use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.
the class SparkProcessAnalyzeTable method handlePartialScanCommand.
/**
* handle partial scan command.
*
* It is composed of PartialScanTask followed by StatsTask.
*/
private void handlePartialScanCommand(TableScanOperator tableScan, ParseContext parseContext, StatsWork statsWork, GenSparkProcContext context, Task<StatsWork> statsTask) throws SemanticException {
String aggregationKey = tableScan.getConf().getStatsAggPrefix();
StringBuilder aggregationKeyBuffer = new StringBuilder(aggregationKey);
List<Path> inputPaths = GenMapRedUtils.getInputPathsForPartialScan(tableScan, aggregationKeyBuffer);
aggregationKey = aggregationKeyBuffer.toString();
// scan work
PartialScanWork scanWork = new PartialScanWork(inputPaths);
scanWork.setMapperCannotSpanPartns(true);
scanWork.setAggKey(aggregationKey);
scanWork.setStatsTmpDir(tableScan.getConf().getTmpStatsDir(), parseContext.getConf());
// stats work
statsWork.setPartialScanAnalyzeCommand(true);
// partial scan task
DriverContext driverCxt = new DriverContext();
@SuppressWarnings("unchecked") Task<PartialScanWork> partialScanTask = TaskFactory.get(scanWork, parseContext.getConf());
partialScanTask.initialize(parseContext.getQueryState(), null, driverCxt, tableScan.getCompilationOpContext());
partialScanTask.setWork(scanWork);
statsWork.setSourceTask(partialScanTask);
// task dependency
context.rootTasks.remove(context.currentTask);
context.rootTasks.add(partialScanTask);
partialScanTask.addDependentTask(statsTask);
}
use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.
the class ColumnTruncateTask method execute.
@Override
public /**
* start a new map-reduce job to do the truncation, almost the same as ExecDriver.
*/
int execute(DriverContext driverContext) {
HiveConf.setVar(job, HiveConf.ConfVars.HIVEINPUTFORMAT, BucketizedHiveInputFormat.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);
}
Path outputPath = this.work.getOutputDir();
Path tempOutPath = Utilities.toTempPath(outputPath);
try {
FileSystem fs = tempOutPath.getFileSystem(job);
if (!fs.exists(tempOutPath)) {
fs.mkdirs(tempOutPath);
}
} catch (IOException e) {
console.printError("Can't make path " + outputPath + " : " + e.getMessage());
return 6;
}
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());
}
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);
// 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();
}
}
ColumnTruncateMapper.jobClose(outputPath, success, job, console, work.getDynPartCtx(), null);
} catch (Exception e) {
LOG.warn("Failed while cleaning up ", e);
} finally {
HadoopJobExecHelper.runningJobs.remove(rj);
}
}
return (returnVal);
}
use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.
the class MergeFileTask method execute.
/**
* start a new map-reduce job to do the merge, almost the same as ExecDriver.
*/
@Override
public int execute(DriverContext driverContext) {
Context ctx = driverContext.getCtx();
boolean ctxCreated = false;
RunningJob rj = null;
int returnVal = 0;
try {
if (ctx == null) {
ctx = new Context(job);
ctxCreated = true;
}
HiveFileFormatUtils.prepareJobOutput(job);
job.setInputFormat(work.getInputformatClass());
job.setOutputFormat(HiveOutputFormatImpl.class);
job.setMapperClass(MergeFileMapper.class);
job.setMapOutputKeyClass(NullWritable.class);
job.setMapOutputValueClass(NullWritable.class);
job.setOutputKeyClass(NullWritable.class);
job.setOutputValueClass(NullWritable.class);
job.setNumReduceTasks(0);
// create the temp directories
Path outputPath = work.getOutputDir();
Path tempOutPath = Utilities.toTempPath(outputPath);
FileSystem fs = tempOutPath.getFileSystem(job);
if (!fs.exists(tempOutPath)) {
fs.mkdirs(tempOutPath);
}
ExecDriver.propagateSplitSettings(job, work);
// set job name
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());
}
// add input path
addInputPaths(job, work);
// serialize work
Utilities.setMapWork(job, work, ctx.getMRTmpPath(), true);
// remove 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");
}
// submit the job
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);
// Finally SUBMIT the JOB!
