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

Example 6 with StatsPublisher

use of org.apache.hadoop.hive.ql.stats.StatsPublisher in project hive by apache.

the class PartialScanMapper method publishStats.

/**
   * Publish statistics.
   * similar to FileSinkOperator.java publishStats()
   *
   * @throws HiveException
   */
private void publishStats() throws HiveException {
    // Initializing a stats publisher
    StatsPublisher statsPublisher = Utilities.getStatsPublisher(jc);
    if (statsPublisher == null) {
        // just return, stats gathering should not block the main query
        LOG.error("StatsPublishing error: StatsPublisher is not initialized.");
        throw new HiveException(ErrorMsg.STATSPUBLISHER_NOT_OBTAINED.getErrorCodedMsg());
    }
    StatsCollectionContext sc = new StatsCollectionContext(jc);
    sc.setStatsTmpDir(jc.get(StatsSetupConst.STATS_TMP_LOC, ""));
    if (!statsPublisher.connect(sc)) {
        // should fail since stats gathering is main purpose of the job
        LOG.error("StatsPublishing error: cannot connect to database");
        throw new HiveException(ErrorMsg.STATSPUBLISHER_CONNECTION_ERROR.getErrorCodedMsg());
    }
    // construct key used to store stats in intermediate db
    String key = statsAggKeyPrefix.endsWith(Path.SEPARATOR) ? statsAggKeyPrefix : statsAggKeyPrefix + Path.SEPARATOR;
    // construct statistics to be stored
    Map<String, String> statsToPublish = new HashMap<String, String>();
    statsToPublish.put(StatsSetupConst.RAW_DATA_SIZE, Long.toString(uncompressedFileSize));
    statsToPublish.put(StatsSetupConst.ROW_COUNT, Long.toString(rowNo));
    if (!statsPublisher.publishStat(key, statsToPublish)) {
        // Not changing the interface to maintain backward compatibility
        throw new HiveException(ErrorMsg.STATSPUBLISHER_PUBLISHING_ERROR.getErrorCodedMsg());
    }
    if (!statsPublisher.closeConnection(sc)) {
        // Not changing the interface to maintain backward compatibility
        throw new HiveException(ErrorMsg.STATSPUBLISHER_CLOSING_ERROR.getErrorCodedMsg());
    }
}

Example 7 with StatsPublisher

use of org.apache.hadoop.hive.ql.stats.StatsPublisher in project hive by apache.

the class PartialScanTask method execute.

@Override
public /**
   * start a new map-reduce job to do partial scan to calculate Stats,
   * almost the same as BlockMergeTask or ExecDriver.
   */
int execute(DriverContext driverContext) {
    HiveConf.setVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT, CombineHiveInputFormat.class.getName());
    success = true;
    HiveFileFormatUtils.prepareJobOutput(job);
    job.setOutputFormat(HiveOutputFormatImpl.class);
    job.setMapperClass(work.getMapperClass());
    Context ctx = driverContext.getCtx();
    boolean ctxCreated = false;
    try {
        if (ctx == null) {
            ctx = new Context(job);
            ctxCreated = true;
        }
    } catch (IOException e) {
        e.printStackTrace();
        console.printError("Error launching map-reduce job", "\n" + org.apache.hadoop.util.StringUtils.stringifyException(e));
        return 5;
    }
    job.setMapOutputKeyClass(NullWritable.class);
    job.setMapOutputValueClass(NullWritable.class);
    if (work.getNumMapTasks() != null) {
        job.setNumMapTasks(work.getNumMapTasks());
    }
    // zero reducers
    job.setNumReduceTasks(0);
    if (work.getMinSplitSize() != null) {
        HiveConf.setLongVar(job, HiveConf.ConfVars.MAPREDMINSPLITSIZE, work.getMinSplitSize().longValue());
    }
    if (work.getInputformat() != null) {
        HiveConf.setVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT, work.getInputformat());
    }
    String inpFormat = HiveConf.getVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT);
    LOG.info("Using " + inpFormat);
    try {
        job.setInputFormat(JavaUtils.loadClass(inpFormat));
    } catch (ClassNotFoundException e) {
        throw new RuntimeException(e.getMessage(), e);
    }
    job.setOutputKeyClass(NullWritable.class);
    job.setOutputValueClass(NullWritable.class);
    int returnVal = 0;
    RunningJob rj = null;
    boolean noName = StringUtils.isEmpty(job.get(MRJobConfig.JOB_NAME));
    String jobName = null;
    if (noName && this.getQueryPlan() != null) {
        int maxlen = conf.getIntVar(HiveConf.ConfVars.HIVEJOBNAMELENGTH);
        jobName = Utilities.abbreviate(this.getQueryPlan().getQueryStr(), maxlen - 6);
    }
    if (noName) {
        // This is for a special case to ensure unit tests pass
        job.set(MRJobConfig.JOB_NAME, jobName != null ? jobName : "JOB" + Utilities.randGen.nextInt());
    }
    // pass aggregation key to mapper
    HiveConf.setVar(job, HiveConf.ConfVars.HIVE_STATS_KEY_PREFIX, work.getAggKey());
    job.set(StatsSetupConst.STATS_TMP_LOC, work.getStatsTmpDir());
    try {
        addInputPaths(job, work);
        MapredWork mrWork = new MapredWork();
        mrWork.setMapWork(work);
        Utilities.setMapRedWork(job, mrWork, ctx.getMRTmpPath());
        // remove the pwd from conf file so that job tracker doesn't show this
        // logs
        String pwd = HiveConf.getVar(job, HiveConf.ConfVars.METASTOREPWD);
        if (pwd != null) {
            HiveConf.setVar(job, HiveConf.ConfVars.METASTOREPWD, "HIVE");
        }
        JobClient jc = new JobClient(job);
        String addedJars = Utilities.getResourceFiles(job, SessionState.ResourceType.JAR);
        if (!addedJars.isEmpty()) {
            job.set("tmpjars", addedJars);
        }
        // make this client wait if job trcker is not behaving well.
        Throttle.checkJobTracker(job, LOG);
        if (work.isGatheringStats()) {
            // initialize stats publishing table
            StatsPublisher statsPublisher;
            StatsFactory factory = StatsFactory.newFactory(job);
            if (factory != null) {
                statsPublisher = factory.getStatsPublisher();
                StatsCollectionContext sc = new StatsCollectionContext(job);
                sc.setStatsTmpDir(work.getStatsTmpDir());
                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());
                    }
                }
            }
        }
        // Finally SUBMIT the JOB!
        rj = jc.submitJob(job);
        this.jobID = rj.getJobID();
        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 {
        try {
            if (ctxCreated) {
                ctx.clear();
            }
            if (rj != null) {
                if (returnVal != 0) {
                    rj.killJob();
                }
            }
        } catch (Exception e) {
            LOG.warn("Failed in cleaning up ", e);
        } finally {
            HadoopJobExecHelper.runningJobs.remove(rj);
        }
    }
    return (returnVal);
}

