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

Example 1 with FileSinkOperator

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

the class GenMapRedUtils method createMRWorkForMergingFiles.

/**
   * @param fsInput The FileSink operator.
   * @param ctx The MR processing context.
   * @param finalName the final destination path the merge job should output.
   * @param dependencyTask
   * @param mvTasks
   * @param conf
   * @param currTask
   * @throws SemanticException

   * create a Map-only merge job using CombineHiveInputFormat for all partitions with
   * following operators:
   *          MR job J0:
   *          ...
   *          |
   *          v
   *          FileSinkOperator_1 (fsInput)
   *          |
   *          v
   *          Merge job J1:
   *          |
   *          v
   *          TableScan (using CombineHiveInputFormat) (tsMerge)
   *          |
   *          v
   *          FileSinkOperator (fsMerge)
   *
   *          Here the pathToPartitionInfo & pathToAlias will remain the same, which means the paths
   *          do
   *          not contain the dynamic partitions (their parent). So after the dynamic partitions are
   *          created (after the first job finished before the moveTask or ConditionalTask start),
   *          we need to change the pathToPartitionInfo & pathToAlias to include the dynamic
   *          partition
   *          directories.
   *
   */
public static void createMRWorkForMergingFiles(FileSinkOperator fsInput, Path finalName, DependencyCollectionTask dependencyTask, List<Task<MoveWork>> mvTasks, HiveConf conf, Task<? extends Serializable> currTask) throws SemanticException {
    //
    // 1. create the operator tree
    //
    FileSinkDesc fsInputDesc = fsInput.getConf();
    // Create a TableScan operator
    RowSchema inputRS = fsInput.getSchema();
    TableScanOperator tsMerge = GenMapRedUtils.createTemporaryTableScanOperator(fsInput.getCompilationOpContext(), inputRS);
    // Create a FileSink operator
    TableDesc ts = (TableDesc) fsInputDesc.getTableInfo().clone();
    FileSinkDesc fsOutputDesc = new FileSinkDesc(finalName, ts, conf.getBoolVar(ConfVars.COMPRESSRESULT));
    FileSinkOperator fsOutput = (FileSinkOperator) OperatorFactory.getAndMakeChild(fsOutputDesc, inputRS, tsMerge);
    // If the input FileSinkOperator is a dynamic partition enabled, the tsMerge input schema
    // needs to include the partition column, and the fsOutput should have
    // a DynamicPartitionCtx to indicate that it needs to dynamically partitioned.
    DynamicPartitionCtx dpCtx = fsInputDesc.getDynPartCtx();
    if (dpCtx != null && dpCtx.getNumDPCols() > 0) {
        // adding DP ColumnInfo to the RowSchema signature
        ArrayList<ColumnInfo> signature = inputRS.getSignature();
        String tblAlias = fsInputDesc.getTableInfo().getTableName();
        for (String dpCol : dpCtx.getDPColNames()) {
            ColumnInfo colInfo = new ColumnInfo(dpCol, // all partition column type should be string
            TypeInfoFactory.stringTypeInfo, tblAlias, // partition column is virtual column
            true);
            signature.add(colInfo);
        }
        inputRS.setSignature(signature);
        // create another DynamicPartitionCtx, which has a different input-to-DP column mapping
        DynamicPartitionCtx dpCtx2 = new DynamicPartitionCtx(dpCtx);
        fsOutputDesc.setDynPartCtx(dpCtx2);
        // update the FileSinkOperator to include partition columns
        usePartitionColumns(fsInputDesc.getTableInfo().getProperties(), dpCtx.getDPColNames());
    } else {
        // non-partitioned table
        fsInputDesc.getTableInfo().getProperties().remove(org.apache.hadoop.hive.metastore.api.hive_metastoreConstants.META_TABLE_PARTITION_COLUMNS);
    }
    //
    // 2. Constructing a conditional task consisting of a move task and a map reduce task
    //
    MoveWork dummyMv = new MoveWork(null, null, null, new LoadFileDesc(fsInputDesc.getFinalDirName(), finalName, true, null, null), false);
    MapWork cplan;
    Serializable work;
    if ((conf.getBoolVar(ConfVars.HIVEMERGERCFILEBLOCKLEVEL) && fsInputDesc.getTableInfo().getInputFileFormatClass().equals(RCFileInputFormat.class)) || (conf.getBoolVar(ConfVars.HIVEMERGEORCFILESTRIPELEVEL) && fsInputDesc.getTableInfo().getInputFileFormatClass().equals(OrcInputFormat.class))) {
        cplan = GenMapRedUtils.createMergeTask(fsInputDesc, finalName, dpCtx != null && dpCtx.getNumDPCols() > 0, fsInput.getCompilationOpContext());
        if (conf.getVar(ConfVars.HIVE_EXECUTION_ENGINE).equals("tez")) {
            work = new TezWork(conf.getVar(HiveConf.ConfVars.HIVEQUERYID), conf);
            cplan.setName("File Merge");
            ((TezWork) work).add(cplan);
        } else if (conf.getVar(ConfVars.HIVE_EXECUTION_ENGINE).equals("spark")) {
            work = new SparkWork(conf.getVar(HiveConf.ConfVars.HIVEQUERYID));
            cplan.setName("Spark Merge File Work");
            ((SparkWork) work).add(cplan);
        } else {
            work = cplan;
        }
    } else {
        cplan = createMRWorkForMergingFiles(conf, tsMerge, fsInputDesc);
        if (conf.getVar(ConfVars.HIVE_EXECUTION_ENGINE).equals("tez")) {
            work = new TezWork(conf.getVar(HiveConf.ConfVars.HIVEQUERYID), conf);
            cplan.setName("File Merge");
            ((TezWork) work).add(cplan);
        } else if (conf.getVar(ConfVars.HIVE_EXECUTION_ENGINE).equals("spark")) {
            work = new SparkWork(conf.getVar(HiveConf.ConfVars.HIVEQUERYID));
            cplan.setName("Spark Merge File Work");
            ((SparkWork) work).add(cplan);
        } else {
            work = new MapredWork();
            ((MapredWork) work).setMapWork(cplan);
        }
    }
    // use CombineHiveInputFormat for map-only merging
    cplan.setInputformat("org.apache.hadoop.hive.ql.io.CombineHiveInputFormat");
    // NOTE: we should gather stats in MR1 rather than MR2 at merge job since we don't
    // know if merge MR2 will be triggered at execution time
    Task<MoveWork> mvTask = GenMapRedUtils.findMoveTask(mvTasks, fsOutput);
    ConditionalTask cndTsk = GenMapRedUtils.createCondTask(conf, currTask, dummyMv, work, fsInputDesc.getFinalDirName(), finalName, mvTask, dependencyTask);
    // keep the dynamic partition context in conditional task resolver context
    ConditionalResolverMergeFilesCtx mrCtx = (ConditionalResolverMergeFilesCtx) cndTsk.getResolverCtx();
    mrCtx.setDPCtx(fsInputDesc.getDynPartCtx());
    mrCtx.setLbCtx(fsInputDesc.getLbCtx());
}

