/** Configures the Reduce plan, the POPackage operator and the reporter thread */
@SuppressWarnings("unchecked")
@Override
protected void setup(Context context) throws IOException, InterruptedException {
super.setup(context);
inIllustrator = inIllustrator(context);
if (inIllustrator) pack = getPack(context);
Configuration jConf = context.getConfiguration();
SpillableMemoryManager.configure(ConfigurationUtil.toProperties(jConf));
context
.getConfiguration()
.set(
PigConstants.TASK_INDEX,
Integer.toString(context.getTaskAttemptID().getTaskID().getId()));
sJobContext = context;
sJobConfInternal.set(context.getConfiguration());
sJobConf = context.getConfiguration();
try {
PigContext.setPackageImportList(
(ArrayList<String>) ObjectSerializer.deserialize(jConf.get("udf.import.list")));
pigContext = (PigContext) ObjectSerializer.deserialize(jConf.get("pig.pigContext"));
// This attempts to fetch all of the generated code from the distributed cache, and resolve
// it
SchemaTupleBackend.initialize(jConf, pigContext);
if (rp == null)
rp = (PhysicalPlan) ObjectSerializer.deserialize(jConf.get("pig.reducePlan"));
stores = PlanHelper.getPhysicalOperators(rp, POStore.class);
if (!inIllustrator)
pack = (POPackage) ObjectSerializer.deserialize(jConf.get("pig.reduce.package"));
// To be removed
if (rp.isEmpty()) log.debug("Reduce Plan empty!");
else {
ByteArrayOutputStream baos = new ByteArrayOutputStream();
rp.explain(baos);
log.debug(baos.toString());
}
pigReporter = new ProgressableReporter();
if (!(rp.isEmpty())) {
roots = rp.getRoots().toArray(new PhysicalOperator[1]);
leaf = rp.getLeaves().get(0);
}
// Get the UDF specific context
MapRedUtil.setupUDFContext(jConf);
} catch (IOException ioe) {
String msg = "Problem while configuring reduce plan.";
throw new RuntimeException(msg, ioe);
}
log.info(
"Aliases being processed per job phase (AliasName[line,offset]): "
+ jConf.get("pig.alias.location"));
Utils.setDefaultTimeZone(PigMapReduce.sJobConfInternal.get());
}
public RequiredFieldResponse pushProjection(RequiredFieldList requiredFieldList)
throws FrontendException {
if (requiredFieldList == null) {
return null;
}
if (requiredFieldList.getFields() != null) {
int lastColumn = -1;
for (RequiredField rf : requiredFieldList.getFields()) {
if (rf.getIndex() > lastColumn) {
lastColumn = rf.getIndex();
}
}
requiredFields = new boolean[lastColumn + 1];
for (RequiredField rf : requiredFieldList.getFields()) {
if (rf.getIndex() != -1) requiredFields[rf.getIndex()] = true;
}
Properties p = UDFContext.getUDFContext().getUDFProperties(this.getClass());
try {
p.setProperty(REQUIRED_FIELDS_SIGNATURE, ObjectSerializer.serialize(requiredFields));
} catch (Exception e) {
throw new RuntimeException("Cannot serialize mRequiredColumns");
}
}
return new RequiredFieldResponse(true);
}
@SuppressWarnings("unchecked")
@Override
public void prepareToRead(RecordReader reader, PigSplit split) throws IOException {
this.reader = reader;
final String resourceSchemaAsStr =
getValueFromUDFContext(this.contextSignature, RESOURCE_SCHEMA_SIGNATURE);
if (resourceSchemaAsStr == null) {
throw new IOException("Could not find schema in UDF context");
}
schema = (ResourceSchema) ObjectSerializer.deserialize(resourceSchemaAsStr);
}
@Override
public void prepareToRead(RecordReader recordReader, PigSplit pigSplit) throws IOException {