rj = jc.submitJob(job);
this.jobID = rj.getJobID();
returnVal = jobExecHelper.progress(rj, jc, ctx);
success = (returnVal == 0);
} catch (Exception e) {
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();
}
}
// get the list of Dynamic partition paths
if (rj != null) {
if (work.getAliasToWork() != null) {
for (Operator<? extends OperatorDesc> op : work.getAliasToWork().values()) {
op.jobClose(job, success);
}
}
}
} catch (Exception e) {
// jobClose needs to execute successfully otherwise fail task
LOG.warn("Job close failed ", e);
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));
}
} finally {
HadoopJobExecHelper.runningJobs.remove(rj);
}
}
return returnVal;
}
use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.
the class TezTask method execute.
@Override
public int execute(DriverContext driverContext) {
int rc = 1;
boolean cleanContext = false;
Context ctx = null;
Ref<TezSessionState> sessionRef = Ref.from(null);
try {
// Get or create Context object. If we create it we have to clean it later as well.
ctx = driverContext.getCtx();
if (ctx == null) {
ctx = new Context(conf);
cleanContext = true;
// some DDL task that directly executes a TezTask does not setup Context and hence TriggerContext.
// Setting queryId is messed up. Some DDL tasks have executionId instead of proper queryId.
String queryId = HiveConf.getVar(conf, HiveConf.ConfVars.HIVEQUERYID);
WmContext wmContext = new WmContext(System.currentTimeMillis(), queryId);
ctx.setWmContext(wmContext);
}
// Need to remove this static hack. But this is the way currently to get a session.
SessionState ss = SessionState.get();
// Note: given that we return pool sessions to the pool in the finally block below, and that
// we need to set the global to null to do that, this "reuse" may be pointless.
TezSessionState session = sessionRef.value = ss.getTezSession();
if (session != null && !session.isOpen()) {
LOG.warn("The session: " + session + " has not been opened");
}
// We only need a username for UGI to use for groups; getGroups will fetch the groups
// based on Hadoop configuration, as documented at
// https://hadoop.apache.org/docs/r2.8.0/hadoop-project-dist/hadoop-common/GroupsMapping.html
String userName = ss.getUserName();
List<String> groups = null;
if (userName == null) {
userName = "anonymous";
} else {
groups = UserGroupInformation.createRemoteUser(ss.getUserName()).getGroups();
}
MappingInput mi = new MappingInput(userName, groups, ss.getHiveVariables().get("wmpool"), ss.getHiveVariables().get("wmapp"));
WmContext wmContext = ctx.getWmContext();
// jobConf will hold all the configuration for hadoop, tez, and hive
JobConf jobConf = utils.createConfiguration(conf);
// Get all user jars from work (e.g. input format stuff).
String[] allNonConfFiles = work.configureJobConfAndExtractJars(jobConf);
// DAG scratch dir. We get a session from the pool so it may be different from Tez one.
// TODO: we could perhaps reuse the same directory for HiveResources?
Path scratchDir = utils.createTezDir(ctx.getMRScratchDir(), conf);
CallerContext callerContext = CallerContext.create("HIVE", queryPlan.getQueryId(), "HIVE_QUERY_ID", queryPlan.getQueryStr());
perfLogger.PerfLogBegin(CLASS_NAME, PerfLogger.TEZ_GET_SESSION);
session = sessionRef.value = WorkloadManagerFederation.getSession(sessionRef.value, conf, mi, getWork().getLlapMode(), wmContext);
perfLogger.PerfLogEnd(CLASS_NAME, PerfLogger.TEZ_GET_SESSION);
try {
ss.setTezSession(session);
LOG.info("Subscribed to counters: {} for queryId: {}", wmContext.getSubscribedCounters(), wmContext.getQueryId());
// Ensure the session is open and has the necessary local resources.
// This would refresh any conf resources and also local resources.
ensureSessionHasResources(session, allNonConfFiles);
// This is a combination of the jar stuff from conf, and not from conf.
