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

Example 11 with LoadTableDesc

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

the class MoveTask method execute.

@Override
public int execute() {
    if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
        Utilities.FILE_OP_LOGGER.trace("Executing MoveWork " + System.identityHashCode(work) + " with " + work.getLoadFileWork() + "; " + work.getLoadTableWork() + "; " + work.getLoadMultiFilesWork());
    }
    if (context.getExplainAnalyze() == AnalyzeState.RUNNING) {
        return 0;
    }
    try (LocalTableLock lock = acquireLockForFileMove(work.getLoadTableWork())) {
        if (checkAndCommitNatively(work, conf)) {
            return 0;
        }
        Hive db = getHive();
        // Do any hive related operations like moving tables and files
        // to appropriate locations
        LoadFileDesc lfd = work.getLoadFileWork();
        if (lfd != null) {
            Path targetPath = lfd.getTargetDir();
            Path sourcePath = lfd.getSourcePath();
            if (targetPath.equals(sourcePath)) {
                Utilities.FILE_OP_LOGGER.debug("MoveTask not moving " + sourcePath);
            } else {
                Utilities.FILE_OP_LOGGER.debug("MoveTask moving " + sourcePath + " to " + targetPath);
                if (lfd.getWriteType() == AcidUtils.Operation.INSERT) {
                    // 'sourcePath' result of 'select ...' part of CTAS statement
                    assert lfd.getIsDfsDir();
                    FileSystem srcFs = sourcePath.getFileSystem(conf);
                    FileStatus[] srcs = srcFs.globStatus(sourcePath);
                    if (srcs != null) {
                        Hive.moveAcidFiles(srcFs, srcs, targetPath, null);
                    } else {
                        LOG.debug("No files found to move from " + sourcePath + " to " + targetPath);
                    }
                } else {
                    moveFile(sourcePath, targetPath, lfd.getIsDfsDir());
                }
            }
        }
        // Multi-file load is for dynamic partitions when some partitions do not
        // need to merge and they can simply be moved to the target directory.
        // This is also used for MM table conversion.
        LoadMultiFilesDesc lmfd = work.getLoadMultiFilesWork();
        if (lmfd != null) {
            boolean isDfsDir = lmfd.getIsDfsDir();
            List<String> targetPrefixes = lmfd.getTargetPrefixes();
            for (int i = 0; i < lmfd.getSourceDirs().size(); ++i) {
                Path srcPath = lmfd.getSourceDirs().get(i);
                Path destPath = lmfd.getTargetDirs().get(i);
                if (destPath.equals(srcPath)) {
                    continue;
                }
                String filePrefix = targetPrefixes == null ? null : targetPrefixes.get(i);
                FileSystem destFs = destPath.getFileSystem(conf);
                if (filePrefix == null) {
                    if (!destFs.exists(destPath.getParent())) {
                        destFs.mkdirs(destPath.getParent());
                    }
                    Utilities.FILE_OP_LOGGER.debug("MoveTask moving (multi-file) " + srcPath + " to " + destPath);
                    moveFile(srcPath, destPath, isDfsDir);
                } else {
                    if (!destFs.exists(destPath)) {
                        destFs.mkdirs(destPath);
                    }
                    FileSystem srcFs = srcPath.getFileSystem(conf);
                    FileStatus[] children = srcFs.listStatus(srcPath);
                    if (children != null) {
                        for (FileStatus child : children) {
                            Path childSrc = child.getPath();
                            Path childDest = new Path(destPath, filePrefix + childSrc.getName());
                            Utilities.FILE_OP_LOGGER.debug("MoveTask moving (multi-file) " + childSrc + " to " + childDest);
                            moveFile(childSrc, childDest, isDfsDir);
                        }
                    } else {
                        Utilities.FILE_OP_LOGGER.debug("MoveTask skipping empty directory (multi-file) " + srcPath);
                    }
                    if (!srcFs.delete(srcPath, false)) {
                        throw new IOException("Couldn't delete " + srcPath + " after moving all the files");
                    }
                }
            }
        }
        // Next we do this for tables and partitions
        LoadTableDesc tbd = work.getLoadTableWork();
        if (tbd != null) {
            logMessage(tbd);
            Table table = db.getTable(tbd.getTable().getTableName());
            checkFileFormats(db, tbd, table);
            boolean isFullAcidOp = work.getLoadTableWork().getWriteType() != AcidUtils.Operation.NOT_ACID && // it seems that LoadTableDesc has Operation.INSERT only for CTAS...
            !tbd.isMmTable();
            // Create a data container
            DataContainer dc = null;
            if (tbd.getPartitionSpec().size() == 0) {
                dc = new DataContainer(table.getTTable());
                if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                    Utilities.FILE_OP_LOGGER.trace("loadTable called from " + tbd.getSourcePath() + " into " + tbd.getTable().getTableName());
                }
                int statementId = tbd.getStmtId();
                if (tbd.isDirectInsert() || tbd.isMmTable()) {
                    statementId = queryPlan.getStatementIdForAcidWriteType(work.getLoadTableWork().getWriteId(), tbd.getMoveTaskId(), work.getLoadTableWork().getWriteType(), tbd.getSourcePath());
                    LOG.debug("The statementId used when loading the dynamic partitions is " + statementId);
                }
                db.loadTable(tbd.getSourcePath(), tbd.getTable().getTableName(), tbd.getLoadFileType(), work.isSrcLocal(), isSkewedStoredAsDirs(tbd), isFullAcidOp, resetStatisticsProps(table), tbd.getWriteId(), statementId, tbd.isInsertOverwrite(), tbd.isDirectInsert());
                if (work.getOutputs() != null) {
                    DDLUtils.addIfAbsentByName(new WriteEntity(table, getWriteType(tbd, work.getLoadTableWork().getWriteType())), work.getOutputs());
                }
            } else {
                LOG.info("Partition is: {}", tbd.getPartitionSpec());
                // Check if the bucketing and/or sorting columns were inferred
                TaskInformation ti = new TaskInformation(this, tbd.getSourcePath().toUri().toString());
                inferTaskInformation(ti);
                // deal with dynamic partitions
                DynamicPartitionCtx dpCtx = tbd.getDPCtx();
                if (dpCtx != null && dpCtx.getNumDPCols() > 0) {
                    // dynamic partitions
                    dc = handleDynParts(db, table, tbd, ti, dpCtx);
                } else {
                    // static partitions
                    dc = handleStaticParts(db, table, tbd, ti);
                }
            }
            if (dc != null) {
                // If we are doing an update or a delete the number of columns in the table will not
                // match the number of columns in the file sink.  For update there will be one too many
                // (because of the ROW__ID), and in the case of the delete there will be just the
                // ROW__ID, which we don't need to worry about from a lineage perspective.
                List<FieldSchema> tableCols = null;
                switch(work.getLoadTableWork().getWriteType()) {
                    case DELETE:
                    case UPDATE:
                        // Pass an empty list as no columns will be written to the file.
                        // TODO I should be able to make this work for update
                        tableCols = new ArrayList<>();
                        break;
                    default:
                        tableCols = table.getCols();
                        break;
                }
                queryState.getLineageState().setLineage(tbd.getSourcePath(), dc, tableCols);
            }
            releaseLocks(tbd);
        }
        return 0;
    } catch (HiveException he) {
        int errorCode = 1;
        if (he.getCanonicalErrorMsg() != ErrorMsg.GENERIC_ERROR) {
            errorCode = he.getCanonicalErrorMsg().getErrorCode();
            if (he.getCanonicalErrorMsg() == ErrorMsg.UNRESOLVED_RT_EXCEPTION) {
                console.printError("Failed with exception " + he.getMessage(), "\n" + StringUtils.stringifyException(he));
            } else {
                console.printError("Failed with exception " + he.getMessage() + "\nRemote Exception: " + he.getRemoteErrorMsg());
                console.printInfo("\n", StringUtils.stringifyException(he), false);
            }
        }
        setException(he);
        errorCode = ReplUtils.handleException(work.isReplication(), he, work.getDumpDirectory(), work.getMetricCollector(), getName(), conf);
        return errorCode;
    } catch (Exception e) {
        console.printError("Failed with exception " + e.getMessage(), "\n" + StringUtils.stringifyException(e));
        setException(e);
        LOG.error("MoveTask failed", e);
        return ReplUtils.handleException(work.isReplication(), e, work.getDumpDirectory(), work.getMetricCollector(), getName(), conf);
    }
}

Example 12 with LoadTableDesc

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

the class GenMapRedUtils method mergeMovePaths.