reader = recordReader;
if (!requiredFieldsInitialized) {
UDFContext udfc = UDFContext.getUDFContext();
Properties p = udfc.getUDFProperties(this.getClass(), new String[] {udfContextSignature});
requiredFields =
(boolean[]) ObjectSerializer.deserialize(p.getProperty(REQUIRED_FIELDS_SIGNATURE));
requiredFieldsInitialized = true;
}
}
public void setConf(Configuration conf) {
try {
mAsc = (boolean[]) ObjectSerializer.deserialize(conf.get("pig.sortOrder"));
} catch (IOException ioe) {
mLog.error("Unable to deserialize pig.sortOrder " + ioe.getMessage());
throw new RuntimeException(ioe);
}
if (mAsc == null) {
mAsc = new boolean[1];
mAsc[0] = true;
}
}
/**
* stolen from JobControlCompiler TODO: refactor it to share this
*
* @param physicalPlan
* @param poLoad
* @param jobConf
* @return
* @throws java.io.IOException
*/
private static JobConf configureLoader(PhysicalPlan physicalPlan, POLoad poLoad, JobConf jobConf)
throws IOException {
// 这部分似乎没用
Job job = new Job(jobConf);
LoadFunc loadFunc = poLoad.getLoadFunc();
loadFunc.setLocation(poLoad.getLFile().getFileName(), job);
// stolen from JobControlCompiler
ArrayList<FileSpec> pigInputs = new ArrayList<FileSpec>();
// Store the inp filespecs
pigInputs.add(poLoad.getLFile());
ArrayList<List<OperatorKey>> inpTargets = Lists.newArrayList();
ArrayList<String> inpSignatures = Lists.newArrayList();
ArrayList<Long> inpLimits = Lists.newArrayList();
// Store the target operators for tuples read
// from this input
List<PhysicalOperator> loadSuccessors = physicalPlan.getSuccessors(poLoad);
List<OperatorKey> loadSuccessorsKeys = Lists.newArrayList();
if (loadSuccessors != null) {
for (PhysicalOperator loadSuccessor : loadSuccessors) {
loadSuccessorsKeys.add(loadSuccessor.getOperatorKey());
}
}
inpTargets.add(loadSuccessorsKeys);
inpSignatures.add(poLoad.getSignature());
inpLimits.add(poLoad.getLimit());
jobConf.set("pig.inputs", ObjectSerializer.serialize(pigInputs));
jobConf.set("pig.inpTargets", ObjectSerializer.serialize(inpTargets));
jobConf.set("pig.inpSignatures", ObjectSerializer.serialize(inpSignatures));
jobConf.set("pig.inpLimits", ObjectSerializer.serialize(inpLimits));
return jobConf;
}
@Override
public ResourceSchema getSchema(String location, Job job) throws IOException {
if (schema != null) {
return schema;
}
final Configuration configuration = job.getConfiguration();
this.initializePhoenixPigConfiguration(location, configuration);
this.schema = PhoenixPigSchemaUtil.getResourceSchema(this.config);
if (LOG.isDebugEnabled()) {
LOG.debug(
String.format(
"Resource Schema generated for location [%s] is [%s]", location, schema.toString()));
}
this.storeInUDFContext(
this.contextSignature, RESOURCE_SCHEMA_SIGNATURE, ObjectSerializer.serialize(schema));
return schema;
}
public void setConf(Configuration conf) {
if (!(conf instanceof JobConf)) {
mLog.warn("Expected jobconf in setConf, got " + conf.getClass().getName());
return;
}
JobConf jconf = (JobConf) conf;
try {
mAsc = (boolean[]) ObjectSerializer.deserialize(jconf.get("pig.sortOrder"));
} catch (IOException ioe) {
mLog.error("Unable to deserialize pig.sortOrder " + ioe.getMessage());
throw new RuntimeException(ioe);
}
if (mAsc == null) {
mAsc = new boolean[1];
mAsc[0] = true;
}
// If there's only one entry in mAsc, it means it's for the whole
// tuple. So we can't be looking for each column.
