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

Example 16 with DriverContext

use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.

the class ExecDriver method execute.

/**
 * Execute a query plan using Hadoop.
 */
@SuppressWarnings({ "deprecation", "unchecked" })
@Override
public int execute(DriverContext driverContext) {
    IOPrepareCache ioPrepareCache = IOPrepareCache.get();
    ioPrepareCache.clear();
    boolean success = true;
    Context ctx = driverContext.getCtx();
    boolean ctxCreated = false;
    Path emptyScratchDir;
    JobClient jc = null;
    if (driverContext.isShutdown()) {
        LOG.warn("Task was cancelled");
        return 5;
    }
    MapWork mWork = work.getMapWork();
    ReduceWork rWork = work.getReduceWork();
    try {
        if (ctx == null) {
            ctx = new Context(job);
            ctxCreated = true;
        }
        emptyScratchDir = ctx.getMRTmpPath();
        FileSystem fs = emptyScratchDir.getFileSystem(job);
        fs.mkdirs(emptyScratchDir);
    } catch (IOException e) {
        e.printStackTrace();
        console.printError("Error launching map-reduce job", "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
        return 5;
    }
    HiveFileFormatUtils.prepareJobOutput(job);
    // See the javadoc on HiveOutputFormatImpl and HadoopShims.prepareJobOutput()
    job.setOutputFormat(HiveOutputFormatImpl.class);
    job.setMapRunnerClass(ExecMapRunner.class);
    job.setMapperClass(ExecMapper.class);
    job.setMapOutputKeyClass(HiveKey.class);
    job.setMapOutputValueClass(BytesWritable.class);
    try {
        String partitioner = HiveConf.getVar(job, ConfVars.HIVEPARTITIONER);
        job.setPartitionerClass(JavaUtils.loadClass(partitioner));
    } catch (ClassNotFoundException e) {
        throw new RuntimeException(e.getMessage(), e);
    }
    propagateSplitSettings(job, mWork);
    job.setNumReduceTasks(rWork != null ? rWork.getNumReduceTasks().intValue() : 0);
    job.setReducerClass(ExecReducer.class);
    // set input format information if necessary
    setInputAttributes(job);
    // Turn on speculative execution for reducers
    boolean useSpeculativeExecReducers = HiveConf.getBoolVar(job, HiveConf.ConfVars.HIVESPECULATIVEEXECREDUCERS);
    job.setBoolean(MRJobConfig.REDUCE_SPECULATIVE, useSpeculativeExecReducers);
    String inpFormat = HiveConf.getVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT);
    if (mWork.isUseBucketizedHiveInputFormat()) {
        inpFormat = BucketizedHiveInputFormat.class.getName();
    }
    LOG.info("Using " + inpFormat);
    try {
        job.setInputFormat(JavaUtils.loadClass(inpFormat));
    } catch (ClassNotFoundException e) {
        throw new RuntimeException(e.getMessage(), e);
    }
    // No-Op - we don't really write anything here ..
    job.setOutputKeyClass(Text.class);
    job.setOutputValueClass(Text.class);
    int returnVal = 0;
    boolean noName = StringUtils.isEmpty(job.get(MRJobConfig.JOB_NAME));
    if (noName) {
        // This is for a special case to ensure unit tests pass
        job.set(MRJobConfig.JOB_NAME, "JOB" + Utilities.randGen.nextInt());
    }
    try {
        MapredLocalWork localwork = mWork.getMapRedLocalWork();
        if (localwork != null && localwork.hasStagedAlias()) {
            if (!ShimLoader.getHadoopShims().isLocalMode(job)) {
                Path localPath = localwork.getTmpPath();
                Path hdfsPath = mWork.getTmpHDFSPath();
                FileSystem hdfs = hdfsPath.getFileSystem(job);
                FileSystem localFS = localPath.getFileSystem(job);
                FileStatus[] hashtableFiles = localFS.listStatus(localPath);
                int fileNumber = hashtableFiles.length;
                String[] fileNames = new String[fileNumber];
                for (int i = 0; i < fileNumber; i++) {
                    fileNames[i] = hashtableFiles[i].getPath().getName();
                }
                // package and compress all the hashtable files to an archive file
                String stageId = this.getId();
                String archiveFileName = Utilities.generateTarFileName(stageId);
                localwork.setStageID(stageId);
                CompressionUtils.tar(localPath.toUri().getPath(), fileNames, archiveFileName);
                Path archivePath = Utilities.generateTarPath(localPath, stageId);
                LOG.info("Archive " + hashtableFiles.length + " hash table files to " + archivePath);
                // upload archive file to hdfs
                Path hdfsFilePath = Utilities.generateTarPath(hdfsPath, stageId);
                short replication = (short) job.getInt("mapred.submit.replication", 10);
                hdfs.copyFromLocalFile(archivePath, hdfsFilePath);
                hdfs.setReplication(hdfsFilePath, replication);
                LOG.info("Upload 1 archive file  from" + archivePath + " to: " + hdfsFilePath);
                // add the archive file to distributed cache
                DistributedCache.createSymlink(job);
                DistributedCache.addCacheArchive(hdfsFilePath.toUri(), job);
                LOG.info("Add 1 archive file to distributed cache. Archive file: " + hdfsFilePath.toUri());
            }
