@Before
public void setUp() throws Exception {
curDir = System.getProperty("user.dir");
inpDir = curDir + File.separatorChar + "test/org/apache/pig/test/data/InputFiles/";
golDir = curDir + File.separatorChar + "test/org/apache/pig/test/data/GoldenFiles/";
if (Util.WINDOWS) {
inpDir = "/" + FileLocalizer.parseCygPath(inpDir, FileLocalizer.STYLE_WINDOWS);
golDir = "/" + FileLocalizer.parseCygPath(golDir, FileLocalizer.STYLE_WINDOWS);
}
}
@Before
public void setUp() throws ExecException {
GenPhyOp.setR(r);
GenPhyOp.setPc(pc);
// Set random seed to generate deterministic temporary paths
FileLocalizer.setR(new Random(1331L));
NodeIdGenerator.reset("");
pigServer = new PigServer(pc);
pigServerMR = new PigServer(pcMR);
}
/**
* . Given a path, which may represent a glob pattern, a directory, comma separated files/glob
* patterns or a file, this method finds the set of relevant metadata files on the storage system.
* The algorithm for finding the metadata file is as follows:
*
* <p>For each object represented by the path (either directly, or via a glob): If object is a
* directory, and path/metaname exists, use that as the metadata file. Else if parentPath/metaname
* exists, use that as the metadata file.
*
* <p>Resolving conflicts, merging the metadata, etc, is not handled by this method and should be
* taken care of by downstream code.
*
* <p>
*
* @param path Path, as passed in to a LoadFunc (may be a Hadoop glob)
* @param metaname Metadata file designation, such as .pig_schema or .pig_stats
* @param conf configuration object
* @return Set of element descriptors for all metadata files associated with the files on the
* path.
*/
protected Set<ElementDescriptor> findMetaFile(String path, String metaname, Configuration conf)
throws IOException {
Set<ElementDescriptor> metaFileSet = new HashSet<ElementDescriptor>();
String[] locations = LoadFunc.getPathStrings(path);
for (String loc : locations) {
DataStorage storage;
storage = new HDataStorage(new Path(loc).toUri(), ConfigurationUtil.toProperties(conf));
String fullPath = FileLocalizer.fullPath(loc, storage);
if (storage.isContainer(fullPath)) {
ElementDescriptor metaFilePath = storage.asElement(fullPath, metaname);
if (exists(metaFilePath)) {
metaFileSet.add(metaFilePath);
}
} else {
ElementDescriptor[] descriptors = storage.asCollection(loc);
for (ElementDescriptor descriptor : descriptors) {
ContainerDescriptor container = null;
if (descriptor instanceof HFile) {
Path descriptorPath = ((HPath) descriptor).getPath();
Path parent = descriptorPath.getParent();
container = new HDirectory((HDataStorage) storage, parent);
} else { // descriptor instanceof HDirectory
container = (HDirectory) descriptor;
}
// if no custom schema, try the parent directory
ElementDescriptor metaFilePath = storage.asElement(container, metaname);
if (exists(metaFilePath)) {
metaFileSet.add(metaFilePath);
}
}
}
}
return metaFileSet;
}
@Test
public void testReducerNumEstimationForOrderBy() throws Exception {
// Skip the test for Tez. Tez use a different mechanism.
// Equivalent test is in TestTezAutoParallelism
Assume.assumeTrue("Skip this test for TEZ", Util.isMapredExecType(cluster.getExecType()));
// use the estimation
pc.getProperties().setProperty("pig.exec.reducers.bytes.per.reducer", "100");
pc.getProperties().setProperty("pig.exec.reducers.max", "10");
String query = "a = load '/passwd';" + "b = order a by $0;" + "store b into 'output';";
PigServer ps = new PigServer(cluster.getExecType(), cluster.getProperties());
PhysicalPlan pp = Util.buildPp(ps, query);
MROperPlan mrPlan = Util.buildMRPlanWithOptimizer(pp, pc);
Configuration conf = ConfigurationUtil.toConfiguration(pc.getProperties());
JobControlCompiler jcc = new JobControlCompiler(pc, conf);
JobControl jobControl = jcc.compile(mrPlan, query);
assertEquals(2, mrPlan.size());
// first job uses a single reducer for the sampling
Util.assertParallelValues(-1, 1, -1, 1, jobControl.getWaitingJobs().get(0).getJobConf());
// Simulate the first job having run so estimation kicks in.