mWholeTuple = (mAsc.length == 1);
}
public static Properties getJobConf(Configuration conf) {
if (conf == null) {
return null;
}
Properties jobConfProperties = null;
try {
jobConfProperties = new Properties();
for (Map.Entry<String, String> entry : conf) {
if (entry.getKey().equals("pig.mapPlan") || entry.getKey().equals("pig.reducePlan")) {
jobConfProperties.setProperty(
entry.getKey(), ObjectSerializer.deserialize(entry.getValue()).toString());
} else if (JOB_CONF_KEYS.contains(entry.getKey())) {
jobConfProperties.setProperty(entry.getKey(), entry.getValue());
}
}
} catch (IOException e) {
logger.warn("Error while reading job conf: " + e.getMessage());
}
return jobConfProperties;
}
/* (non-Javadoc)
* @see org.apache.pig.builtin.PigStorage#getNext()
*/
@Override
public Tuple getNext() throws IOException {
// If SKIP_INPUT_HEADER and this is the first input split, skip header record
// We store its value as a string though, so we can compare
// further records to it. If they are the same (this would
// happen if multiple small files each with a header were combined
// into one split), we know to skip the duplicate header record as well.
if (loadingFirstRecord
&& headerTreatment == Headers.SKIP_INPUT_HEADER
&& (splitIndex == 0 || splitIndex == -1)) {
try {
if (!in.nextKeyValue()) return null;
header = ((Text) in.getCurrentValue()).toString();
} catch (InterruptedException e) {
int errCode = 6018;
String errMsg = "Error while reading input";
throw new ExecException(errMsg, errCode, PigException.REMOTE_ENVIRONMENT, e);
}
}
loadingFirstRecord = false;
mProtoTuple = new ArrayList<Object>();
getNextInQuotedField = false;
boolean evenQuotesSeen = true;
boolean sawEmbeddedRecordDelimiter = false;
byte[] buf = null;
if (!mRequiredColumnsInitialized) {
if (udfContextSignature != null) {
Properties p = UDFContext.getUDFContext().getUDFProperties(this.getClass());
mRequiredColumns =
(boolean[]) ObjectSerializer.deserialize(p.getProperty(udfContextSignature));
}
mRequiredColumnsInitialized = true;
}
// Note: we cannot factor out the check for nextKeyValue() being null,
// because that call overwrites buf with the new line, which is
// bad if we have a field with a newline.
try {
int recordLen = 0;
getNextFieldID = 0;
while (sawEmbeddedRecordDelimiter || getNextFieldID == 0) {
Text value = null;
if (sawEmbeddedRecordDelimiter) {
// Deal with pulling more records from the input, because
// a double quoted embedded newline was encountered in a field.
// Save the length of the record so far, plus one byte for the
// record delimiter (usually newline) that's embedded in the field
// we were working on before falling into this branch:
int prevLineLen = recordLen + 1;
// Save previous line (the one with the field that has the newline) in a new array.
// The last byte will be random; we'll fill in the embedded
// record delimiter (usually newline) below:
byte[] prevLineSaved = Arrays.copyOf(buf, prevLineLen);
prevLineSaved[prevLineLen - 1] = RECORD_DEL;
// Read the continuation of the record, unless EOF:
if (!in.nextKeyValue()) {
return null;
}
value = (Text) in.getCurrentValue();
recordLen = value.getLength();
// Grab the continuation's bytes:
buf = value.getBytes();
// Combine the previous line and the continuation into a new array.
// The following copyOf() does half the job: it allocates all the
// space, and also copies the previous line into that space:
byte[] prevLineAndContinuation = Arrays.copyOf(prevLineSaved, prevLineLen + recordLen);
// Now append the continuation. Parms: fromBuf, fromStartPos, toBuf, toStartPos,
// lengthToCopy:
System.arraycopy(buf, 0, prevLineAndContinuation, prevLineLen, recordLen);
// We'll work with the combination now:
buf = prevLineAndContinuation;
// Do the whole record over from the start:
mProtoTuple.clear();
getNextInQuotedField = false;
evenQuotesSeen = true;
getNextFieldID = 0;
recordLen = prevLineAndContinuation.length;
} else {
// Previous record finished cleanly: start with the next record,
// unless EOF:
if (!in.nextKeyValue()) {
return null;
}
value = (Text) in.getCurrentValue();
// if the line is a duplicate header and 'SKIP_INPUT_HEADER' is set, ignore it
// (this might happen if multiple files each with a header are combined into a single
// split)
if (headerTreatment == Headers.SKIP_INPUT_HEADER && value.toString().equals(header)) {
if (!in.nextKeyValue()) return null;
value = (Text) in.getCurrentValue();
}
buf = value.getBytes();
getNextFieldID = 0;
recordLen = value.getLength();
}
nextTupleSkipChar = false;
ByteBuffer fieldBuffer = ByteBuffer.allocate(recordLen);
sawEmbeddedRecordDelimiter =
processOneInRecord(evenQuotesSeen, buf, recordLen, fieldBuffer);
// The last field is never delimited by a FIELD_DEL, but by
// the end of the record. So we need to add that last field.