        }
        work.configureJobConf(job);
        List<Path> inputPaths = Utilities.getInputPaths(job, mWork, emptyScratchDir, ctx, false);
        Utilities.setInputPaths(job, inputPaths);
        Utilities.setMapRedWork(job, work, ctx.getMRTmpPath());
        if (mWork.getSamplingType() > 0 && rWork != null && job.getNumReduceTasks() > 1) {
            try {
                handleSampling(ctx, mWork, job);
                job.setPartitionerClass(HiveTotalOrderPartitioner.class);
            } catch (IllegalStateException e) {
                console.printInfo("Not enough sampling data.. Rolling back to single reducer task");
                rWork.setNumReduceTasks(1);
                job.setNumReduceTasks(1);
            } catch (Exception e) {
                LOG.error("Sampling error", e);
                console.printError(e.toString(), "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
                rWork.setNumReduceTasks(1);
                job.setNumReduceTasks(1);
            }
        }
        jc = new JobClient(job);
        // make this client wait if job tracker is not behaving well.
        Throttle.checkJobTracker(job, LOG);
        if (mWork.isGatheringStats() || (rWork != null && rWork.isGatheringStats())) {
            // initialize stats publishing table
            StatsPublisher statsPublisher;
            StatsFactory factory = StatsFactory.newFactory(job);
            if (factory != null) {
                statsPublisher = factory.getStatsPublisher();
                List<String> statsTmpDir = Utilities.getStatsTmpDirs(mWork, job);
                if (rWork != null) {
                    statsTmpDir.addAll(Utilities.getStatsTmpDirs(rWork, job));
                }
                StatsCollectionContext sc = new StatsCollectionContext(job);
                sc.setStatsTmpDirs(statsTmpDir);
                if (!statsPublisher.init(sc)) {
                    // creating stats table if not exists
                    if (HiveConf.getBoolVar(job, HiveConf.ConfVars.HIVE_STATS_RELIABLE)) {
                        throw new HiveException(ErrorMsg.STATSPUBLISHER_INITIALIZATION_ERROR.getErrorCodedMsg());
                    }
                }
            }
        }
        Utilities.createTmpDirs(job, mWork);
        Utilities.createTmpDirs(job, rWork);
        SessionState ss = SessionState.get();
        // TODO: why is there a TezSession in MR ExecDriver?
        if (ss != null && HiveConf.getVar(job, ConfVars.HIVE_EXECUTION_ENGINE).equals("tez")) {
            // TODO: this is the only place that uses keepTmpDir. Why?
            TezSessionPoolManager.closeIfNotDefault(ss.getTezSession(), true);
        }
        HiveConfUtil.updateJobCredentialProviders(job);
        // Finally SUBMIT the JOB!
        if (driverContext.isShutdown()) {
            LOG.warn("Task was cancelled");
            return 5;
        }
        rj = jc.submitJob(job);
        if (driverContext.isShutdown()) {
            LOG.warn("Task was cancelled");
            killJob();
            return 5;
        }
        this.jobID = rj.getJobID();
        updateStatusInQueryDisplay();
        returnVal = jobExecHelper.progress(rj, jc, ctx);
        success = (returnVal == 0);
    } catch (Exception e) {
        e.printStackTrace();
        setException(e);
        String mesg = " with exception '" + Utilities.getNameMessage(e) + "'";
        if (rj != null) {
            mesg = "Ended Job = " + rj.getJobID() + mesg;
        } else {
            mesg = "Job Submission failed" + mesg;
        }
        // Has to use full name to make sure it does not conflict with
        // org.apache.commons.lang.StringUtils
        console.printError(mesg, "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
        success = false;
        returnVal = 1;
    } finally {
        Utilities.clearWork(job);
        try {
            if (ctxCreated) {
                ctx.clear();
            }
            if (rj != null) {
                if (returnVal != 0) {
                    killJob();
                }
                jobID = rj.getID().toString();
            }
            if (jc != null) {
                jc.close();
            }
        } catch (Exception e) {
            LOG.warn("Failed while cleaning up ", e);
        } finally {
            HadoopJobExecHelper.runningJobs.remove(rj);
        }
    }
    // get the list of Dynamic partition paths
    try {
        if (rj != null) {
            if (mWork.getAliasToWork() != null) {
                for (Operator<? extends OperatorDesc> op : mWork.getAliasToWork().values()) {
                    op.jobClose(job, success);
                }
            }
            if (rWork != null) {
                rWork.getReducer().jobClose(job, success);
            }
        }
    } catch (Exception e) {
        // jobClose needs to execute successfully otherwise fail task
        if (success) {
            setException(e);
            success = false;
            returnVal = 3;
            String mesg = "Job Commit failed with exception '" + Utilities.getNameMessage(e) + "'";
            console.printError(mesg, "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
        }
    }
    return (returnVal);
}