Example 8 with StatsPublisher

use of org.apache.hadoop.hive.ql.stats.StatsPublisher 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.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();
        if (HiveConf.getVar(job, HiveConf.ConfVars.HIVE_EXECUTION_ENGINE).equals("tez") && ss != null) {
            TezSessionState session = ss.getTezSession();
            TezSessionPoolManager.getInstance().closeIfNotDefault(session, 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");
            if (rj != null) {
                rj.killJob();
                rj = null;
            }
            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) {
                    rj.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 9 with StatsPublisher

use of org.apache.hadoop.hive.ql.stats.StatsPublisher in project hive by apache.

the class DagUtils method createVertex.

/**
   * Create a vertex from a given work object.
   *
   * @param conf JobConf to be used to this execution unit
   * @param work The instance of BaseWork representing the actual work to be performed
   * by this vertex.
   * @param scratchDir HDFS scratch dir for this execution unit.
   * @param appJarLr Local resource for hive-exec.
   * @param additionalLr
   * @param fileSystem FS corresponding to scratchDir and LocalResources
   * @param ctx This query's context
   * @return Vertex
   */
@SuppressWarnings("deprecation")
public Vertex createVertex(JobConf conf, BaseWork work, Path scratchDir, LocalResource appJarLr, List<LocalResource> additionalLr, FileSystem fileSystem, Context ctx, boolean hasChildren, TezWork tezWork, VertexType vertexType) throws Exception {
    Vertex v = null;
    // BaseWork.
    if (work instanceof MapWork) {
        v = createVertex(conf, (MapWork) work, appJarLr, additionalLr, fileSystem, scratchDir, ctx, vertexType);
    } else if (work instanceof ReduceWork) {
        v = createVertex(conf, (ReduceWork) work, appJarLr, additionalLr, fileSystem, scratchDir, ctx);
    } else if (work instanceof MergeJoinWork) {
        v = createVertex(conf, (MergeJoinWork) work, appJarLr, additionalLr, fileSystem, scratchDir, ctx, vertexType);
    } else {
        // something is seriously wrong if this is happening
        throw new HiveException(ErrorMsg.GENERIC_ERROR.getErrorCodedMsg());
    }
    // initialize stats publisher if necessary
    if (work.isGatheringStats()) {
        StatsPublisher statsPublisher;
        StatsFactory factory = StatsFactory.newFactory(conf);
        if (factory != null) {
            StatsCollectionContext sCntxt = new StatsCollectionContext(conf);
            sCntxt.setStatsTmpDirs(Utilities.getStatsTmpDirs(work, conf));
            statsPublisher = factory.getStatsPublisher();
            if (!statsPublisher.init(sCntxt)) {
                // creating stats table if not exists
                if (HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_STATS_RELIABLE)) {
                    throw new HiveException(ErrorMsg.STATSPUBLISHER_INITIALIZATION_ERROR.getErrorCodedMsg());
                }
            }
        }
    }
    // final vertices need to have at least one output
    if (!hasChildren) {
        v.addDataSink("out_" + work.getName(), new DataSinkDescriptor(OutputDescriptor.create(MROutput.class.getName()).setUserPayload(TezUtils.createUserPayloadFromConf(conf)), null, null));
    }
    return v;
}

Example 10 with StatsPublisher

use of org.apache.hadoop.hive.ql.stats.StatsPublisher in project hive by apache.

the class SparkPlanGenerator method initStatsPublisher.

private void initStatsPublisher(BaseWork work) throws HiveException {
    // initialize stats publisher if necessary
    if (work.isGatheringStats()) {
        StatsPublisher statsPublisher;
        StatsFactory factory = StatsFactory.newFactory(jobConf);
        if (factory != null) {
            statsPublisher = factory.getStatsPublisher();
            StatsCollectionContext sc = new StatsCollectionContext(jobConf);
            sc.setStatsTmpDirs(Utilities.getStatsTmpDirs(work, jobConf));
            if (!statsPublisher.init(sc)) {
                // creating stats table if not exists
                if (HiveConf.getBoolVar(jobConf, HiveConf.ConfVars.HIVE_STATS_RELIABLE)) {
                    throw new HiveException(ErrorMsg.STATSPUBLISHER_INITIALIZATION_ERROR.getErrorCodedMsg());
                }
            }
        }
    }
}