Examples with DefaultFetchFormatter org.apache.hadoop.hive.serde2.DefaultFetchFormatter used on opensource projects

Example 1 with DefaultFetchFormatter

use of org.apache.hadoop.hive.serde2.DefaultFetchFormatter in project hive by apache.

the class TaskCompiler method compile.

@SuppressWarnings({ "nls", "unchecked" })
public void compile(final ParseContext pCtx, final List<Task<? extends Serializable>> rootTasks, final HashSet<ReadEntity> inputs, final HashSet<WriteEntity> outputs) throws SemanticException {
    Context ctx = pCtx.getContext();
    GlobalLimitCtx globalLimitCtx = pCtx.getGlobalLimitCtx();
    List<Task<MoveWork>> mvTask = new ArrayList<Task<MoveWork>>();
    List<LoadTableDesc> loadTableWork = pCtx.getLoadTableWork();
    List<LoadFileDesc> loadFileWork = pCtx.getLoadFileWork();
    boolean isCStats = pCtx.getQueryProperties().isAnalyzeRewrite();
    int outerQueryLimit = pCtx.getQueryProperties().getOuterQueryLimit();
    if (pCtx.getFetchTask() != null) {
        if (pCtx.getFetchTask().getTblDesc() == null) {
            return;
        }
        pCtx.getFetchTask().getWork().setHiveServerQuery(SessionState.get().isHiveServerQuery());
        TableDesc resultTab = pCtx.getFetchTask().getTblDesc();
        // then either the ThriftFormatter or the DefaultFetchFormatter should be used.
        if (!resultTab.getSerdeClassName().equalsIgnoreCase(ThriftJDBCBinarySerDe.class.getName())) {
            if (SessionState.get().isHiveServerQuery()) {
                conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, ThriftFormatter.class.getName());
            } else {
                String formatterName = conf.get(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER);
                if (formatterName == null || formatterName.isEmpty()) {
                    conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, DefaultFetchFormatter.class.getName());
                }
            }
        }
        return;
    }
    optimizeOperatorPlan(pCtx, inputs, outputs);
    /*
     * In case of a select, use a fetch task instead of a move task.
     * If the select is from analyze table column rewrite, don't create a fetch task. Instead create
     * a column stats task later.
     */
    if (pCtx.getQueryProperties().isQuery() && !isCStats) {
        if ((!loadTableWork.isEmpty()) || (loadFileWork.size() != 1)) {
            throw new SemanticException(ErrorMsg.INVALID_LOAD_TABLE_FILE_WORK.getMsg());
        }
        LoadFileDesc loadFileDesc = loadFileWork.get(0);
        String cols = loadFileDesc.getColumns();
        String colTypes = loadFileDesc.getColumnTypes();
        String resFileFormat;
        TableDesc resultTab = pCtx.getFetchTableDesc();
        if (resultTab == null) {
            resFileFormat = HiveConf.getVar(conf, HiveConf.ConfVars.HIVEQUERYRESULTFILEFORMAT);
            if (SessionState.get().getIsUsingThriftJDBCBinarySerDe() && (resFileFormat.equalsIgnoreCase("SequenceFile"))) {
                resultTab = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, resFileFormat, ThriftJDBCBinarySerDe.class);
                // Set the fetch formatter to be a no-op for the ListSinkOperator, since we'll
                // read formatted thrift objects from the output SequenceFile written by Tasks.
                conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, NoOpFetchFormatter.class.getName());
            } else {
                resultTab = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, resFileFormat, LazySimpleSerDe.class);
            }
        } else {
            if (resultTab.getProperties().getProperty(serdeConstants.SERIALIZATION_LIB).equalsIgnoreCase(ThriftJDBCBinarySerDe.class.getName())) {
                // Set the fetch formatter to be a no-op for the ListSinkOperator, since we'll
                // read formatted thrift objects from the output SequenceFile written by Tasks.
                conf.set(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER, NoOpFetchFormatter.class.getName());
            }
        }
        FetchWork fetch = new FetchWork(loadFileDesc.getSourcePath(), resultTab, outerQueryLimit);
        boolean isHiveServerQuery = SessionState.get().isHiveServerQuery();
        fetch.setHiveServerQuery(isHiveServerQuery);
        fetch.setSource(pCtx.getFetchSource());
        fetch.setSink(pCtx.getFetchSink());
        if (isHiveServerQuery && null != resultTab && resultTab.getSerdeClassName().equalsIgnoreCase(ThriftJDBCBinarySerDe.class.getName()) && HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_SERVER2_THRIFT_RESULTSET_SERIALIZE_IN_TASKS)) {