List<LocalResource> allNonAppResources = session.getLocalizedResources();
logResources(allNonAppResources);
Map<String, LocalResource> allResources = DagUtils.createTezLrMap(session.getAppJarLr(), allNonAppResources);
// next we translate the TezWork to a Tez DAG
DAG dag = build(jobConf, work, scratchDir, ctx, allResources);
dag.setCallerContext(callerContext);
// Check isShutdown opportunistically; it's never unset.
if (this.isShutdown) {
throw new HiveException("Operation cancelled");
}
DAGClient dagClient = submit(jobConf, dag, sessionRef);
session = sessionRef.value;
boolean wasShutdown = false;
synchronized (dagClientLock) {
assert this.dagClient == null;
wasShutdown = this.isShutdown;
if (!wasShutdown) {
this.dagClient = dagClient;
}
}
if (wasShutdown) {
closeDagClientOnCancellation(dagClient);
throw new HiveException("Operation cancelled");
}
// finally monitor will print progress until the job is done
TezJobMonitor monitor = new TezJobMonitor(work.getAllWork(), dagClient, conf, dag, ctx);
rc = monitor.monitorExecution();
if (rc != 0) {
this.setException(new HiveException(monitor.getDiagnostics()));
}
// fetch the counters
try {
Set<StatusGetOpts> statusGetOpts = EnumSet.of(StatusGetOpts.GET_COUNTERS);
counters = dagClient.getDAGStatus(statusGetOpts).getDAGCounters();
} catch (Exception err) {
// Don't fail execution due to counters - just don't print summary info
LOG.warn("Failed to get counters. Ignoring, summary info will be incomplete. " + err, err);
counters = null;
}
} finally {
// Note: due to TEZ-3846, the session may actually be invalid in case of some errors.
// Currently, reopen on an attempted reuse will take care of that; we cannot tell
// if the session is usable until we try.
// We return this to the pool even if it's unusable; reopen is supposed to handle this.
wmContext = ctx.getWmContext();
try {
if (sessionRef.value != null) {
sessionRef.value.returnToSessionManager();
}
} catch (Exception e) {
LOG.error("Failed to return session: {} to pool", session, e);
throw e;
}
if (!conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("none") && wmContext != null) {
if (conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("json")) {
wmContext.printJson(console);
} else if (conf.getVar(HiveConf.ConfVars.TEZ_SESSION_EVENTS_SUMMARY).equalsIgnoreCase("text")) {
wmContext.print(console);
}
}
}
if (LOG.isInfoEnabled() && counters != null && (HiveConf.getBoolVar(conf, HiveConf.ConfVars.TEZ_EXEC_SUMMARY) || Utilities.isPerfOrAboveLogging(conf))) {
for (CounterGroup group : counters) {
LOG.info(group.getDisplayName() + ":");
for (TezCounter counter : group) {
LOG.info(" " + counter.getDisplayName() + ": " + counter.getValue());
}
}
}
} catch (Exception e) {
LOG.error("Failed to execute tez graph.", e);
// rc will be 1 at this point indicating failure.
} finally {
Utilities.clearWork(conf);
// Clear gWorkMap
for (BaseWork w : work.getAllWork()) {
JobConf workCfg = workToConf.get(w);
if (workCfg != null) {
Utilities.clearWorkMapForConf(workCfg);
}
}
if (cleanContext) {
try {
ctx.clear();
} catch (Exception e) {
/*best effort*/
LOG.warn("Failed to clean up after tez job", e);
}
}
// need to either move tmp files or remove them
DAGClient dagClient = null;
synchronized (dagClientLock) {
dagClient = this.dagClient;
this.dagClient = null;
}
// DagClient as such should have no bearing on jobClose.
if (dagClient != null) {
// rc will only be overwritten if close errors out
rc = close(work, rc, dagClient);
}
}
return rc;
}
use of org.apache.hadoop.hive.ql.DriverContext in project hive by apache.
the class ExecDriver method main.