Example 2 with FileSinkOperator

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

the class GenSparkUtils method getEdgeProperty.

public static SparkEdgeProperty getEdgeProperty(ReduceSinkOperator reduceSink, ReduceWork reduceWork) throws SemanticException {
    SparkEdgeProperty edgeProperty = new SparkEdgeProperty(SparkEdgeProperty.SHUFFLE_NONE);
    edgeProperty.setNumPartitions(reduceWork.getNumReduceTasks());
    String sortOrder = Strings.nullToEmpty(reduceSink.getConf().getOrder()).trim();
    if (hasGBYOperator(reduceSink)) {
        edgeProperty.setShuffleGroup();
        // SHUFFLE_SORT shouldn't be used for this purpose, see HIVE-8542
        if (!sortOrder.isEmpty() && groupByNeedParLevelOrder(reduceSink)) {
            edgeProperty.setMRShuffle();
        }
    }
    if (reduceWork.getReducer() instanceof JoinOperator) {
        //reduce-side join, use MR-style shuffle
        edgeProperty.setMRShuffle();
    }
    //If its a FileSink to bucketed files, also use MR-style shuffle to
    // get compatible taskId for bucket-name
    FileSinkOperator fso = getChildOperator(reduceWork.getReducer(), FileSinkOperator.class);
    if (fso != null) {
        String bucketCount = fso.getConf().getTableInfo().getProperties().getProperty(hive_metastoreConstants.BUCKET_COUNT);
        if (bucketCount != null && Integer.parseInt(bucketCount) > 1) {
            edgeProperty.setMRShuffle();
        }
    }
    // test if we need partition/global order, SHUFFLE_SORT should only be used for global order
    if (edgeProperty.isShuffleNone() && !sortOrder.isEmpty()) {
        if ((reduceSink.getConf().getPartitionCols() == null || reduceSink.getConf().getPartitionCols().isEmpty() || isSame(reduceSink.getConf().getPartitionCols(), reduceSink.getConf().getKeyCols())) && reduceSink.getConf().hasOrderBy()) {
            edgeProperty.setShuffleSort();
        } else {
            edgeProperty.setMRShuffle();
        }
    }
    // simple distribute-by goes here
    if (edgeProperty.isShuffleNone()) {
        edgeProperty.setShuffleGroup();
    }
    return edgeProperty;
}