Example 17 with DriverContext

use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.

the class MapredLocalTask method executeInChildVM.

public int executeInChildVM(DriverContext driverContext) {
    // execute in child jvm
    try {
        // generate the cmd line to run in the child jvm
        Context ctx = driverContext.getCtx();
        String hiveJar = conf.getJar();
        String hadoopExec = conf.getVar(HiveConf.ConfVars.HADOOPBIN);
        conf.setVar(ConfVars.HIVEADDEDJARS, Utilities.getResourceFiles(conf, SessionState.ResourceType.JAR));
        // write out the plan to a local file
        Path planPath = new Path(ctx.getLocalTmpPath(), "plan.xml");
        MapredLocalWork plan = getWork();
        LOG.info("Generating plan file " + planPath.toString());
        OutputStream out = null;
        try {
            out = FileSystem.getLocal(conf).create(planPath);
            SerializationUtilities.serializePlan(plan, out);
            out.close();
            out = null;
        } finally {
            IOUtils.closeQuietly(out);
        }
        String isSilent = "true".equalsIgnoreCase(System.getProperty("test.silent")) ? "-nolog" : "";
        String libJars = ExecDriver.getResource(conf, ResourceType.JAR);
        String libJarsOption = StringUtils.isEmpty(libJars) ? " " : " -libjars " + libJars + " ";
        String jarCmd = hiveJar + " " + ExecDriver.class.getName() + libJarsOption;
        String hiveConfArgs = ExecDriver.generateCmdLine(conf, ctx);
        String cmdLine = hadoopExec + " jar " + jarCmd + " -localtask -plan " + planPath.toString() + " " + isSilent + " " + hiveConfArgs;
        String workDir = (new File(".")).getCanonicalPath();
        String files = Utilities.getResourceFiles(conf, SessionState.ResourceType.FILE);
        if (!files.isEmpty()) {
            cmdLine = cmdLine + " -files " + files;
            workDir = ctx.getLocalTmpPath().toUri().getPath();
            if (!(new File(workDir)).mkdir()) {
                throw new IOException("Cannot create tmp working dir: " + workDir);
            }
            for (String f : StringUtils.split(files, ',')) {
                Path p = new Path(f);
                String target = p.toUri().getPath();
                String link = workDir + Path.SEPARATOR + p.getName();
                if (FileUtil.symLink(target, link) != 0) {
                    throw new IOException("Cannot link to added file: " + target + " from: " + link);
                }
            }
        }
        // Inherit Java system variables
        String hadoopOpts;
        StringBuilder sb = new StringBuilder();
        Properties p = System.getProperties();
        for (String element : HIVE_SYS_PROP) {
            if (p.containsKey(element)) {
                sb.append(" -D" + element + "=" + p.getProperty(element));
            }
        }
        hadoopOpts = sb.toString();
        // Inherit the environment variables
        String[] env;
        Map<String, String> variables = new HashMap<String, String>(System.getenv());
        // The user can specify the hadoop memory
        // if ("local".equals(conf.getVar(HiveConf.ConfVars.HADOOPJT))) {
        // if we are running in local mode - then the amount of memory used
        // by the child jvm can no longer default to the memory used by the
        // parent jvm
        // int hadoopMem = conf.getIntVar(HiveConf.ConfVars.HIVEHADOOPMAXMEM);
        int hadoopMem = conf.getIntVar(HiveConf.ConfVars.HIVEHADOOPMAXMEM);
        if (hadoopMem == 0) {
            // remove env var that would default child jvm to use parent's memory
            // as default. child jvm would use default memory for a hadoop client
            variables.remove(HADOOP_MEM_KEY);
        } else {
            // user specified the memory for local mode hadoop run
            console.printInfo(" set heap size\t" + hadoopMem + "MB");
            variables.put(HADOOP_MEM_KEY, String.valueOf(hadoopMem));
        }
        // } else {
        // nothing to do - we are not running in local mode - only submitting
        // the job via a child process. in this case it's appropriate that the
        // child jvm use the same memory as the parent jvm
        // }
        // Set HADOOP_USER_NAME env variable for child process, so that