/**
 * Merges the given Conditional input path and the linked MoveWork into one only MoveWork.
 * This is an optimization for BlobStore systems to avoid doing two renames or copies that are not necessary.
 *
 * @param condInputPath A path that the ConditionalTask uses as input for its sub-tasks.
 * @param linkedMoveWork A MoveWork that the ConditionalTask uses to link to its sub-tasks.
 * @param lineageState A LineageState used to track what changes.
 * @return A new MoveWork that has the Conditional input path as source and the linkedMoveWork as target.
 */
@VisibleForTesting
protected static MoveWork mergeMovePaths(Path condInputPath, MoveWork linkedMoveWork, LineageState lineageState) {
    MoveWork newWork = new MoveWork(linkedMoveWork);
    LoadFileDesc fileDesc = null;
    LoadTableDesc tableDesc = null;
    if (linkedMoveWork.getLoadFileWork() != null) {
        fileDesc = new LoadFileDesc(linkedMoveWork.getLoadFileWork());
        fileDesc.setSourcePath(condInputPath);
        lineageState.updateDirToOpMap(condInputPath, linkedMoveWork.getLoadFileWork().getSourcePath());
    } else if (linkedMoveWork.getLoadTableWork() != null) {
        tableDesc = new LoadTableDesc(linkedMoveWork.getLoadTableWork());
        tableDesc.setSourcePath(condInputPath);
        lineageState.updateDirToOpMap(condInputPath, linkedMoveWork.getLoadTableWork().getSourcePath());
    } else {
        throw new IllegalArgumentException("Merging a path with a MoveWork with multi-files work is not allowed.");
    }
    newWork.setLoadFileWork(fileDesc);
    newWork.setLoadTableWork(tableDesc);
    return newWork;
}

Example 13 with LoadTableDesc

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

the class TaskCompiler method compile.

