MAPREDUCE-5785. Derive heap size or mapreduce.*.memory.mb automatically. (Gera Shegalov and Karthik Kambatla via gera)

hadoop-mapreduce-project/CHANGES.txt

@@ -3,6 +3,8 @@ Hadoop MapReduce Change Log
 Trunk (Unreleased)
 
   INCOMPATIBLE CHANGES
+    MAPREDUCE-5785. Derive heap size or mapreduce.*.memory.mb automatically.
+    (Gera Shegalov and Karthik Kambatla via gera)
 
   NEW FEATURES
 

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapred/MapReduceChildJVM.java

@@ -27,6 +27,7 @@ import java.util.Vector;
 import org.apache.hadoop.fs.Path;
 import org.apache.hadoop.mapred.TaskLog.LogName;
 import org.apache.hadoop.mapreduce.MRJobConfig;
+import org.apache.hadoop.mapreduce.TaskType;
 import org.apache.hadoop.mapreduce.TypeConverter;
 import org.apache.hadoop.mapreduce.v2.util.MRApps;
 import org.apache.hadoop.yarn.api.ApplicationConstants;
@@ -99,36 +100,7 @@ public class MapReduceChildJVM {
   }
 
   private static String getChildJavaOpts(JobConf jobConf, boolean isMapTask) {
-    String userClasspath = "";
-    String adminClasspath = "";
-    if (isMapTask) {
-      userClasspath = 
-          jobConf.get(
-              JobConf.MAPRED_MAP_TASK_JAVA_OPTS, 
-              jobConf.get(
-                  JobConf.MAPRED_TASK_JAVA_OPTS, 
-                  JobConf.DEFAULT_MAPRED_TASK_JAVA_OPTS)
-          );
-      adminClasspath = 
-          jobConf.get(
-              MRJobConfig.MAPRED_MAP_ADMIN_JAVA_OPTS,
-              MRJobConfig.DEFAULT_MAPRED_ADMIN_JAVA_OPTS);
-    } else {
-      userClasspath =
-          jobConf.get(
-              JobConf.MAPRED_REDUCE_TASK_JAVA_OPTS, 
-              jobConf.get(
-                  JobConf.MAPRED_TASK_JAVA_OPTS,
-                  JobConf.DEFAULT_MAPRED_TASK_JAVA_OPTS)
-              );
-      adminClasspath =
-          jobConf.get(
-              MRJobConfig.MAPRED_REDUCE_ADMIN_JAVA_OPTS,
-              MRJobConfig.DEFAULT_MAPRED_ADMIN_JAVA_OPTS);
-    }
-    
-    // Add admin classpath first so it can be overridden by user.
-    return adminClasspath + " " + userClasspath;
+    return jobConf.getTaskJavaOpts(isMapTask ? TaskType.MAP : TaskType.REDUCE);
   }
 
   public static List<String> getVMCommand(

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/main/java/org/apache/hadoop/mapreduce/v2/app/job/impl/TaskAttemptImpl.java

@@ -563,19 +563,8 @@ public abstract class TaskAttemptImpl implements
     stateMachine = stateMachineFactory.make(this);
   }
 
-  private int getMemoryRequired(Configuration conf, TaskType taskType) {
-    int memory = 1024;
-    if (taskType == TaskType.MAP)  {
-      memory =
-          conf.getInt(MRJobConfig.MAP_MEMORY_MB,
-              MRJobConfig.DEFAULT_MAP_MEMORY_MB);
-    } else if (taskType == TaskType.REDUCE) {
-      memory =
-          conf.getInt(MRJobConfig.REDUCE_MEMORY_MB,
-              MRJobConfig.DEFAULT_REDUCE_MEMORY_MB);
-    }
-    
-    return memory;
+  private int getMemoryRequired(JobConf conf, TaskType taskType) {
+    return conf.getMemoryRequired(TypeConverter.fromYarn(taskType));
   }
 
   private int getCpuRequired(Configuration conf, TaskType taskType) {

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/test/java/org/apache/hadoop/mapreduce/v2/app/job/impl/TestMapReduceChildJVM.java

@@ -18,8 +18,10 @@
 
 package org.apache.hadoop.mapreduce.v2.app.job.impl;
 
+import java.util.ArrayList;
 import java.util.Map;
 