Example 3 with FileSinkOperator

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

the class DriverContext method finished.

public void finished(TaskRunner runner) {
    if (statsTasks.isEmpty() || !(runner.getTask() instanceof MapRedTask)) {
        return;
    }
    MapRedTask mapredTask = (MapRedTask) runner.getTask();
    MapWork mapWork = mapredTask.getWork().getMapWork();
    ReduceWork reduceWork = mapredTask.getWork().getReduceWork();
    List<Operator> operators = new ArrayList<Operator>(mapWork.getAliasToWork().values());
    if (reduceWork != null) {
        operators.add(reduceWork.getReducer());
    }
    final List<String> statKeys = new ArrayList<String>(1);
    NodeUtils.iterate(operators, FileSinkOperator.class, new Function<FileSinkOperator>() {

        @Override
        public void apply(FileSinkOperator fsOp) {
            if (fsOp.getConf().isGatherStats()) {
                statKeys.add(fsOp.getConf().getStatsAggPrefix());
            }
        }
    });
    for (String statKey : statKeys) {
        statsTasks.get(statKey).getWork().setSourceTask(mapredTask);
    }
}

Example 4 with FileSinkOperator

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

the class TezCompiler method generateTaskTree.

@Override
protected void generateTaskTree(List<Task<? extends Serializable>> rootTasks, ParseContext pCtx, List<Task<MoveWork>> mvTask, Set<ReadEntity> inputs, Set<WriteEntity> outputs) throws SemanticException {
    PerfLogger perfLogger = SessionState.getPerfLogger();
    perfLogger.PerfLogBegin(this.getClass().getName(), PerfLogger.TEZ_COMPILER);
    ParseContext tempParseContext = getParseContext(pCtx, rootTasks);
    GenTezUtils utils = new GenTezUtils();
    GenTezWork genTezWork = new GenTezWork(utils);
    GenTezProcContext procCtx = new GenTezProcContext(conf, tempParseContext, mvTask, rootTasks, inputs, outputs);
    // create a walker which walks the tree in a DFS manner while maintaining
    // the operator stack.
    // The dispatcher generates the plan from the operator tree
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("Split Work - ReduceSink", ReduceSinkOperator.getOperatorName() + "%"), genTezWork);
    opRules.put(new RuleRegExp("No more walking on ReduceSink-MapJoin", MapJoinOperator.getOperatorName() + "%"), new ReduceSinkMapJoinProc());
    opRules.put(new RuleRegExp("Recognize a Sorted Merge Join operator to setup the right edge and" + " stop traversing the DummyStore-MapJoin", CommonMergeJoinOperator.getOperatorName() + "%"), new MergeJoinProc());
    opRules.put(new RuleRegExp("Split Work + Move/Merge - FileSink", FileSinkOperator.getOperatorName() + "%"), new CompositeProcessor(new FileSinkProcessor(), genTezWork));
    opRules.put(new RuleRegExp("Split work - DummyStore", DummyStoreOperator.getOperatorName() + "%"), genTezWork);
    opRules.put(new RuleRegExp("Handle Potential Analyze Command", TableScanOperator.getOperatorName() + "%"), new ProcessAnalyzeTable(utils));
    opRules.put(new RuleRegExp("Remember union", UnionOperator.getOperatorName() + "%"), new UnionProcessor());
    opRules.put(new RuleRegExp("AppMasterEventOperator", AppMasterEventOperator.getOperatorName() + "%"), new AppMasterEventProcessor());
    // The dispatcher fires the processor corresponding to the closest matching
    // rule and passes the context along
    Dispatcher disp = new DefaultRuleDispatcher(null, opRules, procCtx);
    List<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pCtx.getTopOps().values());
    GraphWalker ogw = new GenTezWorkWalker(disp, procCtx);
    ogw.startWalking(topNodes, null);
    // we need to specify the reserved memory for each work that contains Map Join
    for (List<BaseWork> baseWorkList : procCtx.mapJoinWorkMap.values()) {
        for (BaseWork w : baseWorkList) {
            // work should be the smallest unit for memory allocation
            w.setReservedMemoryMB((int) (conf.getLongVar(ConfVars.HIVECONVERTJOINNOCONDITIONALTASKTHRESHOLD) / (1024 * 1024)));
        }
    }
    // we need to clone some operator plans and remove union operators still
    int indexForTezUnion = 0;
    for (BaseWork w : procCtx.workWithUnionOperators) {
        GenTezUtils.removeUnionOperators(procCtx, w, indexForTezUnion++);
    }
    // then we make sure the file sink operators are set up right
    for (FileSinkOperator fileSink : procCtx.fileSinkSet) {
        GenTezUtils.processFileSink(procCtx, fileSink);
    }
    // Connect any edges required for min/max pushdown
    if (pCtx.getRsToRuntimeValuesInfoMap().size() > 0) {
        for (ReduceSinkOperator rs : pCtx.getRsToRuntimeValuesInfoMap().keySet()) {
            // Process min/max
            GenTezUtils.processDynamicSemiJoinPushDownOperator(procCtx, pCtx.getRsToRuntimeValuesInfoMap().get(rs), rs);
        }
    }
    // and finally we hook up any events that need to be sent to the tez AM
    LOG.debug("There are " + procCtx.eventOperatorSet.size() + " app master events.");
    for (AppMasterEventOperator event : procCtx.eventOperatorSet) {
        LOG.debug("Handling AppMasterEventOperator: " + event);
        GenTezUtils.processAppMasterEvent(procCtx, event);
    }
    perfLogger.PerfLogEnd(this.getClass().getName(), PerfLogger.TEZ_COMPILER, "generateTaskTree");
}