@SuppressWarnings("nls")
public void compile(final ParseContext pCtx, final List<Task<?>> rootTasks, final Set<ReadEntity> inputs, final Set<WriteEntity> outputs) throws SemanticException {
    Context ctx = pCtx.getContext();
    GlobalLimitCtx globalLimitCtx = pCtx.getGlobalLimitCtx();
    List<Task<MoveWork>> mvTask = new ArrayList<>();
    List<LoadTableDesc> loadTableWork = pCtx.getLoadTableWork();
    List<LoadFileDesc> loadFileWork = pCtx.getLoadFileWork();
    boolean isCStats = pCtx.getQueryProperties().isAnalyzeRewrite();
    int outerQueryLimit = pCtx.getQueryProperties().getOuterQueryLimit();
    boolean directInsertCtas = false;
    if (pCtx.getCreateTable() != null && pCtx.getCreateTable().getStorageHandler() != null) {
        try {
            directInsertCtas = HiveUtils.getStorageHandler(conf, pCtx.getCreateTable().getStorageHandler()).directInsertCTAS();
        } catch (HiveException e) {
            throw new SemanticException("Failed to load storage handler:  " + e.getMessage());
        }
    }
    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;
    }
    if (!pCtx.getQueryProperties().isAnalyzeCommand()) {
        LOG.debug("Skipping optimize operator plan for analyze command.");
        optimizeOperatorPlan(pCtx);
    }
    /*
     * 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();
        TableDesc resultTab = pCtx.getFetchTableDesc();
        boolean shouldSetOutputFormatter = false;
        if (resultTab == null) {
            ResultFileFormat resFileFormat = conf.getResultFileFormat();
            String fileFormat;
            Class<? extends Deserializer> serdeClass;
            if (SessionState.get().getIsUsingThriftJDBCBinarySerDe() && resFileFormat == ResultFileFormat.SEQUENCEFILE) {
                fileFormat = resFileFormat.toString();
                serdeClass = ThriftJDBCBinarySerDe.class;
                shouldSetOutputFormatter = true;
            } else if (resFileFormat == ResultFileFormat.SEQUENCEFILE) {
                // file format is changed so that IF file sink provides list of files to fetch from (instead
                // of whole directory) list status is done on files (which is what HiveSequenceFileInputFormat does)
                fileFormat = "HiveSequenceFile";
                serdeClass = LazySimpleSerDe.class;
            } else {
                // All other cases we use the defined file format and LazySimpleSerde
                fileFormat = resFileFormat.toString();
                serdeClass = LazySimpleSerDe.class;
            }
            resultTab = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, fileFormat, serdeClass);
        } else {
            shouldSetOutputFormatter = resultTab.getProperties().getProperty(serdeConstants.SERIALIZATION_LIB).equalsIgnoreCase(ThriftJDBCBinarySerDe.class.getName());
        }
        if (shouldSetOutputFormatter) {
            // Set the fetch formatter to be a no-op for the ListSinkOperator, since we will
            // 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);
        }
        // The idea here is to keep an object reference both in FileSink and in FetchTask for list of files
        // to be fetched. During Job close file sink will populate the list and fetch task later will use it
        // to fetch the results.
        Collection<Operator<?>> tableScanOps = Lists.<Operator<?>>newArrayList(pCtx.getTopOps().values());
        Set<FileSinkOperator> fsOps = OperatorUtils.findOperators(tableScanOps, FileSinkOperator.class);
        if (fsOps != null && fsOps.size() == 1) {
            FileSinkOperator op = fsOps.iterator().next();
            Set<FileStatus> filesToFetch = new HashSet<>();
            op.getConf().setFilesToFetch(filesToFetch);
            fetch.setFilesToFetch(filesToFetch);
        }
        pCtx.setFetchTask((FetchTask) TaskFactory.get(fetch));
        // 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));
            mvTask.add(tsk);
        }
        boolean oneLoadFileForCtas = true;
        for (LoadFileDesc lfd : loadFileWork) {
            if (pCtx.getQueryProperties().isCTAS() || pCtx.getQueryProperties().isMaterializedView()) {
                if (!oneLoadFileForCtas) {
                    // should not have more than 1 load file for CTAS.
                    throw new SemanticException("One query is not expected to contain multiple CTAS loads statements");
                }
                setLoadFileLocation(pCtx, lfd);
                oneLoadFileForCtas = false;
            }
            mvTask.add(TaskFactory.get(new MoveWork(null, null, null, lfd, false)));
        }
    }
    generateTaskTree(rootTasks, pCtx, mvTask, inputs, outputs);
    // For each task, set the key descriptor for the reducer
    for (Task<?> rootTask : rootTasks) {
        GenMapRedUtils.setKeyAndValueDescForTaskTree(rootTask);
    }
    // to be used, please do so
    for (Task<?> 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.
     * As per HIVE-15903, we will also collect table stats when user computes column stats.
     * That means, if isCStats || !pCtx.getColumnStatsAutoGatherContexts().isEmpty()
     * We need to collect table stats
     * if isCStats, we need to include a basic stats task
     * else it is ColumnStatsAutoGather, which should have a move task with a stats task already.
     */
    if (isCStats || !pCtx.getColumnStatsAutoGatherContexts().isEmpty()) {
        // map from tablename to task (ColumnStatsTask which includes a BasicStatsTask)
        Map<String, StatsTask> map = new LinkedHashMap<>();
        if (isCStats) {
            if (rootTasks == null || rootTasks.size() != 1 || pCtx.getTopOps() == null || pCtx.getTopOps().size() != 1) {
                throw new SemanticException("Can not find correct root task!");
            }
            try {
                Task<?> root = rootTasks.iterator().next();
                StatsTask tsk = (StatsTask) genTableStats(pCtx, pCtx.getTopOps().values().iterator().next(), root, outputs);
                root.addDependentTask(tsk);
                map.put(extractTableFullName(tsk), tsk);
            } catch (HiveException e) {
                throw new SemanticException(e);
            }
            genColumnStatsTask(pCtx.getAnalyzeRewrite(), loadFileWork, map, outerQueryLimit, 0);
        } else {
            Set<Task<?>> leafTasks = new LinkedHashSet<Task<?>>();
            getLeafTasks(rootTasks, leafTasks);
            List<Task<?>> nonStatsLeafTasks = new ArrayList<>();
            for (Task<?> tsk : leafTasks) {
                // map table name to the correct ColumnStatsTask
                if (tsk instanceof StatsTask) {
                    map.put(extractTableFullName((StatsTask) tsk), (StatsTask) tsk);
                } else {
                    nonStatsLeafTasks.add(tsk);
                }
            }
            // add cStatsTask as a dependent of all the nonStatsLeafTasks
            for (Task<?> tsk : nonStatsLeafTasks) {
                for (Task<?> cStatsTask : map.values()) {
                    tsk.addDependentTask(cStatsTask);
                }
            }
            for (ColumnStatsAutoGatherContext columnStatsAutoGatherContext : pCtx.getColumnStatsAutoGatherContexts()) {
                if (!columnStatsAutoGatherContext.isInsertInto()) {
                    genColumnStatsTask(columnStatsAutoGatherContext.getAnalyzeRewrite(), columnStatsAutoGatherContext.getLoadFileWork(), map, outerQueryLimit, 0);
                } else {
                    int numBitVector;
                    try {
                        numBitVector = HiveStatsUtils.getNumBitVectorsForNDVEstimation(conf);
                    } catch (Exception e) {
                        throw new SemanticException(e.getMessage());
                    }
                    genColumnStatsTask(columnStatsAutoGatherContext.getAnalyzeRewrite(), columnStatsAutoGatherContext.getLoadFileWork(), map, outerQueryLimit, numBitVector);
                }
            }
        }
    }
    decideExecMode(rootTasks, ctx, globalLimitCtx);
    // ahead of time by the non-native table
    if (pCtx.getQueryProperties().isCTAS() && !pCtx.getCreateTable().isMaterialization() && !directInsertCtas) {
        // generate a DDL task and make it a dependent task of the leaf
        CreateTableDesc crtTblDesc = pCtx.getCreateTable();
        crtTblDesc.validate(conf);
        Task<?> crtTblTask = TaskFactory.get(new DDLWork(inputs, outputs, crtTblDesc));
        patchUpAfterCTASorMaterializedView(rootTasks, inputs, outputs, crtTblTask, CollectionUtils.isEmpty(crtTblDesc.getPartColNames()));
    } else if (pCtx.getQueryProperties().isMaterializedView()) {
        // generate a DDL task and make it a dependent task of the leaf
        CreateMaterializedViewDesc viewDesc = pCtx.getCreateViewDesc();
        Task<?> crtViewTask = TaskFactory.get(new DDLWork(inputs, outputs, viewDesc));
        patchUpAfterCTASorMaterializedView(rootTasks, inputs, outputs, crtViewTask, CollectionUtils.isEmpty(viewDesc.getPartColNames()));
    } else if (pCtx.getMaterializedViewUpdateDesc() != null) {
        // If there is a materialized view update desc, we create introduce it at the end
        // of the tree.
        MaterializedViewUpdateDesc materializedViewDesc = pCtx.getMaterializedViewUpdateDesc();
        DDLWork ddlWork = new DDLWork(inputs, outputs, materializedViewDesc);
        Set<Task<?>> leafTasks = new LinkedHashSet<Task<?>>();
        getLeafTasks(rootTasks, leafTasks);
        Task<?> materializedViewTask = TaskFactory.get(ddlWork, conf);
        for (Task<?> task : leafTasks) {
            task.addDependentTask(materializedViewTask);
        }
    }
    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());
    }
    Interner<TableDesc> interner = Interners.newStrongInterner();
    // Perform Final chores on generated Map works
    // 1.  Intern the table descriptors
    // 2.  Derive final explain attributes based on previous compilation.
    GenMapRedUtils.finalMapWorkChores(rootTasks, pCtx.getConf(), interner);
}

