class GenMapRedUtils method createMergeTask.

/**
   * Create a block level merge task for RCFiles or stripe level merge task for
   * ORCFiles
   *
   * @param fsInputDesc
   * @param finalName
   * @param ctx
   * @param inputFormatClass
   * @return MergeWork if table is stored as RCFile or ORCFile,
   *         null otherwise
   */
public static MapWork createMergeTask(FileSinkDesc fsInputDesc, Path finalName, boolean hasDynamicPartitions, CompilationOpContext ctx) throws SemanticException {
    Path inputDir = fsInputDesc.getFinalDirName();
    TableDesc tblDesc = fsInputDesc.getTableInfo();
    List<Path> inputDirs = new ArrayList<Path>(1);
    ArrayList<String> inputDirstr = new ArrayList<String>(1);
    // in case of dynamic partitioning and list bucketing
    if (!hasDynamicPartitions && !GenMapRedUtils.isSkewedStoredAsDirs(fsInputDesc)) {
        inputDirs.add(inputDir);
    }
    inputDirstr.add(inputDir.toString());
    // internal input format class for CombineHiveInputFormat
    final Class<? extends InputFormat> internalIFClass;
    if (tblDesc.getInputFileFormatClass().equals(RCFileInputFormat.class)) {
        internalIFClass = RCFileBlockMergeInputFormat.class;
    } else if (tblDesc.getInputFileFormatClass().equals(OrcInputFormat.class)) {
        internalIFClass = OrcFileStripeMergeInputFormat.class;
    } else {
        throw new SemanticException("createMergeTask called on a table with file" + " format other than RCFile or ORCFile");
    }
    // create the merge file work
    MergeFileWork work = new MergeFileWork(inputDirs, finalName, hasDynamicPartitions, tblDesc.getInputFileFormatClass().getName());
    LinkedHashMap<Path, ArrayList<String>> pathToAliases = new LinkedHashMap<>();
    pathToAliases.put(inputDir, inputDirstr);
    work.setMapperCannotSpanPartns(true);
    work.setPathToAliases(pathToAliases);
    PartitionDesc pDesc = new PartitionDesc(tblDesc, null);
    pDesc.setInputFileFormatClass(internalIFClass);
    work.addPathToPartitionInfo(inputDir, pDesc);
    work.setListBucketingCtx(fsInputDesc.getLbCtx());
    // create alias to work which contains the merge operator
    LinkedHashMap<String, Operator<? extends OperatorDesc>> aliasToWork = new LinkedHashMap<String, Operator<? extends OperatorDesc>>();
    Operator<? extends OperatorDesc> mergeOp = null;
    final FileMergeDesc fmd;
    if (tblDesc.getInputFileFormatClass().equals(RCFileInputFormat.class)) {
        fmd = new RCFileMergeDesc();
    } else {
        fmd = new OrcFileMergeDesc();
    }
    fmd.setDpCtx(fsInputDesc.getDynPartCtx());
    fmd.setOutputPath(finalName);
    fmd.setHasDynamicPartitions(work.hasDynamicPartitions());
    fmd.setListBucketingAlterTableConcatenate(work.isListBucketingAlterTableConcatenate());
    int lbLevel = work.getListBucketingCtx() == null ? 0 : work.getListBucketingCtx().calculateListBucketingLevel();
    fmd.setListBucketingDepth(lbLevel);
    mergeOp = OperatorFactory.get(ctx, fmd);
    aliasToWork.put(inputDir.toString(), mergeOp);
    work.setAliasToWork(aliasToWork);
    return work;
}

Example 15 with TableDesc

use of org.apache.hadoop.hive.ql.plan.TableDesc in project hive by apache.

the class MapJoinProcessor method convertSMBJoinToMapJoin.

/**
   * convert a sortmerge join to a a map-side join.
   *
   * @param opParseCtxMap
   * @param smbJoinOp
   *          join operator
   * @param joinTree
   *          qb join tree
   * @param bigTablePos
   *          position of the source to be read as part of map-reduce framework. All other sources
   *          are cached in memory
   * @param noCheckOuterJoin
   */
public static MapJoinOperator convertSMBJoinToMapJoin(HiveConf hconf, SMBMapJoinOperator smbJoinOp, int bigTablePos, boolean noCheckOuterJoin) throws SemanticException {
    // Create a new map join operator
    SMBJoinDesc smbJoinDesc = smbJoinOp.getConf();
    List<ExprNodeDesc> keyCols = smbJoinDesc.getKeys().get(Byte.valueOf((byte) 0));
    TableDesc keyTableDesc = PlanUtils.getMapJoinKeyTableDesc(hconf, PlanUtils.getFieldSchemasFromColumnList(keyCols, MAPJOINKEY_FIELDPREFIX));
    MapJoinDesc mapJoinDesc = new MapJoinDesc(smbJoinDesc.getKeys(), keyTableDesc, smbJoinDesc.getExprs(), smbJoinDesc.getValueTblDescs(), smbJoinDesc.getValueTblDescs(), smbJoinDesc.getOutputColumnNames(), bigTablePos, smbJoinDesc.getConds(), smbJoinDesc.getFilters(), smbJoinDesc.isNoOuterJoin(), smbJoinDesc.getDumpFilePrefix());
    mapJoinDesc.setStatistics(smbJoinDesc.getStatistics());
    RowSchema joinRS = smbJoinOp.getSchema();
    // The mapjoin has the same schema as the join operator
    MapJoinOperator mapJoinOp = (MapJoinOperator) OperatorFactory.getAndMakeChild(smbJoinOp.getCompilationOpContext(), mapJoinDesc, joinRS, new ArrayList<Operator<? extends OperatorDesc>>());
    // change the children of the original join operator to point to the map
    // join operator
    List<Operator<? extends OperatorDesc>> childOps = smbJoinOp.getChildOperators();
    for (Operator<? extends OperatorDesc> childOp : childOps) {
        childOp.replaceParent(smbJoinOp, mapJoinOp);
    }
    mapJoinOp.setChildOperators(childOps);
    smbJoinOp.setChildOperators(null);
    // change the parent of the original SMBjoin operator to point to the map
    // join operator
    List<Operator<? extends OperatorDesc>> parentOps = smbJoinOp.getParentOperators();
    for (Operator<? extends OperatorDesc> parentOp : parentOps) {
        parentOp.replaceChild(smbJoinOp, mapJoinOp);
    }
    mapJoinOp.setParentOperators(parentOps);
    smbJoinOp.setParentOperators(null);
    return mapJoinOp;
}