@SuppressWarnings("unchecked")
public static void main(String[] args) throws IOException, HiveException {
String planFileName = null;
String jobConfFileName = null;
boolean noLog = false;
String files = null;
String libjars = null;
boolean localtask = false;
try {
for (int i = 0; i < args.length; i++) {
if (args[i].equals("-plan")) {
planFileName = args[++i];
} else if (args[i].equals("-jobconffile")) {
jobConfFileName = args[++i];
} else if (args[i].equals("-nolog")) {
noLog = true;
} else if (args[i].equals("-files")) {
files = args[++i];
} else if (args[i].equals("-libjars")) {
libjars = args[++i];
} else if (args[i].equals("-localtask")) {
localtask = true;
}
}
} catch (IndexOutOfBoundsException e) {
System.err.println("Missing argument to option");
printUsage();
}
JobConf conf;
if (localtask) {
conf = new JobConf(MapredLocalTask.class);
} else {
conf = new JobConf(ExecDriver.class);
}
if (jobConfFileName != null) {
conf.addResource(new Path(jobConfFileName));
}
// Initialize the resources from command line
if (files != null) {
conf.set("tmpfiles", files);
}
if (libjars != null) {
conf.set("tmpjars", libjars);
}
if (UserGroupInformation.isSecurityEnabled()) {
String hadoopAuthToken = System.getenv(UserGroupInformation.HADOOP_TOKEN_FILE_LOCATION);
if (hadoopAuthToken != null) {
conf.set("mapreduce.job.credentials.binary", hadoopAuthToken);
}
}
boolean isSilent = HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVESESSIONSILENT);
String queryId = HiveConf.getVar(conf, HiveConf.ConfVars.HIVEQUERYID, "").trim();
if (queryId.isEmpty()) {
queryId = "unknown-" + System.currentTimeMillis();
HiveConf.setVar(conf, HiveConf.ConfVars.HIVEQUERYID, queryId);
}
System.setProperty(HiveConf.ConfVars.HIVEQUERYID.toString(), queryId);
LogUtils.registerLoggingContext(conf);
if (noLog) {
// If started from main(), and noLog is on, we should not output
// any logs. To turn the log on, please set -Dtest.silent=false
org.apache.logging.log4j.Logger logger = org.apache.logging.log4j.LogManager.getRootLogger();
NullAppender appender = NullAppender.createNullAppender();
appender.addToLogger(logger.getName(), Level.ERROR);
appender.start();
} else {
setupChildLog4j(conf);
}
Logger LOG = LoggerFactory.getLogger(ExecDriver.class.getName());
LogHelper console = new LogHelper(LOG, isSilent);
if (planFileName == null) {
console.printError("Must specify Plan File Name");
printUsage();
}
// that it's easy to find reason for local mode execution failures
for (Appender appender : ((org.apache.logging.log4j.core.Logger) LogManager.getRootLogger()).getAppenders().values()) {
if (appender instanceof FileAppender) {
console.printInfo("Execution log at: " + ((FileAppender) appender).getFileName());
} else if (appender instanceof RollingFileAppender) {
console.printInfo("Execution log at: " + ((RollingFileAppender) appender).getFileName());
}
}
// the plan file should always be in local directory
Path p = new Path(planFileName);
FileSystem fs = FileSystem.getLocal(conf);
InputStream pathData = fs.open(p);
// child process. so we add it here explicitly
try {
// see also - code in CliDriver.java
ClassLoader loader = conf.getClassLoader();
if (StringUtils.isNotBlank(libjars)) {
loader = Utilities.addToClassPath(loader, StringUtils.split(libjars, ","));
}
conf.setClassLoader(loader);
// Also set this to the Thread ContextClassLoader, so new threads will
// inherit
// this class loader, and propagate into newly created Configurations by
// those
// new threads.
Thread.currentThread().setContextClassLoader(loader);
} catch (Exception e) {
throw new HiveException(e.getMessage(), e);
}
int ret;
if (localtask) {
memoryMXBean = ManagementFactory.getMemoryMXBean();
MapredLocalWork plan = SerializationUtilities.deserializePlan(pathData, MapredLocalWork.class);
MapredLocalTask ed = new MapredLocalTask(plan, conf, isSilent);
ret = ed.executeInProcess(new DriverContext());
} else {
MapredWork plan = SerializationUtilities.deserializePlan(pathData, MapredWork.class);
ExecDriver ed = new ExecDriver(plan, conf, isSilent);
ret = ed.execute(new DriverContext());
}
if (ret != 0) {
System.exit(ret);
}
}
Aggregations