// The '!sawEmbeddedRecordDelimiter' handles the case of
// embedded newlines; we are amidst a field, not at
// the final record:
if (!sawEmbeddedRecordDelimiter) readField(fieldBuffer, getNextFieldID++);
} // end while
} catch (InterruptedException e) {
int errCode = 6018;
String errMsg = "Error while reading input";
throw new ExecException(errMsg, errCode, PigException.REMOTE_ENVIRONMENT, e);
}
Tuple t = mTupleFactory.newTupleNoCopy(mProtoTuple);
return t;
}
/**
* The Main-Class for the Pig Jar that will provide a shell and setup a classpath appropriate for
* executing Jar files.
*
* @param args -jar can be used to add additional jar files (colon separated). - will start a
* shell. -e will execute the rest of the command line as if it was input to the shell.
* @throws IOException
*/
public static void main(String args[]) {
int rc = 1;
Properties properties = new Properties();
PropertiesUtil.loadPropertiesFromFile(properties);
boolean verbose = false;
boolean gruntCalled = false;
String logFileName = null;
try {
BufferedReader pin = null;
boolean debug = false;
boolean dryrun = false;
ArrayList<String> params = new ArrayList<String>();
ArrayList<String> paramFiles = new ArrayList<String>();
HashSet<String> optimizerRules = new HashSet<String>();
CmdLineParser opts = new CmdLineParser(args);
opts.registerOpt('4', "log4jconf", CmdLineParser.ValueExpected.REQUIRED);
opts.registerOpt('b', "brief", CmdLineParser.ValueExpected.NOT_ACCEPTED);
opts.registerOpt('c', "cluster", CmdLineParser.ValueExpected.REQUIRED);
opts.registerOpt('d', "debug", CmdLineParser.ValueExpected.REQUIRED);
opts.registerOpt('e', "execute", CmdLineParser.ValueExpected.NOT_ACCEPTED);
opts.registerOpt('f', "file", CmdLineParser.ValueExpected.REQUIRED);
opts.registerOpt('h', "help", CmdLineParser.ValueExpected.NOT_ACCEPTED);
opts.registerOpt('i', "version", CmdLineParser.ValueExpected.OPTIONAL);
opts.registerOpt('j', "jar", CmdLineParser.ValueExpected.REQUIRED);
opts.registerOpt('l', "logfile", CmdLineParser.ValueExpected.REQUIRED);
opts.registerOpt('m', "param_file", CmdLineParser.ValueExpected.OPTIONAL);
opts.registerOpt('o', "hod", CmdLineParser.ValueExpected.NOT_ACCEPTED);
opts.registerOpt('p', "param", CmdLineParser.ValueExpected.OPTIONAL);
opts.registerOpt('r', "dryrun", CmdLineParser.ValueExpected.NOT_ACCEPTED);
opts.registerOpt('t', "optimizer_off", CmdLineParser.ValueExpected.REQUIRED);
opts.registerOpt('v', "verbose", CmdLineParser.ValueExpected.NOT_ACCEPTED);
opts.registerOpt('w', "warning", CmdLineParser.ValueExpected.NOT_ACCEPTED);
opts.registerOpt('x', "exectype", CmdLineParser.ValueExpected.REQUIRED);
opts.registerOpt('F', "stop_on_failure", CmdLineParser.ValueExpected.NOT_ACCEPTED);
opts.registerOpt('M', "no_multiquery", CmdLineParser.ValueExpected.NOT_ACCEPTED);
ExecMode mode = ExecMode.UNKNOWN;
String file = null;
ExecType execType = ExecType.MAPREDUCE;
String execTypeString = properties.getProperty("exectype");
if (execTypeString != null && execTypeString.length() > 0) {
execType = PigServer.parseExecType(execTypeString);
}
String cluster = "local";
String clusterConfigured = properties.getProperty("cluster");
if (clusterConfigured != null && clusterConfigured.length() > 0) {
cluster = clusterConfigured;
}
// by default warning aggregation is on
properties.setProperty("aggregate.warning", "" + true);
// by default multiquery optimization is on
properties.setProperty("opt.multiquery", "" + true);
// by default we keep going on error on the backend
properties.setProperty("stop.on.failure", "" + false);
char opt;
while ((opt = opts.getNextOpt()) != CmdLineParser.EndOfOpts) {
switch (opt) {
case '4':
String log4jconf = opts.getValStr();
if (log4jconf != null) {
properties.setProperty(LOG4J_CONF, log4jconf);
}
break;
case 'b':
properties.setProperty(BRIEF, "true");
break;
case 'c':