Example 14 with LoadTableDesc

use of org.apache.hadoop.hive.ql.plan.LoadTableDesc 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 destType = qbm.getDestTypeForAlias(dest);
    // destination table if any
    Table destinationTable = null;
    // true for full ACID table and MM table
    boolean destTableIsTransactional;
    // should the destination table be written to using ACID
    boolean destTableIsFullAcid;
    // should we add files directly to the final path
    boolean isDirectInsert = false;
    AcidUtils.Operation acidOperation = null;
    boolean destTableIsTemporary = false;
    boolean destTableIsMaterialization = false;
    // destination partition if any
    Partition destinationPartition = null;
    // the intermediate destination directory
    Path queryTmpdir = null;
    String moveTaskId = null;
    // the final destination directory
    Path destinationPath = null;
    TableDesc tableDescriptor = null;
    StructObjectInspector specificRowObjectInspector = null;
    int currentTableId = 0;
    boolean isLocal = false;
    SortBucketRSCtx rsCtx = new SortBucketRSCtx();
    DynamicPartitionCtx dpCtx = null;
    LoadTableDesc ltd = null;
    ListBucketingCtx lbCtx = null;
    Map<String, String> partSpec = null;
    boolean isMmTable = false, isMmCreate = false, isNonNativeTable = false;
    Long writeId = null;
    HiveTxnManager txnMgr = getTxnMgr();
    switch(destType.intValue()) {
        case QBMetaData.DEST_TABLE:
            {
                destinationTable = qbm.getDestTableForAlias(dest);
                destTableIsTransactional = AcidUtils.isTransactionalTable(destinationTable);
                destTableIsFullAcid = AcidUtils.isFullAcidTable(destinationTable);
                destTableIsTemporary = destinationTable.isTemporary();
                // Is the user trying to insert into a external tables
                checkExternalTable(destinationTable);
                partSpec = qbm.getPartSpecForAlias(dest);
                destinationPath = destinationTable.getPath();
                checkImmutableTable(qb, destinationTable, destinationPath, false);
                // check for partition
                List<FieldSchema> parts = destinationTable.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) {
                        destinationTable.validatePartColumnNames(partSpec, false);
                        dpCtx = new DynamicPartitionCtx(partSpec, conf.getVar(HiveConf.ConfVars.DEFAULTPARTITIONNAME), conf.getIntVar(HiveConf.ConfVars.DYNAMICPARTITIONMAXPARTSPERNODE));
                        qbm.setDPCtx(dest, dpCtx);
                    }
                }
                // Check for dynamic partitions.
                dpCtx = checkDynPart(qb, qbm, destinationTable, partSpec, dest);
                if (dpCtx != null && dpCtx.getSPPath() != null) {
                    destinationPath = new Path(destinationTable.getPath(), dpCtx.getSPPath());
                }
                isNonNativeTable = destinationTable.isNonNative();
                isMmTable = AcidUtils.isInsertOnlyTable(destinationTable.getParameters());
                AcidUtils.Operation acidOp = AcidUtils.Operation.NOT_ACID;
                // this table_desc does not contain the partitioning columns
                tableDescriptor = Utilities.getTableDesc(destinationTable);
                if (!isNonNativeTable) {
                    if (destTableIsTransactional) {
                        acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                    }
                }
                isDirectInsert = isDirectInsert(destTableIsFullAcid, acidOp);
                acidOperation = acidOp;
                queryTmpdir = getTmpDir(isNonNativeTable, isMmTable, isDirectInsert, destinationPath);
                moveTaskId = getMoveTaskId();
                if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                    Utilities.FILE_OP_LOGGER.trace("create filesink w/DEST_TABLE specifying " + queryTmpdir + " from " + destinationPath);
                }
                if (dpCtx != null) {
                    // set the root of the temporary path where dynamic partition columns will populate
                    dpCtx.setRootPath(queryTmpdir);
                }
                // Add NOT NULL constraint check
                input = genConstraintsPlan(dest, qb, input);
                if (!qb.getIsQuery()) {
                    input = genConversionSelectOperator(dest, qb, input, destinationTable.getDeserializer(), dpCtx, parts);
                }
                if (destinationTable.isMaterializedView() && mvRebuildMode == MaterializationRebuildMode.INSERT_OVERWRITE_REBUILD) {
                    // Data organization (DISTRIBUTED, SORTED, CLUSTERED) for materialized view
                    // TODO: We only do this for a full rebuild
                    String sortColsStr = destinationTable.getProperty(Constants.MATERIALIZED_VIEW_SORT_COLUMNS);
                    String distributeColsStr = destinationTable.getProperty(Constants.MATERIALIZED_VIEW_DISTRIBUTE_COLUMNS);
                    if (sortColsStr != null || distributeColsStr != null) {
                        input = genMaterializedViewDataOrgPlan(destinationTable, sortColsStr, distributeColsStr, inputRR, input);
                    }
                } else {
                    // Add sorting/bucketing if needed
                    input = genBucketingSortingDest(dest, input, qb, tableDescriptor, destinationTable, rsCtx);
                }
                idToTableNameMap.put(String.valueOf(destTableId), destinationTable.getTableName());
                currentTableId = destTableId;
                destTableId++;
                // NOTE: specify Dynamic partitions in dest_tab for WriteEntity
                if (!isNonNativeTable || destinationTable.getStorageHandler().commitInMoveTask()) {
                    if (destTableIsTransactional) {
                        acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                        checkAcidConstraints();
                    } else {
                        lbCtx = constructListBucketingCtx(destinationTable.getSkewedColNames(), destinationTable.getSkewedColValues(), destinationTable.getSkewedColValueLocationMaps(), destinationTable.isStoredAsSubDirectories());
                    }
                    try {
                        if (ctx.getExplainConfig() != null) {
                            // For explain plan, txn won't be opened and doesn't make sense to allocate write id
                            writeId = null;
                        } else {
                            if (isMmTable) {
                                writeId = txnMgr.getTableWriteId(destinationTable.getDbName(), destinationTable.getTableName());
                            } else {
                                writeId = acidOp == Operation.NOT_ACID ? null : txnMgr.getTableWriteId(destinationTable.getDbName(), destinationTable.getTableName());
                            }
                        }
                    } catch (LockException ex) {
                        throw new SemanticException("Failed to allocate write Id", ex);
                    }
                    boolean isReplace = !qb.getParseInfo().isInsertIntoTable(destinationTable.getDbName(), destinationTable.getTableName());
                    ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, dpCtx, acidOp, isReplace, writeId);
                    if (writeId != null) {
                        ltd.setStmtId(txnMgr.getCurrentStmtId());
                    }
                    ltd.setMoveTaskId(moveTaskId);
                    // For Acid table, Insert Overwrite shouldn't replace the table content. We keep the old
                    // deltas and base and leave them up to the cleaner to clean up
                    boolean isInsertInto = qb.getParseInfo().isInsertIntoTable(destinationTable.getDbName(), destinationTable.getTableName());
                    LoadFileType loadType;
                    if (isDirectInsert) {
                        loadType = LoadFileType.IGNORE;
                    } else if (!isInsertInto && !destTableIsTransactional) {
                        loadType = LoadFileType.REPLACE_ALL;
                    } else {
                        loadType = LoadFileType.KEEP_EXISTING;
                    }
                    ltd.setLoadFileType(loadType);
                    ltd.setInsertOverwrite(!isInsertInto);
                    ltd.setIsDirectInsert(isDirectInsert);
                    ltd.setLbCtx(lbCtx);
                    loadTableWork.add(ltd);
                } else {
                    // This is a non-native table.
                    // We need to set stats as inaccurate.
                    setStatsForNonNativeTable(destinationTable.getDbName(), destinationTable.getTableName());
                    // true if it is insert overwrite.
                    boolean overwrite = !qb.getParseInfo().isInsertIntoTable(String.format("%s.%s", destinationTable.getDbName(), destinationTable.getTableName()));
                    createPreInsertDesc(destinationTable, overwrite);
                    ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, partSpec == null ? ImmutableMap.of() : partSpec);