+import org.apache.hadoop.mapreduce.TaskType;
 import org.junit.Assert;
 
 import org.apache.commons.logging.Log;
@@ -56,8 +58,8 @@ public class TestMapReduceChildJVM {
     Assert.assertEquals(
       "[" + MRApps.crossPlatformify("JAVA_HOME") + "/bin/java" +
       " -Djava.net.preferIPv4Stack=true" +
-      " -Dhadoop.metrics.log.level=WARN" +
-      "  -Xmx200m -Djava.io.tmpdir=" + MRApps.crossPlatformify("PWD") + "/tmp" +
+      " -Dhadoop.metrics.log.level=WARN " +
+      "  -Xmx820m -Djava.io.tmpdir=" + MRApps.crossPlatformify("PWD") + "/tmp" +
       " -Dlog4j.configuration=container-log4j.properties" +
       " -Dyarn.app.container.log.dir=<LOG_DIR>" +
       " -Dyarn.app.container.log.filesize=0" +
@@ -67,7 +69,7 @@ public class TestMapReduceChildJVM {
       " attempt_0_0000_m_000000_0" +
       " 0" +
       " 1><LOG_DIR>/stdout" +
-      " 2><LOG_DIR>/stderr ]", app.myCommandLine);
+      " 2><LOG_DIR>/stderr ]", app.launchCmdList.get(0));
     
     Assert.assertTrue("HADOOP_ROOT_LOGGER not set for job",
       app.cmdEnvironment.containsKey("HADOOP_ROOT_LOGGER"));
@@ -119,8 +121,8 @@ public class TestMapReduceChildJVM {
     Assert.assertEquals(
         "[" + MRApps.crossPlatformify("JAVA_HOME") + "/bin/java" +
             " -Djava.net.preferIPv4Stack=true" +
-            " -Dhadoop.metrics.log.level=WARN" +
-            "  -Xmx200m -Djava.io.tmpdir=" + MRApps.crossPlatformify("PWD") + "/tmp" +
+            " -Dhadoop.metrics.log.level=WARN " +
+            "  -Xmx820m -Djava.io.tmpdir=" + MRApps.crossPlatformify("PWD") + "/tmp" +
             " -Dlog4j.configuration=container-log4j.properties" +
             " -Dyarn.app.container.log.dir=<LOG_DIR>" +
             " -Dyarn.app.container.log.filesize=0" +
@@ -134,7 +136,7 @@ public class TestMapReduceChildJVM {
             " attempt_0_0000_r_000000_0" +
             " 0" +
             " 1><LOG_DIR>/stdout" +
-            " 2><LOG_DIR>/stderr ]", app.myCommandLine);
+            " 2><LOG_DIR>/stderr ]", app.launchCmdList.get(0));
 