// Needed away to specify the cluster to run the MR job on
// Bug 831708 - fixed
String clusterParameter = opts.getValStr();
if (clusterParameter != null && clusterParameter.length() > 0) {
cluster = clusterParameter;
}
break;
case 'd':
String logLevel = opts.getValStr();
if (logLevel != null) {
properties.setProperty(DEBUG, logLevel);
}
debug = true;
break;
case 'e':
mode = ExecMode.STRING;
break;
case 'f':
mode = ExecMode.FILE;
file = opts.getValStr();
break;
case 'F':
properties.setProperty("stop.on.failure", "" + true);
break;
case 'h':
usage();
return;
case 'i':
System.out.println(getVersionString());
return;
case 'j':
String jarsString = opts.getValStr();
if (jarsString != null) {
properties.setProperty(JAR, jarsString);
}
break;
case 'l':
// call to method that validates the path to the log file
// and sets up the file to store the client side log file
String logFileParameter = opts.getValStr();
if (logFileParameter != null && logFileParameter.length() > 0) {
logFileName = validateLogFile(logFileParameter, null);
} else {
logFileName = validateLogFile(logFileName, null);
}
properties.setProperty("pig.logfile", logFileName);
break;
case 'm':
paramFiles.add(opts.getValStr());
break;
case 'M':
// turns off multiquery optimization
properties.setProperty("opt.multiquery", "" + false);
break;
case 'o':
// TODO sgroschupf using system properties is always a very bad idea
String gateway = System.getProperty("ssh.gateway");
if (gateway == null || gateway.length() == 0) {
properties.setProperty("hod.server", "local");
} else {
properties.setProperty("hod.server", System.getProperty("ssh.gateway"));
}
break;
case 'p':
String val = opts.getValStr();
params.add(opts.getValStr());
break;
case 'r':
// currently only used for parameter substitution
// will be extended in the future
dryrun = true;
break;
case 't':
optimizerRules.add(opts.getValStr());
break;
case 'v':
properties.setProperty(VERBOSE, "" + true);
verbose = true;
break;
case 'w':
properties.setProperty("aggregate.warning", "" + false);
break;
case 'x':
try {
execType = PigServer.parseExecType(opts.getValStr());
} catch (IOException e) {
throw new RuntimeException("ERROR: Unrecognized exectype.", e);
}
break;
default:
{
Character cc = new Character(opt);
throw new AssertionError("Unhandled option " + cc.toString());
}
}
}
// configure logging
configureLog4J(properties);
// create the context with the parameter
PigContext pigContext = new PigContext(execType, properties);
if (logFileName == null) {
logFileName = validateLogFile(null, null);
}
pigContext.getProperties().setProperty("pig.logfile", logFileName);
if (optimizerRules.size() > 0) {
pigContext
.getProperties()
.setProperty("pig.optimizer.rules", ObjectSerializer.serialize(optimizerRules));
}
LogicalPlanBuilder.classloader = pigContext.createCl(null);
// construct the parameter substitution preprocessor
Grunt grunt = null;
BufferedReader in;
String substFile = null;
switch (mode) {
case FILE:
{
// Run, using the provided file as a pig file
in = new BufferedReader(new FileReader(file));
// run parameter substitution preprocessor first
substFile = file + ".substituted";
pin = runParamPreprocessor(in, params, paramFiles, substFile, debug || dryrun);
if (dryrun) {
log.info("Dry run completed. Substituted pig script is at " + substFile);
return;
}
logFileName = validateLogFile(logFileName, file);
pigContext.getProperties().setProperty("pig.logfile", logFileName);
// Set job name based on name of the script
pigContext
.getProperties()
.setProperty(PigContext.JOB_NAME, "PigLatin:" + new File(file).getName());
if (!debug) {
new File(substFile).deleteOnExit();
}
grunt = new Grunt(pin, pigContext);
gruntCalled = true;
int results[] = grunt.exec();
rc = getReturnCodeForStats(results);
return;
}
case STRING:
{
// Gather up all the remaining arguments into a string and pass them into
// grunt.