                    ltd.setInsertOverwrite(overwrite);
                    ltd.setLoadFileType(overwrite ? LoadFileType.REPLACE_ALL : LoadFileType.KEEP_EXISTING);
                }
                if (destinationTable.isMaterializedView()) {
                    materializedViewUpdateDesc = new MaterializedViewUpdateDesc(destinationTable.getFullyQualifiedName(), false, false, true);
                }
                WriteEntity output = generateTableWriteEntity(dest, destinationTable, partSpec, ltd, dpCtx);
                ctx.getLoadTableOutputMap().put(ltd, output);
                break;
            }
        case QBMetaData.DEST_PARTITION:
            {
                destinationPartition = qbm.getDestPartitionForAlias(dest);
                destinationTable = destinationPartition.getTable();
                destTableIsTransactional = AcidUtils.isTransactionalTable(destinationTable);
                destTableIsFullAcid = AcidUtils.isFullAcidTable(destinationTable);
                checkExternalTable(destinationTable);
                Path partPath = destinationPartition.getDataLocation();
                checkImmutableTable(qb, destinationTable, partPath, true);
                // Previous behavior (HIVE-1707) used to replace the partition's dfs with the table's dfs.
                // The changes in HIVE-19891 appears to no longer support that behavior.
                destinationPath = partPath;
                if (MetaStoreUtils.isArchived(destinationPartition.getTPartition())) {
                    try {
                        String conflictingArchive = ArchiveUtils.conflictingArchiveNameOrNull(db, destinationTable, destinationPartition.getSpec());
                        String message = String.format("Insert conflict with existing archive: %s", conflictingArchive);
                        throw new SemanticException(message);
                    } catch (SemanticException err) {
                        throw err;
                    } catch (HiveException err) {
                        throw new SemanticException(err);
                    }
                }
                isNonNativeTable = destinationTable.isNonNative();
                isMmTable = AcidUtils.isInsertOnlyTable(destinationTable.getParameters());
                AcidUtils.Operation acidOp = AcidUtils.Operation.NOT_ACID;
                // this table_desc does not contain the partitioning columns
                tableDescriptor = Utilities.getTableDesc(destinationTable);
                if (!isNonNativeTable) {
                    if (destTableIsTransactional) {
                        acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                    }
                }
                isDirectInsert = isDirectInsert(destTableIsFullAcid, acidOp);
                acidOperation = acidOp;
                queryTmpdir = getTmpDir(isNonNativeTable, isMmTable, isDirectInsert, destinationPath);
                moveTaskId = getMoveTaskId();
                if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                    Utilities.FILE_OP_LOGGER.trace("create filesink w/DEST_PARTITION specifying " + queryTmpdir + " from " + destinationPath);
                }
                // Add NOT NULL constraint check
                input = genConstraintsPlan(dest, qb, input);
                if (!qb.getIsQuery()) {
                    input = genConversionSelectOperator(dest, qb, input, destinationTable.getDeserializer(), dpCtx, null);
                }
                if (destinationTable.isMaterializedView() && mvRebuildMode == MaterializationRebuildMode.INSERT_OVERWRITE_REBUILD) {
                    // Data organization (DISTRIBUTED, SORTED, CLUSTERED) for materialized view
                    // TODO: We only do this for a full rebuild
                    String sortColsStr = destinationTable.getProperty(Constants.MATERIALIZED_VIEW_SORT_COLUMNS);
                    String distributeColsStr = destinationTable.getProperty(Constants.MATERIALIZED_VIEW_DISTRIBUTE_COLUMNS);
                    if (sortColsStr != null || distributeColsStr != null) {
                        input = genMaterializedViewDataOrgPlan(destinationTable, sortColsStr, distributeColsStr, inputRR, input);
                    }
                } else {
                    // Add sorting/bucketing if needed
                    input = genBucketingSortingDest(dest, input, qb, tableDescriptor, destinationTable, rsCtx);
                }
                idToTableNameMap.put(String.valueOf(destTableId), destinationTable.getTableName());
                currentTableId = destTableId;
                destTableId++;
                if (destTableIsTransactional) {
                    acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                    checkAcidConstraints();
                } else {
                    // Transactional tables can't be list bucketed or have skewed cols
                    lbCtx = constructListBucketingCtx(destinationPartition.getSkewedColNames(), destinationPartition.getSkewedColValues(), destinationPartition.getSkewedColValueLocationMaps(), destinationPartition.isStoredAsSubDirectories());
                }
                try {
                    if (ctx.getExplainConfig() != null) {
                        // For explain plan, txn won't be opened and doesn't make sense to allocate write id
                        writeId = null;
                    } else {
                        if (isMmTable) {
                            writeId = txnMgr.getTableWriteId(destinationTable.getDbName(), destinationTable.getTableName());
                        } else {
                            writeId = (acidOp == Operation.NOT_ACID) ? null : txnMgr.getTableWriteId(destinationTable.getDbName(), destinationTable.getTableName());
                        }
                    }
                } catch (LockException ex) {
                    throw new SemanticException("Failed to allocate write Id", ex);
                }
                ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, destinationPartition.getSpec(), acidOp, writeId);
                if (writeId != null) {
                    ltd.setStmtId(txnMgr.getCurrentStmtId());
                }
                // For the current context for generating File Sink Operator, it is either INSERT INTO or INSERT OVERWRITE.
                // So the next line works.
                boolean isInsertInto = !qb.getParseInfo().isDestToOpTypeInsertOverwrite(dest);
                // For Acid table, Insert Overwrite shouldn't replace the table content. We keep the old
                // deltas and base and leave them up to the cleaner to clean up
                LoadFileType loadType;
                if (isDirectInsert) {
                    loadType = LoadFileType.IGNORE;
                } else if (!isInsertInto && !destTableIsTransactional) {
                    loadType = LoadFileType.REPLACE_ALL;
                } else {
                    loadType = LoadFileType.KEEP_EXISTING;
                }
                ltd.setLoadFileType(loadType);
                ltd.setInsertOverwrite(!isInsertInto);
                ltd.setIsDirectInsert(isDirectInsert);
                ltd.setLbCtx(lbCtx);
                ltd.setMoveTaskId(moveTaskId);
                loadTableWork.add(ltd);
                if (!outputs.add(new WriteEntity(destinationPartition, determineWriteType(ltd, dest)))) {
                    throw new SemanticException(ErrorMsg.OUTPUT_SPECIFIED_MULTIPLE_TIMES.getMsg(destinationTable.getTableName() + "@" + destinationPartition.getName()));
                }
                break;
            }
        case QBMetaData.DEST_LOCAL_FILE:
            isLocal = true;
        // fall through
        case QBMetaData.DEST_DFS_FILE:
            {
                destinationPath = getDestinationFilePath(qbm.getDestFileForAlias(dest), isMmTable);
                // CTAS case: the file output format and serde are defined by the create
                // table command rather than taking the default value
                List<FieldSchema> fieldSchemas = null;
                List<FieldSchema> partitionColumns = null;
                List<String> partitionColumnNames = null;
                List<FieldSchema> sortColumns = null;
                List<String> sortColumnNames = null;
                List<FieldSchema> distributeColumns = null;
                List<String> distributeColumnNames = null;
                List<ColumnInfo> fileSinkColInfos = null;
                List<ColumnInfo> sortColInfos = null;
                List<ColumnInfo> distributeColInfos = null;
                TableName tableName = null;
                Map<String, String> tblProps = null;
                CreateTableDesc tblDesc = qb.getTableDesc();
                CreateMaterializedViewDesc viewDesc = qb.getViewDesc();
                if (tblDesc != null) {
                    fieldSchemas = new ArrayList<>();
                    partitionColumns = new ArrayList<>();
                    partitionColumnNames = tblDesc.getPartColNames();
                    fileSinkColInfos = new ArrayList<>();