        // it also runs with hadoop permissions for the user the job is running as
        // This will be used by hadoop only in unsecure(/non kerberos) mode
        String endUserName = Utils.getUGI().getShortUserName();
        LOG.debug("setting HADOOP_USER_NAME\t" + endUserName);
        variables.put("HADOOP_USER_NAME", endUserName);
        if (variables.containsKey(HADOOP_OPTS_KEY)) {
            variables.put(HADOOP_OPTS_KEY, variables.get(HADOOP_OPTS_KEY) + hadoopOpts);
        } else {
            variables.put(HADOOP_OPTS_KEY, hadoopOpts);
        }
        // Hiveserver2 using "-hiveconf hive.hadoop.classpath=%HIVE_LIB%". This is to combine path(s).
        if (HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_HADOOP_CLASSPATH) != null) {
            if (variables.containsKey("HADOOP_CLASSPATH")) {
                variables.put("HADOOP_CLASSPATH", variables.get("HADOOP_CLASSPATH") + ";" + HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_HADOOP_CLASSPATH));
            } else {
                variables.put("HADOOP_CLASSPATH", HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_HADOOP_CLASSPATH));
            }
        }
        if (variables.containsKey(MapRedTask.HIVE_DEBUG_RECURSIVE)) {
            MapRedTask.configureDebugVariablesForChildJVM(variables);
        }
        if (UserGroupInformation.isSecurityEnabled() && UserGroupInformation.isLoginKeytabBased()) {
            // If kerberos security is enabled, and HS2 doAs is enabled,
            // then additional params need to be set so that the command is run as
            // intended user
            secureDoAs = new SecureCmdDoAs(conf);
            secureDoAs.addEnv(variables);
        }
        // have different settings from those of HiveServer2.
        if (variables.containsKey(HIVE_LOCAL_TASK_CHILD_OPTS_KEY)) {
            String childOpts = variables.get(HIVE_LOCAL_TASK_CHILD_OPTS_KEY);
            if (childOpts == null) {
                childOpts = "";
            }
            String clientOpts = variables.put(HADOOP_CLIENT_OPTS, childOpts);
            String tmp = variables.get(HADOOP_OPTS_KEY);
            if (tmp != null && !StringUtils.isBlank(clientOpts)) {
                tmp = tmp.replace(clientOpts, childOpts);
                variables.put(HADOOP_OPTS_KEY, tmp);
            }
        }
        env = new String[variables.size()];
        int pos = 0;
        for (Map.Entry<String, String> entry : variables.entrySet()) {
            String name = entry.getKey();
            String value = entry.getValue();
            env[pos++] = name + "=" + value;
            LOG.debug("Setting env: " + name + "=" + LogUtils.maskIfPassword(name, value));
        }
        LOG.info("Executing: " + cmdLine);
        // Run ExecDriver in another JVM
        executor = Runtime.getRuntime().exec(cmdLine, env, new File(workDir));
        final LogRedirector.LogSourceCallback callback = () -> {
            return executor.isAlive();
        };
        LogRedirector.redirect(Thread.currentThread().getName() + "-LocalTask-" + getName() + "-stdout", new LogRedirector(executor.getInputStream(), LOG, callback));
        LogRedirector.redirect(Thread.currentThread().getName() + "-LocalTask-" + getName() + "-stderr", new LogRedirector(executor.getErrorStream(), LOG, callback));
        CachingPrintStream errPrintStream = new CachingPrintStream(System.err);
        StreamPrinter outPrinter = new StreamPrinter(executor.getInputStream(), null, System.out);
        StreamPrinter errPrinter = new StreamPrinter(executor.getErrorStream(), null, errPrintStream);
        outPrinter.start();
        errPrinter.start();
        int exitVal = jobExecHelper.progressLocal(executor, getId());
        // wait for stream threads to finish
        outPrinter.join();
        errPrinter.join();
        if (exitVal != 0) {
            LOG.error("Execution failed with exit status: " + exitVal);
            if (SessionState.get() != null) {
                SessionState.get().addLocalMapRedErrors(getId(), errPrintStream.getOutput());
            }
        } else {
            LOG.info("Execution completed successfully");
        }
        return exitVal;
    } catch (Exception e) {
        LOG.error("Exception: ", e);
        return (1);
    } finally {
        if (secureDoAs != null) {
            secureDoAs.close();
        }
    }
}