     Assert.assertTrue("HADOOP_ROOT_LOGGER not set for job",
         app.cmdEnvironment.containsKey("HADOOP_ROOT_LOGGER"));
@@ -161,8 +163,8 @@ public class TestMapReduceChildJVM {
     Assert.assertEquals(
       "[" + MRApps.crossPlatformify("JAVA_HOME") + "/bin/java" +
       " -Djava.net.preferIPv4Stack=true" +
-      " -Dhadoop.metrics.log.level=WARN" +
-      "  -Xmx200m -Djava.io.tmpdir=" + MRApps.crossPlatformify("PWD") + "/tmp" +
+      " -Dhadoop.metrics.log.level=WARN " +
+      "  -Xmx820m -Djava.io.tmpdir=" + MRApps.crossPlatformify("PWD") + "/tmp" +
       " -Dlog4j.configuration=" + testLogPropertieFile +
       " -Dyarn.app.container.log.dir=<LOG_DIR>" +
       " -Dyarn.app.container.log.filesize=0" +
@@ -172,12 +174,81 @@ public class TestMapReduceChildJVM {
       " attempt_0_0000_m_000000_0" +
       " 0" +
       " 1><LOG_DIR>/stdout" +
-      " 2><LOG_DIR>/stderr ]", app.myCommandLine);
+      " 2><LOG_DIR>/stderr ]", app.launchCmdList.get(0));
+  }
+
+  @Test
+  public void testAutoHeapSizes() throws Exception {
+    // Don't specify heap size or memory-mb
+    testAutoHeapSize(-1, -1, null);
+
+    // Don't specify heap size
+    testAutoHeapSize(512, 768, null);
+    testAutoHeapSize(100, 768, null);
+    testAutoHeapSize(512, 100, null);
+    // Specify heap size
+    testAutoHeapSize(512, 768, "-Xmx100m");
+    testAutoHeapSize(512, 768, "-Xmx500m");
+
+    // Specify heap size but not the memory
+    testAutoHeapSize(-1, -1, "-Xmx100m");
+    testAutoHeapSize(-1, -1, "-Xmx500m");
+  }
+
+  private void testAutoHeapSize(int mapMb, int redMb, String xmxArg)
+      throws Exception {
+    JobConf conf = new JobConf();
+    float heapRatio = conf.getFloat(MRJobConfig.HEAP_MEMORY_MB_RATIO,
+        MRJobConfig.DEFAULT_HEAP_MEMORY_MB_RATIO);
+
+    // Verify map and reduce java opts are not set by default
+    Assert.assertNull("Default map java opts!",
+        conf.get(MRJobConfig.MAP_JAVA_OPTS));
+    Assert.assertNull("Default reduce java opts!",
+        conf.get(MRJobConfig.REDUCE_JAVA_OPTS));
+    // Set the memory-mbs and java-opts
+    if (mapMb > 0) {
+      conf.setInt(MRJobConfig.MAP_MEMORY_MB, mapMb);
+    } else {
+      mapMb = conf.getMemoryRequired(TaskType.MAP);
+    }
+
+    if (redMb > 0) {
+      conf.setInt(MRJobConfig.REDUCE_MEMORY_MB, redMb);
+    } else {
+      redMb = conf.getMemoryRequired(TaskType.REDUCE);
+    }
+    if (xmxArg != null) {
+      conf.set(MRJobConfig.MAP_JAVA_OPTS, xmxArg);
+      conf.set(MRJobConfig.REDUCE_JAVA_OPTS, xmxArg);
+    }
+
+    // Submit job to let unspecified fields be picked up
+    MyMRApp app = new MyMRApp(1, 1, true, this.getClass().getName(), true);
+    Job job = app.submit(conf);
+    app.waitForState(job, JobState.SUCCEEDED);
+    app.verifyCompleted();
+
+    // Go through the tasks and verify the values are as expected
+    for (String cmd : app.launchCmdList) {
+      final boolean isMap = cmd.contains("_m_");
+      int heapMb;
+      if (xmxArg == null) {
+        heapMb = (int)(Math.ceil((isMap ? mapMb : redMb) * heapRatio));
+      } else {
+        final String javaOpts = conf.get(isMap
+            ? MRJobConfig.MAP_JAVA_OPTS
+            : MRJobConfig.REDUCE_JAVA_OPTS);
+        heapMb = JobConf.parseMaximumHeapSizeMB(javaOpts);
+      }
+      Assert.assertEquals("Incorrect heapsize in the command opts",
+          heapMb, JobConf.parseMaximumHeapSizeMB(cmd));
+    }
   }
 
   private static final class MyMRApp extends MRApp {
 
-    private String myCommandLine;
+    private ArrayList<String> launchCmdList = new ArrayList<>();
     private Map<String, String> cmdEnvironment;
 
     public MyMRApp(int maps, int reduces, boolean autoComplete,
@@ -196,7 +267,7 @@ public class TestMapReduceChildJVM {
                 launchEvent.getContainerLaunchContext();
             String cmdString = launchContext.getCommands().toString();
             LOG.info("launchContext " + cmdString);
-            myCommandLine = cmdString;
+            launchCmdList.add(cmdString);
             cmdEnvironment = launchContext.getEnvironment();
           }
           super.handle(event);

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapred/JobConf.java

@@ -20,8 +20,10 @@ package org.apache.hadoop.mapred;
 
 
 import java.io.IOException;
+import java.util.regex.Matcher;
 import java.util.regex.Pattern;
 
+import com.google.common.annotations.VisibleForTesting;
 import org.apache.commons.logging.Log;
 import org.apache.commons.logging.LogFactory;
 import org.apache.hadoop.classification.InterfaceAudience;
@@ -44,6 +46,7 @@ import org.apache.hadoop.mapred.lib.KeyFieldBasedComparator;
 import org.apache.hadoop.mapred.lib.KeyFieldBasedPartitioner;
 import org.apache.hadoop.mapreduce.MRConfig;
 import org.apache.hadoop.mapreduce.MRJobConfig;
+import org.apache.hadoop.mapreduce.TaskType;
 import org.apache.hadoop.mapreduce.filecache.DistributedCache;
 import org.apache.hadoop.mapreduce.util.ConfigUtil;
 import org.apache.hadoop.security.Credentials;
@@ -114,6 +117,8 @@ import org.apache.log4j.Level;
 public class JobConf extends Configuration {
 
   private static final Log LOG = LogFactory.getLog(JobConf.class);
+  private static final Pattern JAVA_OPTS_XMX_PATTERN =
+          Pattern.compile(".*(?:^|\\s)-Xmx(\\d+)([gGmMkK]?)(?:$|\\s).*");
 