StringBuffer sb = new StringBuffer();
String remainders[] = opts.getRemainingArgs();
for (int i = 0; i < remainders.length; i++) {
if (i != 0) sb.append(' ');
sb.append(remainders[i]);
}
in = new BufferedReader(new StringReader(sb.toString()));
grunt = new Grunt(in, pigContext);
gruntCalled = true;
int results[] = grunt.exec();
rc = getReturnCodeForStats(results);
return;
}
default:
break;
}
// If we're here, we don't know yet what they want. They may have just
// given us a jar to execute, they might have given us a pig script to
// execute, or they might have given us a dash (or nothing) which means to
// run grunt interactive.
String remainders[] = opts.getRemainingArgs();
if (remainders == null) {
// Interactive
mode = ExecMode.SHELL;
ConsoleReader reader = new ConsoleReader(System.in, new OutputStreamWriter(System.out));
reader.setDefaultPrompt("grunt> ");
final String HISTORYFILE = ".pig_history";
String historyFile = System.getProperty("user.home") + File.separator + HISTORYFILE;
reader.setHistory(new History(new File(historyFile)));
ConsoleReaderInputStream inputStream = new ConsoleReaderInputStream(reader);
grunt = new Grunt(new BufferedReader(new InputStreamReader(inputStream)), pigContext);
grunt.setConsoleReader(reader);
gruntCalled = true;
grunt.run();
rc = 0;
return;
} else {
// They have a pig script they want us to run.
if (remainders.length > 1) {
throw new RuntimeException(
"You can only run one pig script " + "at a time from the command line.");
}
mode = ExecMode.FILE;
in = new BufferedReader(new FileReader(remainders[0]));
// run parameter substitution preprocessor first
substFile = remainders[0] + ".substituted";
pin = runParamPreprocessor(in, params, paramFiles, substFile, debug || dryrun);
if (dryrun) {
log.info("Dry run completed. Substituted pig script is at " + substFile);
return;
}
logFileName = validateLogFile(logFileName, remainders[0]);
pigContext.getProperties().setProperty("pig.logfile", logFileName);
if (!debug) {
new File(substFile).deleteOnExit();
}
// Set job name based on name of the script
pigContext
.getProperties()
.setProperty(PigContext.JOB_NAME, "PigLatin:" + new File(remainders[0]).getName());
grunt = new Grunt(pin, pigContext);
gruntCalled = true;
int[] results = grunt.exec();
rc = getReturnCodeForStats(results);
return;
}
// Per Utkarsh and Chris invocation of jar file via pig depricated.
} catch (ParseException e) {
usage();
rc = 2;
} catch (NumberFormatException e) {
usage();
rc = 2;
} catch (PigException pe) {
if (pe.retriable()) {
rc = 1;
} else {
rc = 2;
}
if (!gruntCalled) {
LogUtils.writeLog(pe, logFileName, log, verbose);
}
} catch (Throwable e) {
rc = 2;
if (!gruntCalled) {
LogUtils.writeLog(e, logFileName, log, verbose);
}
} finally {
// clear temp files
FileLocalizer.deleteTempFiles();
PerformanceTimerFactory.getPerfTimerFactory().dumpTimers();
System.exit(rc);
}
}