Example 5 with FileSinkOperator

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

the class SparkCompiler method generateTaskTree.

/**
 * TODO: need to turn on rules that's commented out and add more if necessary.
 */
@Override
protected void generateTaskTree(List<Task<? extends Serializable>> rootTasks, ParseContext pCtx, List<Task<MoveWork>> mvTask, Set<ReadEntity> inputs, Set<WriteEntity> outputs) throws SemanticException {
    PERF_LOGGER.PerfLogBegin(CLASS_NAME, PerfLogger.SPARK_GENERATE_TASK_TREE);
    GenSparkUtils utils = GenSparkUtils.getUtils();
    utils.resetSequenceNumber();
    ParseContext tempParseContext = getParseContext(pCtx, rootTasks);
    GenSparkProcContext procCtx = new GenSparkProcContext(conf, tempParseContext, mvTask, rootTasks, inputs, outputs, pCtx.getTopOps());
    // -------------------------------- First Pass ---------------------------------- //
    // Identify SparkPartitionPruningSinkOperators, and break OP tree if necessary
    Map<Rule, NodeProcessor> opRules = new LinkedHashMap<Rule, NodeProcessor>();
    opRules.put(new RuleRegExp("Clone OP tree for PartitionPruningSink", SparkPartitionPruningSinkOperator.getOperatorName() + "%"), new SplitOpTreeForDPP());
    Dispatcher disp = new DefaultRuleDispatcher(null, opRules, procCtx);
    GraphWalker ogw = new GenSparkWorkWalker(disp, procCtx);
    List<Node> topNodes = new ArrayList<Node>();
    topNodes.addAll(pCtx.getTopOps().values());
    ogw.startWalking(topNodes, null);
    // -------------------------------- Second Pass ---------------------------------- //
    // Process operator tree in two steps: first we process the extra op trees generated
    // in the first pass. Then we process the main op tree, and the result task will depend
    // on the task generated in the first pass.
    topNodes.clear();
    topNodes.addAll(procCtx.topOps.values());
    generateTaskTreeHelper(procCtx, topNodes);
    // the partitions used.
    if (!procCtx.clonedPruningTableScanSet.isEmpty()) {
        SparkTask pruningTask = SparkUtilities.createSparkTask(conf);
        SparkTask mainTask = procCtx.currentTask;
        pruningTask.addDependentTask(procCtx.currentTask);
        procCtx.rootTasks.remove(procCtx.currentTask);
        procCtx.rootTasks.add(pruningTask);
        procCtx.currentTask = pruningTask;
        topNodes.clear();
        topNodes.addAll(procCtx.clonedPruningTableScanSet);
        generateTaskTreeHelper(procCtx, topNodes);
        procCtx.currentTask = mainTask;
    }
    // we need to clone some operator plans and remove union operators still
    for (BaseWork w : procCtx.workWithUnionOperators) {
        GenSparkUtils.getUtils().removeUnionOperators(procCtx, w);
    }
    // we need to fill MapWork with 'local' work and bucket information for SMB Join.
    GenSparkUtils.getUtils().annotateMapWork(procCtx);
    // finally make sure the file sink operators are set up right
    for (FileSinkOperator fileSink : procCtx.fileSinkSet) {
        GenSparkUtils.getUtils().processFileSink(procCtx, fileSink);
    }
    // Process partition pruning sinks
    for (Operator<?> prunerSink : procCtx.pruningSinkSet) {
        utils.processPartitionPruningSink(procCtx, (SparkPartitionPruningSinkOperator) prunerSink);
    }
    PERF_LOGGER.PerfLogEnd(CLASS_NAME, PerfLogger.SPARK_GENERATE_TASK_TREE);
}