}
@Override
public int execute(DriverContext driverContext) {
Context ctx = driverContext.getCtx();
boolean ctxCreated = false;
try {
if (ctx == null) {
ctx = new Context(conf);
ctxCreated = true;
}
// estimate number of reducers
setNumberOfReducers();
// auto-determine local mode if allowed
if (!ctx.isLocalOnlyExecutionMode() &&
conf.getBoolVar(HiveConf.ConfVars.LOCALMODEAUTO)) {
if (inputSummary == null) {
inputSummary = Utilities.getInputSummary(driverContext.getCtx(), work.getMapWork(), null);
}
// set the values of totalInputFileSize and totalInputNumFiles, estimating them
// if percentage block sampling is being used
double samplePercentage = Utilities.getHighestSamplePercentage(work.getMapWork());
totalInputFileSize = Utilities.getTotalInputFileSize(inputSummary, work.getMapWork(), samplePercentage);
totalInputNumFiles = Utilities.getTotalInputNumFiles(inputSummary, work.getMapWork(), samplePercentage);
// at this point the number of reducers is precisely defined in the plan
int numReducers = work.getReduceWork() == null ? 0 : work.getReduceWork().getNumReduceTasks();
if (LOG.isDebugEnabled()) {
LOG.debug("Task: " + getId() + ", Summary: " +
totalInputFileSize + "," + totalInputNumFiles + ","
+ numReducers);
}
String reason = MapRedTask.isEligibleForLocalMode(conf, numReducers,
totalInputFileSize, totalInputNumFiles);
if (reason == null) {
// clone configuration before modifying it on per-task basis
cloneConf();
ShimLoader.getHadoopShims().setJobLauncherRpcAddress(conf, "local");
console.printInfo("Selecting local mode for task: " + getId());
this.setLocalMode(true);
} else {
console.printInfo("Cannot run job locally: " + reason);
this.setLocalMode(false);
}
}
runningViaChild = ShimLoader.getHadoopShims().isLocalMode(conf) ||
conf.getBoolVar(HiveConf.ConfVars.SUBMITVIACHILD);
if(!runningViaChild) {
// we are not running this mapred task via child jvm
// so directly invoke ExecDriver
return super.execute(driverContext);
}
// we need to edit the configuration to setup cmdline. clone it first
cloneConf();
// propagate input format if necessary
super.setInputAttributes(conf);
// enable assertion
String hadoopExec = conf.getVar(HiveConf.ConfVars.HADOOPBIN);
String hiveJar = conf.getJar();
String libJarsOption;
String addedJars = Utilities.getResourceFiles(conf, SessionState.ResourceType.JAR);
conf.setVar(ConfVars.HIVEADDEDJARS, addedJars);
String auxJars = conf.getAuxJars();
// Put auxjars and addedjars together into libjars
if (StringUtils.isEmpty(addedJars)) {
if (StringUtils.isEmpty(auxJars)) {
libJarsOption = " ";
} else {
libJarsOption = " -libjars " + auxJars + " ";
}
} else {
if (StringUtils.isEmpty(auxJars)) {
libJarsOption = " -libjars " + addedJars + " ";
} else {
libJarsOption = " -libjars " + addedJars + "," + auxJars + " ";
}
}
// Generate the hiveConfArgs after potentially adding the jars
String hiveConfArgs = generateCmdLine(conf, ctx);
// write out the plan to a local file
Path planPath = new Path(ctx.getLocalTmpFileURI(), "plan.xml");
OutputStream out = FileSystem.getLocal(conf).create(planPath);
MapredWork plan = getWork();
LOG.info("Generating plan file " + planPath.toString());
Utilities.serializePlan(plan, out);
String isSilent = "true".equalsIgnoreCase(System
.getProperty("test.silent")) ? "-nolog" : "";
String jarCmd;
if (ShimLoader.getHadoopShims().usesJobShell()) {
jarCmd = libJarsOption + hiveJar + " " + ExecDriver.class.getName();
} else {
jarCmd = hiveJar + " " + ExecDriver.class.getName() + libJarsOption;
}
String cmdLine = hadoopExec + " jar " + jarCmd + " -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 = (new Path(ctx.getLocalTmpFileURI())).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);
}
}
}
LOG.info("Executing: " + cmdLine);
// 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(System.getenv());
// The user can specify the hadoop memory
if (ShimLoader.getHadoopShims().isLocalMode(conf)) {
// 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);
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
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
}
if (variables.containsKey(HADOOP_OPTS_KEY)) {
variables.put(HADOOP_OPTS_KEY, variables.get(HADOOP_OPTS_KEY)
+ hadoopOpts);
} else {
variables.put(HADOOP_OPTS_KEY, hadoopOpts);
}
if(variables.containsKey(HIVE_DEBUG_RECURSIVE)) {
configureDebugVariablesForChildJVM(variables);
}
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;
}
// Run ExecDriver in another JVM
executor = Runtime.getRuntime().exec(cmdLine, env, new File(workDir));
CachingPrintStream errPrintStream =
new CachingPrintStream(SessionState.getConsole().getChildErrStream());
StreamPrinter outPrinter = new StreamPrinter(
executor.getInputStream(), null,
SessionState.getConsole().getChildOutStream());
StreamPrinter errPrinter = new StreamPrinter(
executor.getErrorStream(), null,
errPrintStream);
outPrinter.start();
errPrinter.start();
int exitVal = jobExecHelper.progressLocal(executor, getId());
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) {
e.printStackTrace();
LOG.error("Exception: " + e.getMessage());
return (1);
} finally {
try {
// creating the context can create a bunch of files. So make
// sure to clear it out
if(ctxCreated) {
ctx.clear();
}
} catch (Exception e) {
LOG.error("Exception: " + e.getMessage());
}