            fetch.setIsUsingThriftJDBCBinarySerDe(true);
        } else {
            fetch.setIsUsingThriftJDBCBinarySerDe(false);
        }
        pCtx.setFetchTask((FetchTask) TaskFactory.get(fetch, conf));
        // For the FetchTask, the limit optimization requires we fetch all the rows
        // in memory and count how many rows we get. It's not practical if the
        // limit factor is too big
        int fetchLimit = HiveConf.getIntVar(conf, HiveConf.ConfVars.HIVELIMITOPTMAXFETCH);
        if (globalLimitCtx.isEnable() && globalLimitCtx.getGlobalLimit() > fetchLimit) {
            LOG.info("For FetchTask, LIMIT " + globalLimitCtx.getGlobalLimit() + " > " + fetchLimit + ". Doesn't qualify limit optimization.");
            globalLimitCtx.disableOpt();
        }
        if (outerQueryLimit == 0) {
            // Believe it or not, some tools do generate queries with limit 0 and than expect
            // query to run quickly. Lets meet their requirement.
            LOG.info("Limit 0. No query execution needed.");
            return;
        }
    } else if (!isCStats) {
        for (LoadTableDesc ltd : loadTableWork) {
            Task<MoveWork> tsk = TaskFactory.get(new MoveWork(null, null, ltd, null, false), conf);
            mvTask.add(tsk);
            // Check to see if we are stale'ing any indexes and auto-update them if we want
            if (HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVEINDEXAUTOUPDATE)) {
                IndexUpdater indexUpdater = new IndexUpdater(loadTableWork, inputs, conf);
                try {
                    List<Task<? extends Serializable>> indexUpdateTasks = indexUpdater.generateUpdateTasks();
                    for (Task<? extends Serializable> updateTask : indexUpdateTasks) {
                        tsk.addDependentTask(updateTask);
                    }
                } catch (HiveException e) {
                    console.printInfo("WARNING: could not auto-update stale indexes, which are not in sync");
                }
            }
        }
        boolean oneLoadFile = true;
        for (LoadFileDesc lfd : loadFileWork) {
            if (pCtx.getQueryProperties().isCTAS() || pCtx.getQueryProperties().isMaterializedView()) {
                // should not have more than 1 load file for
                assert (oneLoadFile);
                // CTAS
                // make the movetask's destination directory the table's destination.
                Path location;
                String loc = pCtx.getQueryProperties().isCTAS() ? pCtx.getCreateTable().getLocation() : pCtx.getCreateViewDesc().getLocation();
                if (loc == null) {
                    // get the default location
                    Path targetPath;
                    try {
                        String protoName = null;
                        if (pCtx.getQueryProperties().isCTAS()) {
                            protoName = pCtx.getCreateTable().getTableName();
                        } else if (pCtx.getQueryProperties().isMaterializedView()) {
                            protoName = pCtx.getCreateViewDesc().getViewName();
                        }
                        String[] names = Utilities.getDbTableName(protoName);
                        if (!db.databaseExists(names[0])) {
                            throw new SemanticException("ERROR: The database " + names[0] + " does not exist.");
                        }
                        Warehouse wh = new Warehouse(conf);
                        targetPath = wh.getTablePath(db.getDatabase(names[0]), names[1]);
                    } catch (HiveException e) {
                        throw new SemanticException(e);
                    } catch (MetaException e) {
                        throw new SemanticException(e);
                    }
                    location = targetPath;
                } else {
                    location = new Path(loc);
                }
                lfd.setTargetDir(location);
                oneLoadFile = false;
            }
            mvTask.add(TaskFactory.get(new MoveWork(null, null, null, lfd, false), conf));
        }
    }
    generateTaskTree(rootTasks, pCtx, mvTask, inputs, outputs);
    // For each task, set the key descriptor for the reducer
    for (Task<? extends Serializable> rootTask : rootTasks) {
        GenMapRedUtils.setKeyAndValueDescForTaskTree(rootTask);
    }
    // to be used, please do so
    for (Task<? extends Serializable> rootTask : rootTasks) {
        setInputFormat(rootTask);
    }