                    destTableIsTemporary = tblDesc.isTemporary();
                    destTableIsMaterialization = tblDesc.isMaterialization();
                    tableName = TableName.fromString(tblDesc.getDbTableName(), null, tblDesc.getDatabaseName());
                    tblProps = tblDesc.getTblProps();
                } else if (viewDesc != null) {
                    fieldSchemas = new ArrayList<>();
                    partitionColumns = new ArrayList<>();
                    partitionColumnNames = viewDesc.getPartColNames();
                    sortColumns = new ArrayList<>();
                    sortColumnNames = viewDesc.getSortColNames();
                    distributeColumns = new ArrayList<>();
                    distributeColumnNames = viewDesc.getDistributeColNames();
                    fileSinkColInfos = new ArrayList<>();
                    sortColInfos = new ArrayList<>();
                    distributeColInfos = new ArrayList<>();
                    destTableIsTemporary = false;
                    destTableIsMaterialization = false;
                    tableName = HiveTableName.ofNullableWithNoDefault(viewDesc.getViewName());
                    tblProps = viewDesc.getTblProps();
                }
                destTableIsTransactional = tblProps != null && AcidUtils.isTablePropertyTransactional(tblProps);
                if (destTableIsTransactional) {
                    try {
                        if (ctx.getExplainConfig() != null) {
                            // For explain plan, txn won't be opened and doesn't make sense to allocate write id
                            writeId = 0L;
                        } else {
                            writeId = txnMgr.getTableWriteId(tableName.getDb(), tableName.getTable());
                        }
                    } catch (LockException ex) {
                        throw new SemanticException("Failed to allocate write Id", ex);
                    }
                    if (AcidUtils.isInsertOnlyTable(tblProps, true)) {
                        isMmTable = isMmCreate = true;
                        if (tblDesc != null) {
                            tblDesc.setInitialMmWriteId(writeId);
                        } else {
                            viewDesc.setInitialMmWriteId(writeId);
                        }
                    }
                }
                if (isLocal) {
                    assert !isMmTable;
                    // for local directory - we always write to map-red intermediate
                    // store and then copy to local fs
                    queryTmpdir = ctx.getMRTmpPath();
                } else {
                    // no copy is required. we may want to revisit this policy in future
                    try {
                        Path qPath = FileUtils.makeQualified(destinationPath, conf);
                        queryTmpdir = isMmTable ? qPath : ctx.getTempDirForFinalJobPath(qPath);
                        if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
                            Utilities.FILE_OP_LOGGER.trace("Setting query directory " + queryTmpdir + " from " + destinationPath + " (" + isMmTable + ")");
                        }
                    } catch (Exception e) {
                        throw new SemanticException("Error creating temporary folder on: " + destinationPath, e);
                    }
                }
                // Check for dynamic partitions.
                final String cols, colTypes;
                final boolean isPartitioned;
                if (dpCtx != null) {
                    throw new SemanticException("Dynamic partition context has already been created, this should not happen");
                }
                if (!CollectionUtils.isEmpty(partitionColumnNames)) {
                    ColsAndTypes ct = deriveFileSinkColTypes(inputRR, partitionColumnNames, sortColumnNames, distributeColumnNames, fieldSchemas, partitionColumns, sortColumns, distributeColumns, fileSinkColInfos, sortColInfos, distributeColInfos);
                    cols = ct.cols;
                    colTypes = ct.colTypes;
                    dpCtx = new DynamicPartitionCtx(partitionColumnNames, conf.getVar(HiveConf.ConfVars.DEFAULTPARTITIONNAME), conf.getIntVar(HiveConf.ConfVars.DYNAMICPARTITIONMAXPARTSPERNODE));
                    qbm.setDPCtx(dest, dpCtx);
                    // set the root of the temporary path where dynamic partition columns will populate
                    dpCtx.setRootPath(queryTmpdir);
                    isPartitioned = true;
                } else {
                    ColsAndTypes ct = deriveFileSinkColTypes(inputRR, sortColumnNames, distributeColumnNames, fieldSchemas, sortColumns, distributeColumns, sortColInfos, distributeColInfos);
                    cols = ct.cols;
                    colTypes = ct.colTypes;
                    isPartitioned = false;
                }
                // update the create table descriptor with the resulting schema.
                if (tblDesc != null) {
                    tblDesc.setCols(new ArrayList<>(fieldSchemas));
                    tblDesc.setPartCols(new ArrayList<>(partitionColumns));
                } else if (viewDesc != null) {
                    viewDesc.setSchema(new ArrayList<>(fieldSchemas));
                    viewDesc.setPartCols(new ArrayList<>(partitionColumns));
                    if (viewDesc.isOrganized()) {
                        viewDesc.setSortCols(new ArrayList<>(sortColumns));
                        viewDesc.setDistributeCols(new ArrayList<>(distributeColumns));
                    }
                }
                boolean isDestTempFile = true;
                if (ctx.isMRTmpFileURI(destinationPath.toUri().toString()) == false && ctx.isResultCacheDir(destinationPath) == false) {
                    // not a temp dir and not a result cache dir
                    idToTableNameMap.put(String.valueOf(destTableId), destinationPath.toUri().toString());
                    currentTableId = destTableId;
                    destTableId++;
                    isDestTempFile = false;
                }
                if (tblDesc == null) {
                    if (viewDesc != null) {
                        tableDescriptor = PlanUtils.getTableDesc(viewDesc, cols, colTypes);
                    } else if (qb.getIsQuery()) {
                        Class<? extends Deserializer> serdeClass = LazySimpleSerDe.class;
                        String fileFormat = conf.getResultFileFormat().toString();
                        if (SessionState.get().getIsUsingThriftJDBCBinarySerDe()) {
                            serdeClass = ThriftJDBCBinarySerDe.class;
                            fileFormat = ResultFileFormat.SEQUENCEFILE.toString();
                            // 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 if (fileFormat.equals(PlanUtils.LLAP_OUTPUT_FORMAT_KEY)) {
                            // If this output format is Llap, check to see if Arrow is requested
                            boolean useArrow = HiveConf.getBoolVar(conf, HiveConf.ConfVars.LLAP_OUTPUT_FORMAT_ARROW);
                            serdeClass = useArrow ? ArrowColumnarBatchSerDe.class : LazyBinarySerDe2.class;
                        }
                        tableDescriptor = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes, fileFormat, serdeClass);
                    } else {
                        tableDescriptor = PlanUtils.getDefaultTableDesc(qb.getDirectoryDesc(), cols, colTypes);
                    }
                } else {
                    tableDescriptor = PlanUtils.getTableDesc(tblDesc, cols, colTypes);
                }
                // if available, set location in table desc properties
                if (tblDesc != null && tblDesc.getLocation() != null && tableDescriptor != null && !tableDescriptor.getProperties().containsKey(hive_metastoreConstants.META_TABLE_LOCATION)) {
                    tableDescriptor.getProperties().setProperty(hive_metastoreConstants.META_TABLE_LOCATION, tblDesc.getLocation());
                }
                // We need a specific rowObjectInspector in this case
                try {
                    specificRowObjectInspector = (StructObjectInspector) tableDescriptor.getDeserializer(conf).getObjectInspector();
                } catch (Exception e) {
                    throw new SemanticException(e.getMessage(), e);
                }
                boolean isDfsDir = (destType == QBMetaData.DEST_DFS_FILE);
                try {
                    destinationTable = tblDesc != null ? tblDesc.toTable(conf) : viewDesc != null ? viewDesc.toTable(conf) : null;
                } catch (HiveException e) {
                    throw new SemanticException(e);
                }
                destTableIsFullAcid = AcidUtils.isFullAcidTable(destinationTable);
                // Data organization (DISTRIBUTED, SORTED, CLUSTERED) for materialized view
                if (viewDesc != null && viewDesc.isOrganized()) {
                    input = genMaterializedViewDataOrgPlan(sortColInfos, distributeColInfos, inputRR, input);
                }
                moveTaskId = getMoveTaskId();