MapReduceOper sort = mrPlan.getLeaves().get(0);
jcc.updateMROpPlan(jobControl.getReadyJobs());
FileLocalizer.create(sort.getQuantFile(), pc);
jobControl = jcc.compile(mrPlan, query);
sort = mrPlan.getLeaves().get(0);
long reducer =
Math.min(
(long) Math.ceil(new File("test/org/apache/pig/test/data/passwd").length() / 100.0),
10);
assertEquals(reducer, sort.getRequestedParallelism());
// the second job estimates reducers
Util.assertParallelValues(
-1, -1, reducer, reducer, jobControl.getWaitingJobs().get(0).getJobConf());
// use the PARALLEL key word, it will override the estimated reducer number
query = "a = load '/passwd';" + "b = order a by $0 PARALLEL 2;" + "store b into 'output';";
pp = Util.buildPp(ps, query);
mrPlan = Util.buildMRPlanWithOptimizer(pp, pc);
assertEquals(2, mrPlan.size());
sort = mrPlan.getLeaves().get(0);
assertEquals(2, sort.getRequestedParallelism());
// the estimation won't take effect when it apply to non-dfs or the files doesn't exist, such as
// hbase
query =
"a = load 'hbase://passwd' using org.apache.pig.backend.hadoop.hbase.HBaseStorage('c:f1 c:f2');"
+ "b = order a by $0 ;"
+ "store b into 'output';";
pp = Util.buildPp(ps, query);
mrPlan = Util.buildMRPlanWithOptimizer(pp, pc);
assertEquals(2, mrPlan.size());
sort = mrPlan.getLeaves().get(0);
// the requested parallel will be -1 if users don't set any of default_parallel, paralllel
// and the estimation doesn't take effect. MR framework will finally set it to 1.
assertEquals(-1, sort.getRequestedParallelism());
// test order by with three jobs (after optimization)
query =
"a = load '/passwd';"
+ "b = foreach a generate $0, $1, $2;"
+ "c = order b by $0;"
+ "store c into 'output';";
pp = Util.buildPp(ps, query);
mrPlan = Util.buildMRPlanWithOptimizer(pp, pc);
assertEquals(3, mrPlan.size());
// Simulate the first 2 jobs having run so estimation kicks in.
sort = mrPlan.getLeaves().get(0);
FileLocalizer.create(sort.getQuantFile(), pc);
jobControl = jcc.compile(mrPlan, query);
Util.copyFromLocalToCluster(
cluster,
"test/org/apache/pig/test/data/passwd",
((POLoad) sort.mapPlan.getRoots().get(0)).getLFile().getFileName());
// First job is just foreach with projection, mapper-only job, so estimate gets ignored
Util.assertParallelValues(-1, -1, -1, 0, jobControl.getWaitingJobs().get(0).getJobConf());
jcc.updateMROpPlan(jobControl.getReadyJobs());
jobControl = jcc.compile(mrPlan, query);
jcc.updateMROpPlan(jobControl.getReadyJobs());
// Second job is a sampler, which requests and gets 1 reducer
Util.assertParallelValues(-1, 1, -1, 1, jobControl.getWaitingJobs().get(0).getJobConf());
jobControl = jcc.compile(mrPlan, query);
sort = mrPlan.getLeaves().get(0);
assertEquals(reducer, sort.getRequestedParallelism());
// Third job is the order, which uses the estimated number of reducers
Util.assertParallelValues(
-1, -1, reducer, reducer, jobControl.getWaitingJobs().get(0).getJobConf());
}
/**
* 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);
}
}
/**
* This method copies all class files present in the local temp directory to the distributed
* cache. All copied files will have a symlink of their name. No files will be copied if the
* current job is being run from local mode.
*
* @param pigContext
* @param conf
*/
private void internalCopyAllGeneratedToDistributedCache() {
LOG.info("Starting process to move generated code to distributed cacche");
if (pigContext.getExecType().isLocal()) {
String codePath = codeDir.getAbsolutePath();
LOG.info(
"Distributed cache not supported or needed in local mode. Setting key ["
+ LOCAL_CODE_DIR
+ "] with code temp directory: "
+ codePath);
conf.set(LOCAL_CODE_DIR, codePath);
return;
} else {
// This let's us avoid NPE in some of the non-traditional pipelines
String codePath = codeDir.getAbsolutePath();
conf.set(LOCAL_CODE_DIR, codePath);
}
DistributedCache.createSymlink(conf); // we will read using symlinks
StringBuilder serialized = new StringBuilder();
boolean first = true;
// We attempt to copy over every file in the generated code temp directory
for (File f : codeDir.listFiles()) {
if (first) {
first = false;
} else {
serialized.append(",");
}
String symlink = f.getName(); // the class name will also be the symlink
serialized.append(symlink);
Path src = new Path(f.toURI());
Path dst;
try {
dst = FileLocalizer.getTemporaryPath(pigContext);
} catch (IOException e) {
throw new RuntimeException("Error getting temporary path in HDFS", e);
}
FileSystem fs;
try {
fs = dst.getFileSystem(conf);
} catch (IOException e) {
throw new RuntimeException("Unable to get FileSystem", e);
}
try {
fs.copyFromLocalFile(src, dst);
fs.setReplication(dst, (short) conf.getInt(MRConfiguration.SUMIT_REPLICATION, 3));
} catch (IOException e) {
throw new RuntimeException(
"Unable to copy from local filesystem to HDFS, src = " + src + ", dst = " + dst, e);
}
String destination = dst.toString() + "#" + symlink;
try {
DistributedCache.addCacheFile(new URI(destination), conf);
} catch (URISyntaxException e) {
throw new RuntimeException("Unable to add file to distributed cache: " + destination, e);
}
LOG.info("File successfully added to the distributed cache: " + symlink);
}
String toSer = serialized.toString();
LOG.info(
"Setting key ["
+ GENERATED_CLASSES_KEY
+ "] with classes to deserialize ["
+ toSer
+ "]");
// we must set a key in the job conf so individual jobs know to resolve the shipped classes
conf.set(GENERATED_CLASSES_KEY, toSer);
}