    optimizeTaskPlan(rootTasks, pCtx, ctx);
    /*
     * If the query was the result of analyze table column compute statistics rewrite, create
     * a column stats task instead of a fetch task to persist stats to the metastore.
     */
    if (isCStats || !pCtx.getColumnStatsAutoGatherContexts().isEmpty()) {
        Set<Task<? extends Serializable>> leafTasks = new LinkedHashSet<Task<? extends Serializable>>();
        getLeafTasks(rootTasks, leafTasks);
        if (isCStats) {
            genColumnStatsTask(pCtx.getAnalyzeRewrite(), loadFileWork, leafTasks, outerQueryLimit, 0);
        } else {
            for (ColumnStatsAutoGatherContext columnStatsAutoGatherContext : pCtx.getColumnStatsAutoGatherContexts()) {
                if (!columnStatsAutoGatherContext.isInsertInto()) {
                    genColumnStatsTask(columnStatsAutoGatherContext.getAnalyzeRewrite(), columnStatsAutoGatherContext.getLoadFileWork(), leafTasks, outerQueryLimit, 0);
                } else {
                    int numBitVector;
                    try {
                        numBitVector = HiveStatsUtils.getNumBitVectorsForNDVEstimation(conf);
                    } catch (Exception e) {
                        throw new SemanticException(e.getMessage());
                    }
                    genColumnStatsTask(columnStatsAutoGatherContext.getAnalyzeRewrite(), columnStatsAutoGatherContext.getLoadFileWork(), leafTasks, outerQueryLimit, numBitVector);
                }
            }
        }
    }
    decideExecMode(rootTasks, ctx, globalLimitCtx);
    if (pCtx.getQueryProperties().isCTAS() && !pCtx.getCreateTable().isMaterialization()) {
        // generate a DDL task and make it a dependent task of the leaf
        CreateTableDesc crtTblDesc = pCtx.getCreateTable();
        crtTblDesc.validate(conf);
        Task<? extends Serializable> crtTblTask = TaskFactory.get(new DDLWork(inputs, outputs, crtTblDesc), conf);
        patchUpAfterCTASorMaterializedView(rootTasks, outputs, crtTblTask);
    } else if (pCtx.getQueryProperties().isMaterializedView()) {
        // generate a DDL task and make it a dependent task of the leaf
        CreateViewDesc viewDesc = pCtx.getCreateViewDesc();
        Task<? extends Serializable> crtViewTask = TaskFactory.get(new DDLWork(inputs, outputs, viewDesc), conf);
        patchUpAfterCTASorMaterializedView(rootTasks, outputs, crtViewTask);
    }
    if (globalLimitCtx.isEnable() && pCtx.getFetchTask() != null) {
        LOG.info("set least row check for FetchTask: " + globalLimitCtx.getGlobalLimit());
        pCtx.getFetchTask().getWork().setLeastNumRows(globalLimitCtx.getGlobalLimit());
    }
    if (globalLimitCtx.isEnable() && globalLimitCtx.getLastReduceLimitDesc() != null) {
        LOG.info("set least row check for LimitDesc: " + globalLimitCtx.getGlobalLimit());
        globalLimitCtx.getLastReduceLimitDesc().setLeastRows(globalLimitCtx.getGlobalLimit());
        List<ExecDriver> mrTasks = Utilities.getMRTasks(rootTasks);
        for (ExecDriver tsk : mrTasks) {
            tsk.setRetryCmdWhenFail(true);
        }
        List<SparkTask> sparkTasks = Utilities.getSparkTasks(rootTasks);
        for (SparkTask sparkTask : sparkTasks) {
            sparkTask.setRetryCmdWhenFail(true);
        }
    }
    Interner<TableDesc> interner = Interners.newStrongInterner();
    for (Task<? extends Serializable> rootTask : rootTasks) {
        GenMapRedUtils.internTableDesc(rootTask, interner);
        GenMapRedUtils.deriveFinalExplainAttributes(rootTask, pCtx.getConf());
    }
}

Example 2 with DefaultFetchFormatter

use of org.apache.hadoop.hive.serde2.DefaultFetchFormatter in project hive by apache.

the class ListSinkOperator method initializeFetcher.

private FetchFormatter initializeFetcher(Configuration conf) throws Exception {
    String formatterName = conf.get(SerDeUtils.LIST_SINK_OUTPUT_FORMATTER);
    FetchFormatter fetcher;
    if (formatterName != null && !formatterName.isEmpty()) {
        Class<? extends FetchFormatter> fetcherClass = Class.forName(formatterName, true, Utilities.getSessionSpecifiedClassLoader()).asSubclass(FetchFormatter.class);
        fetcher = ReflectionUtils.newInstance(fetcherClass, null);
    } else {
        fetcher = new DefaultFetchFormatter();
    }
    // selectively used by fetch formatter
    Properties props = new Properties();
    props.put(serdeConstants.SERIALIZATION_FORMAT, "" + Utilities.tabCode);
    props.put(serdeConstants.SERIALIZATION_NULL_FORMAT, getConf().getSerializationNullFormat());
    fetcher.initialize(conf, props);
    return fetcher;
}