                if (isPartitioned) {
                    // Create a SELECT that may reorder the columns if needed
                    RowResolver rowResolver = new RowResolver();
                    List<ExprNodeDesc> columnExprs = new ArrayList<>();
                    List<String> colNames = new ArrayList<>();
                    Map<String, ExprNodeDesc> colExprMap = new HashMap<>();
                    for (int i = 0; i < fileSinkColInfos.size(); i++) {
                        ColumnInfo ci = fileSinkColInfos.get(i);
                        ExprNodeDesc columnExpr = new ExprNodeColumnDesc(ci);
                        String name = getColumnInternalName(i);
                        rowResolver.put("", name, new ColumnInfo(name, columnExpr.getTypeInfo(), "", false));
                        columnExprs.add(columnExpr);
                        colNames.add(name);
                        colExprMap.put(name, columnExpr);
                    }
                    input = putOpInsertMap(OperatorFactory.getAndMakeChild(new SelectDesc(columnExprs, colNames), new RowSchema(rowResolver.getColumnInfos()), input), rowResolver);
                    input.setColumnExprMap(colExprMap);
                    // If this is a partitioned CTAS or MV statement, we are going to create a LoadTableDesc
                    // object. Although the table does not exist in metastore, we will swap the CreateTableTask
                    // and MoveTask resulting from this LoadTable so in this specific case, first we create
                    // the metastore table, then we move and commit the partitions. At least for the time being,
                    // this order needs to be enforced because metastore expects a table to exist before we can
                    // add any partitions to it.
                    isNonNativeTable = tableDescriptor.isNonNative();
                    if (!isNonNativeTable || destinationTable.getStorageHandler().commitInMoveTask()) {
                        AcidUtils.Operation acidOp = AcidUtils.Operation.NOT_ACID;
                        if (destTableIsTransactional) {
                            acidOp = getAcidType(tableDescriptor.getOutputFileFormatClass(), dest, isMmTable);
                            checkAcidConstraints();
                        }
                        // isReplace = false in case concurrent operation is executed
                        ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, dpCtx, acidOp, false, writeId);
                        if (writeId != null) {
                            ltd.setStmtId(txnMgr.getCurrentStmtId());
                        }
                        ltd.setLoadFileType(LoadFileType.KEEP_EXISTING);
                        ltd.setInsertOverwrite(false);
                        loadTableWork.add(ltd);
                    } else {
                        // This is a non-native table.
                        // We need to set stats as inaccurate.
                        setStatsForNonNativeTable(tableDescriptor.getDbName(), tableDescriptor.getTableName());
                        ltd = new LoadTableDesc(queryTmpdir, tableDescriptor, dpCtx.getPartSpec());
                        ltd.setInsertOverwrite(false);
                        ltd.setLoadFileType(LoadFileType.KEEP_EXISTING);
                    }
                    ltd.setMoveTaskId(moveTaskId);
                    ltd.setMdTable(destinationTable);
                    WriteEntity output = generateTableWriteEntity(dest, destinationTable, dpCtx.getPartSpec(), ltd, dpCtx);
                    ctx.getLoadTableOutputMap().put(ltd, output);
                } else {
                    // Create LFD even for MM CTAS - it's a no-op move, but it still seems to be used for stats.
                    LoadFileDesc loadFileDesc = new LoadFileDesc(tblDesc, viewDesc, queryTmpdir, destinationPath, isDfsDir, cols, colTypes, // there is a change here - prev version had 'transactional', one before 'acid'
                    destTableIsFullAcid ? Operation.INSERT : Operation.NOT_ACID, isMmCreate);
                    loadFileDesc.setMoveTaskId(moveTaskId);
                    loadFileWork.add(loadFileDesc);
                    try {
                        Path qualifiedPath = destinationPath.getFileSystem(conf).makeQualified(destinationPath);
                        if (!outputs.add(new WriteEntity(qualifiedPath, !isDfsDir, isDestTempFile))) {
                            throw new SemanticException(ErrorMsg.OUTPUT_SPECIFIED_MULTIPLE_TIMES.getMsg(destinationPath.toUri().toString()));
                        }
                    } catch (IOException ex) {
                        throw new SemanticException("Error while getting the full qualified path for the given directory: " + ex.getMessage());
                    }
                }
                break;
            }
        default:
            throw new SemanticException("Unknown destination type: " + destType);
    }
    inputRR = opParseCtx.get(input).getRowResolver();
    List<ColumnInfo> vecCol = new ArrayList<ColumnInfo>();
    if (updating(dest) || deleting(dest)) {
        vecCol.add(new ColumnInfo(VirtualColumn.ROWID.getName(), VirtualColumn.ROWID.getTypeInfo(), "", true));
    } else {
        try {
            // If we already have a specific inspector (view or directory as a target) use that
            // Otherwise use the table deserializer to get the inspector
            StructObjectInspector rowObjectInspector = specificRowObjectInspector != null ? specificRowObjectInspector : (StructObjectInspector) destinationTable.getDeserializer().getObjectInspector();
            List<? extends StructField> fields = rowObjectInspector.getAllStructFieldRefs();
            for (StructField field : fields) {
                vecCol.add(new ColumnInfo(field.getFieldName(), TypeInfoUtils.getTypeInfoFromObjectInspector(field.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 = (destinationTable == null || !((destinationTable.getNumBuckets() > 0) || (destinationTable.getSortCols() != null && destinationTable.getSortCols().size() > 0)));
    // If this table is working with ACID semantics, turn off merging
    canBeMerged &= !destTableIsFullAcid;
    // Generate the partition columns from the parent input
    if (destType == QBMetaData.DEST_TABLE || destType == QBMetaData.DEST_PARTITION) {
        genPartnCols(dest, input, qb, tableDescriptor, destinationTable, rsCtx);
    }
    FileSinkDesc fileSinkDesc = createFileSinkDesc(dest, tableDescriptor, destinationPartition, // this was 1/4 acid
    destinationPath, // this was 1/4 acid
    currentTableId, // this was 1/4 acid
    destTableIsFullAcid, // this was 1/4 acid
    destTableIsTemporary, destTableIsMaterialization, queryTmpdir, rsCtx, dpCtx, lbCtx, fsRS, canBeMerged, destinationTable, writeId, isMmCreate, destType, qb, isDirectInsert, acidOperation, moveTaskId);
    if (isMmCreate) {
        // Add FSD so that the LoadTask compilation could fix up its path to avoid the move.
        if (tableDesc != null) {
            tableDesc.setWriter(fileSinkDesc);
        } else {
            createVwDesc.setWriter(fileSinkDesc);
        }
    }
    if (fileSinkDesc.getInsertOverwrite()) {
        if (ltd != null) {
            ltd.setInsertOverwrite(true);
        }
    }
    if (null != tableDescriptor && useBatchingSerializer(tableDescriptor.getSerdeClassName())) {
        fileSinkDesc.setIsUsingBatchingSerDe(true);
    } else {
        fileSinkDesc.setIsUsingBatchingSerDe(false);
    }
    Operator output = putOpInsertMap(OperatorFactory.getAndMakeChild(fileSinkDesc, fsRS, input), inputRR);
    // in case of a merge statement.
    if (!isDirectInsert || acidOperation == AcidUtils.Operation.INSERT) {
        handleLineage(ltd, output);
    }
    setWriteIdForSurrogateKeys(ltd, input);
    LOG.debug("Created FileSink Plan for clause: {}dest_path: {} row schema: {}", dest, destinationPath, inputRR);
    FileSinkOperator fso = (FileSinkOperator) output;
    fso.getConf().setTable(destinationTable);
    // and it is an insert overwrite or insert into table
    if (conf.getBoolVar(ConfVars.HIVESTATSAUTOGATHER) && conf.getBoolVar(ConfVars.HIVESTATSCOLAUTOGATHER) && destinationTable != null && (!destinationTable.isNonNative() || destinationTable.getStorageHandler().commitInMoveTask()) && !destTableIsTemporary && !destTableIsMaterialization && ColumnStatsAutoGatherContext.canRunAutogatherStats(fso)) {
        if (destType == QBMetaData.DEST_TABLE) {
            genAutoColumnStatsGatheringPipeline(destinationTable, partSpec, input, qb.getParseInfo().isInsertIntoTable(destinationTable.getDbName(), destinationTable.getTableName()), false);
        } else if (destType == QBMetaData.DEST_PARTITION) {
            genAutoColumnStatsGatheringPipeline(destinationTable, destinationPartition.getSpec(), input, qb.getParseInfo().isInsertIntoTable(destinationTable.getDbName(), destinationTable.getTableName()), false);
        } else if (destType == QBMetaData.DEST_LOCAL_FILE || destType == QBMetaData.DEST_DFS_FILE) {
            // CTAS or CMV statement
            genAutoColumnStatsGatheringPipeline(destinationTable, null, input, false, true);
        }
    }
    return output;
}