Example 18 with DriverContext

use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.

the class DDLTask method truncateTable.

private int truncateTable(Hive db, TruncateTableDesc truncateTableDesc) throws HiveException {
    if (truncateTableDesc.getColumnIndexes() != null) {
        ColumnTruncateWork truncateWork = new ColumnTruncateWork(truncateTableDesc.getColumnIndexes(), truncateTableDesc.getInputDir(), truncateTableDesc.getOutputDir());
        truncateWork.setListBucketingCtx(truncateTableDesc.getLbCtx());
        truncateWork.setMapperCannotSpanPartns(true);
        DriverContext driverCxt = new DriverContext();
        ColumnTruncateTask taskExec = new ColumnTruncateTask();
        taskExec.initialize(queryState, null, driverCxt, null);
        taskExec.setWork(truncateWork);
        taskExec.setQueryPlan(this.getQueryPlan());
        subtask = taskExec;
        int ret = taskExec.execute(driverCxt);
        if (subtask.getException() != null) {
            setException(subtask.getException());
        }
        return ret;
    }
    String tableName = truncateTableDesc.getTableName();
    Map<String, String> partSpec = truncateTableDesc.getPartSpec();
    if (!allowOperationInReplicationScope(db, tableName, partSpec, truncateTableDesc.getReplicationSpec())) {
        // or the existing table is newer than our update.
        if (LOG.isDebugEnabled()) {
            LOG.debug("DDLTask: Truncate Table/Partition is skipped as table {} / partition {} is newer than update", tableName, (partSpec == null) ? "null" : FileUtils.makePartName(new ArrayList<>(partSpec.keySet()), new ArrayList<>(partSpec.values())));
        }
        return 0;
    }
    try {
        db.truncateTable(tableName, partSpec);
    } catch (Exception e) {
        throw new HiveException(e, ErrorMsg.GENERIC_ERROR);
    }
    return 0;
}

Example 19 with DriverContext

use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.

the class DDLTask method mergeFiles.

