/**
* Copyright 2010-2011 TransPac Software, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.scaleunlimited.cascading.hadoop;
import java.io.IOException;
import java.util.Map;
import java.util.Properties;
import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.mapred.ClusterStatus;
import org.apache.hadoop.mapred.JobClient;
import org.apache.hadoop.mapred.JobConf;
import org.apache.hadoop.mapred.JobTracker.State;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import cascading.flow.FlowProcess;
import cascading.flow.FlowProcessWrapper;
import cascading.flow.hadoop.HadoopFlowProcess;
import cascading.flow.hadoop.util.HadoopUtil;
import cascading.property.AppProps;
import com.scaleunlimited.cascading.Level;
import com.scaleunlimited.cascading.LoggingFlowProcess;
public class HadoopUtils {
private static final Logger LOGGER = LoggerFactory.getLogger(HadoopUtils.class);
private static final long STATUS_CHECK_INTERVAL = 10000;
public static void safeRemove(FileSystem fs, Path path) {
if ((fs != null) && (path != null)) {
try {
fs.delete(path, true);
} catch (Throwable t) {
// Ignore
}
}
}
/**
* Return the number of reducers, and thus the max number of parallel reduce tasks.
*
* @param conf
* @return number of reducers
* @throws IOException
* @throws InterruptedException
*/
public static int getNumReducers(JobConf conf) throws IOException, InterruptedException {
// TODO the call to getMaxReduceTasks always returns 1 in MR2.
ClusterStatus status = safeGetClusterStatus(conf);
return status.getMaxReduceTasks();
}
public static int getTaskTrackers(JobConf conf) throws IOException, InterruptedException {
ClusterStatus status = safeGetClusterStatus(conf);
return status.getTaskTrackers();
}
public static JobConf getDefaultJobConf() throws IOException, InterruptedException {
JobConf conf = new JobConf();
// We explicitly set task counts to 1 for local so that code which depends on
// things like the reducer count runs properly.
if (isJobLocal(conf)) {
conf.setNumMapTasks(1);
conf.setNumReduceTasks(1);
} else {
conf.setNumReduceTasks(getNumReducers(conf));
// TODO - By default we want to use 0.95 * the number of reduce slots, as per
// Hadoop wiki. But we want to round, versus truncate, to avoid setting it to
// 0 if we have one reducer. This way it only impacts you if you have more
// than 10 reducers.
// conf.setNumReduceTasks((getNumReducers(conf) * 95) / 100);
}
conf.setMapSpeculativeExecution(false);
conf.setReduceSpeculativeExecution(false);
return conf;
}
public static void setLoggingProperties(Properties props, Level cascadingLevel, Level bixoLevel) {
props.put("log4j.logger", String.format("cascading=%s,bixo=%s", cascadingLevel, bixoLevel));
}
public static Map<Object, Object> getDefaultProperties(Class appJarClass, boolean debugging, JobConf conf) {
Map<Object, Object> properties = HadoopUtil.createProperties(conf);
// Use special Cascading hack to control logging levels for code running as Hadoop jobs
if (debugging) {
properties.put("log4j.logger", "cascading=DEBUG,bixo=TRACE");
} else {
properties.put("log4j.logger", "cascading=INFO,bixo=INFO");
}
AppProps.setApplicationJarClass(properties, appJarClass);
return properties;
}
public static boolean isJobLocal(JobConf conf) {
return isConfigLocal(conf);
}
public static boolean isConfigLocal(Configuration conf) {
// First see if we have the new MR2 setting
String hostname = conf.get("yarn.resourcemanager.hostname");
if (hostname != null) {
return hostname.equals("0.0.0.0");
} else {
// MR1 approach
return conf.get("mapred.job.tracker").equalsIgnoreCase("local");
}
}
@SuppressWarnings("rawtypes")
public static boolean isHadoopFlowProcess(FlowProcess fp) {
return (undelegate(fp) instanceof HadoopFlowProcess);
}
@SuppressWarnings("rawtypes")
public static FlowProcess undelegate(FlowProcess fp) {
FlowProcess delegate = fp;
if (delegate instanceof LoggingFlowProcess) {
delegate = ((LoggingFlowProcess)delegate).getDelegate();
}
int delegateNestingLevel = 0;
while (delegate instanceof FlowProcessWrapper) {
if (++delegateNestingLevel > 100) {
throw new RuntimeException("FlowProcessWrapper seems to have circular nesting references");
}
delegate = FlowProcessWrapper.undelegate(delegate);
}
return delegate;
}
/**
* Utility routine that tries to ensure the cluster is "stable" (slaves have reported in) so
* that it's safe to call things like maxReduceTasks.
*
* @param conf
* @return
* @throws IOException
* @throws InterruptedException
*/
@SuppressWarnings("deprecation")
private static ClusterStatus safeGetClusterStatus(JobConf conf) throws IOException, InterruptedException {
JobClient jobClient = new JobClient(conf);
int numTaskTrackers = -1;
while (true) {
ClusterStatus status = jobClient.getClusterStatus();
// TODO there isn't a "job tracker" in MR2, just a resource manager, and a transient
// application manager for running a Hadoop job.
if (status.getJobTrackerState() == State.RUNNING) {
int curTaskTrackers = status.getTaskTrackers();
if (curTaskTrackers == numTaskTrackers) {
return status;
} else {
// Things are still settling down, so keep looping.
if (numTaskTrackers != -1) {
LOGGER.trace(String.format("Got incremental update to number of task trackers (%d to %d)", numTaskTrackers, curTaskTrackers));
}
numTaskTrackers = curTaskTrackers;
}
}
if (!isJobLocal(conf)) {
LOGGER.trace("Sleeping during status check");
Thread.sleep(STATUS_CHECK_INTERVAL);
}
}
}
}