Example 15 with LoadTableDesc

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

the class ImportSemanticAnalyzer method addSinglePartition.

private static Task<?> addSinglePartition(ImportTableDesc tblDesc, Table table, Warehouse wh, AlterTableAddPartitionDesc addPartitionDesc, ReplicationSpec replicationSpec, EximUtil.SemanticAnalyzerWrapperContext x, Long writeId, int stmtId, boolean isReplication, String dumpRoot, ReplicationMetricCollector metricCollector) throws MetaException, IOException, HiveException {
    AlterTableAddPartitionDesc.PartitionDesc partSpec = addPartitionDesc.getPartitions().get(0);
    boolean isSkipTrash = false;
    boolean needRecycle = false;
    if (shouldSkipDataCopyInReplScope(tblDesc, replicationSpec) || (tblDesc.isExternal() && tblDesc.getLocation() == null)) {
        x.getLOG().debug("Adding AddPart and skipped data copy for partition " + partSpecToString(partSpec.getPartSpec()));
        // addPartitionDesc already has the right partition location
        @SuppressWarnings("unchecked") Task<?> addPartTask = TaskFactory.get(new DDLWork(x.getInputs(), x.getOutputs(), addPartitionDesc, isReplication, dumpRoot, metricCollector), x.getConf());
        return addPartTask;
    } else {
        String srcLocation = partSpec.getLocation();
        if (replicationSpec.isInReplicationScope() && !ReplicationSpec.Type.IMPORT.equals(replicationSpec.getReplSpecType())) {
            Path partLocation = new Path(partSpec.getLocation());
            Path dataDirBase = partLocation.getParent();
            String bucketDir = partLocation.getName();
            for (int i = 1; i < partSpec.getPartSpec().size(); i++) {
                bucketDir = dataDirBase.getName() + File.separator + bucketDir;
                dataDirBase = dataDirBase.getParent();
            }
            String relativePartDataPath = EximUtil.DATA_PATH_NAME + File.separator + bucketDir;
            srcLocation = new Path(dataDirBase, relativePartDataPath).toString();
        }
        fixLocationInPartSpec(tblDesc, table, wh, replicationSpec, partSpec, x);
        x.getLOG().debug("adding dependent CopyWork/AddPart/MoveWork for partition " + partSpecToString(partSpec.getPartSpec()) + " with source location: " + srcLocation);
        Path tgtLocation = new Path(partSpec.getLocation());
        LoadFileType loadFileType;
        Path destPath;
        if (replicationSpec.isInReplicationScope()) {
            loadFileType = LoadFileType.IGNORE;
            destPath = tgtLocation;
            isSkipTrash = MetaStoreUtils.isSkipTrash(table.getParameters());
            if (table.isTemporary()) {
                needRecycle = false;
            } else {
                org.apache.hadoop.hive.metastore.api.Database db = x.getHive().getDatabase(table.getDbName());
                needRecycle = db != null && ReplChangeManager.shouldEnableCm(db, table.getTTable());
            }
        } else {
            loadFileType = replicationSpec.isReplace() ? LoadFileType.REPLACE_ALL : LoadFileType.OVERWRITE_EXISTING;
            // Replication scope the write id will be invalid
            boolean useStagingDirectory = !AcidUtils.isTransactionalTable(table.getParameters()) || replicationSpec.isInReplicationScope();
            destPath = useStagingDirectory ? x.getCtx().getExternalTmpPath(tgtLocation) : new Path(tgtLocation, AcidUtils.deltaSubdir(writeId, writeId, stmtId));
        }
        Path moveTaskSrc = !AcidUtils.isTransactionalTable(table.getParameters()) || replicationSpec.isInReplicationScope() ? destPath : tgtLocation;
        if (Utilities.FILE_OP_LOGGER.isTraceEnabled()) {
            Utilities.FILE_OP_LOGGER.trace("adding import work for partition with source location: " + srcLocation + "; target: " + tgtLocation + "; copy dest " + destPath + "; mm " + writeId + " for " + partSpecToString(partSpec.getPartSpec()) + ": " + (AcidUtils.isFullAcidTable(table) ? "acid" : (AcidUtils.isInsertOnlyTable(table) ? "mm" : "flat")));
        }
        Task<?> copyTask = null;
        if (replicationSpec.isInReplicationScope()) {
            boolean copyAtLoad = x.getConf().getBoolVar(HiveConf.ConfVars.REPL_RUN_DATA_COPY_TASKS_ON_TARGET);
            copyTask = ReplCopyTask.getLoadCopyTask(replicationSpec, new Path(srcLocation), destPath, x.getConf(), isSkipTrash, needRecycle, copyAtLoad, dumpRoot, metricCollector);
        } else {
            copyTask = TaskFactory.get(new CopyWork(new Path(srcLocation), destPath, false, dumpRoot, metricCollector, isReplication));
        }
        Task<?> addPartTask = null;
        if (x.getEventType() != DumpType.EVENT_COMMIT_TXN) {
            // During replication, by the time we are applying commit transaction event, we expect
            // the partition/s to be already added or altered by previous events. So no need to
            // create add partition event again.
            addPartTask = TaskFactory.get(new DDLWork(x.getInputs(), x.getOutputs(), addPartitionDesc, isReplication, dumpRoot, metricCollector), x.getConf());
        }
        MoveWork moveWork = new MoveWork(x.getInputs(), x.getOutputs(), null, null, false, dumpRoot, metricCollector, isReplication);
        // See setLoadFileType and setIsAcidIow calls elsewhere for an example.
        if (replicationSpec.isInReplicationScope() && AcidUtils.isTransactionalTable(tblDesc.getTblProps())) {
            LoadMultiFilesDesc loadFilesWork = new LoadMultiFilesDesc(Collections.singletonList(destPath), Collections.singletonList(tgtLocation), true, null, null);
            moveWork.setMultiFilesDesc(loadFilesWork);
            moveWork.setNeedCleanTarget(replicationSpec.isReplace());
        } else {
            LoadTableDesc loadTableWork = new LoadTableDesc(moveTaskSrc, Utilities.getTableDesc(table), partSpec.getPartSpec(), loadFileType, writeId);
            loadTableWork.setStmtId(stmtId);
            loadTableWork.setInheritTableSpecs(false);
            moveWork.setLoadTableWork(loadTableWork);
        }
        if (loadFileType == LoadFileType.IGNORE) {
            // update which is again done in load operations as part of move task.
            if (x.getEventType() == DumpType.EVENT_INSERT) {
                copyTask.addDependentTask(TaskFactory.get(moveWork, x.getConf()));
            } else {
                if (addPartTask != null) {
                    copyTask.addDependentTask(addPartTask);
                }
            }
            return copyTask;
        }
        Task<?> loadPartTask = TaskFactory.get(moveWork, x.getConf());
        copyTask.addDependentTask(loadPartTask);
        if (addPartTask != null) {
            addPartTask.addDependentTask(loadPartTask);
            x.getTasks().add(copyTask);
            return addPartTask;
        }
        return copyTask;
    }
}