/**
 * First, make sure the source table/partition is not
 * archived/indexes/non-rcfile. If either of these is true, throw an
 * exception.
 *
 * The way how it does the merge is to create a BlockMergeTask from the
 * mergeFilesDesc.
 *
 * @param db
 * @param mergeFilesDesc
 * @return
 * @throws HiveException
 */
private int mergeFiles(Hive db, AlterTablePartMergeFilesDesc mergeFilesDesc, DriverContext driverContext) throws HiveException {
    ListBucketingCtx lbCtx = mergeFilesDesc.getLbCtx();
    boolean lbatc = lbCtx == null ? false : lbCtx.isSkewedStoredAsDir();
    int lbd = lbCtx == null ? 0 : lbCtx.calculateListBucketingLevel();
    // merge work only needs input and output.
    MergeFileWork mergeWork = new MergeFileWork(mergeFilesDesc.getInputDir(), mergeFilesDesc.getOutputDir(), mergeFilesDesc.getInputFormatClass().getName(), mergeFilesDesc.getTableDesc());
    LinkedHashMap<Path, ArrayList<String>> pathToAliases = new LinkedHashMap<>();
    ArrayList<String> inputDirstr = new ArrayList<String>(1);
    inputDirstr.add(mergeFilesDesc.getInputDir().toString());
    pathToAliases.put(mergeFilesDesc.getInputDir().get(0), inputDirstr);
    mergeWork.setPathToAliases(pathToAliases);
    mergeWork.setListBucketingCtx(mergeFilesDesc.getLbCtx());
    mergeWork.resolveConcatenateMerge(db.getConf());
    mergeWork.setMapperCannotSpanPartns(true);
    mergeWork.setSourceTableInputFormat(mergeFilesDesc.getInputFormatClass().getName());
    final FileMergeDesc fmd;
    if (mergeFilesDesc.getInputFormatClass().equals(RCFileInputFormat.class)) {
        fmd = new RCFileMergeDesc();
    } else {
        // safe to assume else is ORC as semantic analyzer will check for RC/ORC
        fmd = new OrcFileMergeDesc();
    }
    fmd.setDpCtx(null);
    fmd.setHasDynamicPartitions(false);
    fmd.setListBucketingAlterTableConcatenate(lbatc);
    fmd.setListBucketingDepth(lbd);
    fmd.setOutputPath(mergeFilesDesc.getOutputDir());
    CompilationOpContext opContext = driverContext.getCtx().getOpContext();
    Operator<? extends OperatorDesc> mergeOp = OperatorFactory.get(opContext, fmd);
    LinkedHashMap<String, Operator<? extends OperatorDesc>> aliasToWork = new LinkedHashMap<String, Operator<? extends OperatorDesc>>();
    aliasToWork.put(mergeFilesDesc.getInputDir().toString(), mergeOp);
    mergeWork.setAliasToWork(aliasToWork);
    DriverContext driverCxt = new DriverContext();
    Task<?> task;
    if (conf.getVar(ConfVars.HIVE_EXECUTION_ENGINE).equals("tez")) {
        TezWork tezWork = new TezWork(queryState.getQueryId(), conf);
        mergeWork.setName("File Merge");
        tezWork.add(mergeWork);
        task = new TezTask();
        ((TezTask) task).setWork(tezWork);
    } else {
        task = new MergeFileTask();
        ((MergeFileTask) task).setWork(mergeWork);
    }
    // initialize the task and execute
    task.initialize(queryState, getQueryPlan(), driverCxt, opContext);
    subtask = task;
    int ret = task.execute(driverCxt);
    if (subtask.getException() != null) {
        setException(subtask.getException());
    }
    return ret;
}