   static{
     ConfigUtil.loadResources();
@@ -247,9 +252,9 @@ public class JobConf extends Configuration {
    */
   public static final String MAPRED_REDUCE_TASK_JAVA_OPTS = 
     JobContext.REDUCE_JAVA_OPTS;
-  
-  public static final String DEFAULT_MAPRED_TASK_JAVA_OPTS = "-Xmx200m";
-  
+
+  public static final String DEFAULT_MAPRED_TASK_JAVA_OPTS = "";
+
   /**
    * @deprecated
    * Configuration key to set the maximum virtual memory available to the child
@@ -2022,7 +2027,123 @@ public class JobConf extends Configuration {
       LOG.warn(JobConf.deprecatedString(JobConf.MAPRED_REDUCE_TASK_ULIMIT));
     }
   }
-  
+
+  private String getConfiguredTaskJavaOpts(TaskType taskType) {
+    String userClasspath = "";
+    String adminClasspath = "";
+    if (taskType == TaskType.MAP) {
+      userClasspath = get(MAPRED_MAP_TASK_JAVA_OPTS,
+          get(MAPRED_TASK_JAVA_OPTS, DEFAULT_MAPRED_TASK_JAVA_OPTS));
+      adminClasspath = get(MRJobConfig.MAPRED_MAP_ADMIN_JAVA_OPTS,
+          MRJobConfig.DEFAULT_MAPRED_ADMIN_JAVA_OPTS);
+    } else {
+      userClasspath = get(MAPRED_REDUCE_TASK_JAVA_OPTS,
+          get(MAPRED_TASK_JAVA_OPTS, DEFAULT_MAPRED_TASK_JAVA_OPTS));
+      adminClasspath = get(MRJobConfig.MAPRED_REDUCE_ADMIN_JAVA_OPTS,
+          MRJobConfig.DEFAULT_MAPRED_ADMIN_JAVA_OPTS);
+    }
+
+    return adminClasspath + " " + userClasspath;
+  }
+
+  @Private
+  public String getTaskJavaOpts(TaskType taskType) {
+    String javaOpts = getConfiguredTaskJavaOpts(taskType);
+
+    if (!javaOpts.contains("-Xmx")) {
+      float heapRatio = getFloat(MRJobConfig.HEAP_MEMORY_MB_RATIO,
+          MRJobConfig.DEFAULT_HEAP_MEMORY_MB_RATIO);
+
+      if (heapRatio > 1.0f || heapRatio < 0) {
+        LOG.warn("Invalid value for " + MRJobConfig.HEAP_MEMORY_MB_RATIO
+            + ", using the default.");
+        heapRatio = MRJobConfig.DEFAULT_HEAP_MEMORY_MB_RATIO;
+      }
+
+      int taskContainerMb = getMemoryRequired(taskType);
+      int taskHeapSize = (int)Math.ceil(taskContainerMb * heapRatio);
+
+      String xmxArg = String.format("-Xmx%dm", taskHeapSize);
+      LOG.info("Task java-opts do not specify heap size. Setting task attempt" +
+          " jvm max heap size to " + xmxArg);
+
+      javaOpts += " " + xmxArg;
+    }
+
+    return javaOpts;
+  }
+
+  /**
+   * Parse the Maximum heap size from the java opts as specified by the -Xmx option
+   * Format: -Xmx<size>[g|G|m|M|k|K]
+   * @param javaOpts String to parse to read maximum heap size
+   * @return Maximum heap size in MB or -1 if not specified
+   */
+  @Private
+  @VisibleForTesting
+  public static int parseMaximumHeapSizeMB(String javaOpts) {
+    // Find the last matching -Xmx following word boundaries
+    Matcher m = JAVA_OPTS_XMX_PATTERN.matcher(javaOpts);
+    if (m.matches()) {
+      int size = Integer.parseInt(m.group(1));
+      if (size <= 0) {
+        return -1;
+      }
+      if (m.group(2).isEmpty()) {
+        // -Xmx specified in bytes
+        return size / (1024 * 1024);
+      }
+      char unit = m.group(2).charAt(0);
+      switch (unit) {
+        case 'g':
+        case 'G':
+          // -Xmx specified in GB
+          return size * 1024;
+        case 'm':
+        case 'M':
+          // -Xmx specified in MB
+          return size;
+        case 'k':
+        case 'K':
+          // -Xmx specified in KB
+          return size / 1024;
+      }
+    }
+    // -Xmx not specified
+    return -1;
+  }
+
+  private int getMemoryRequiredHelper(
+      String configName, int defaultValue, int heapSize, float heapRatio) {
+    int memory = getInt(configName, -1);
+    if (memory <= 0) {
+      if (heapSize > 0) {
+        memory = (int) Math.ceil(heapSize / heapRatio);
+        LOG.info("Figured value for " + configName + " from javaOpts");
+      } else {
+        memory = defaultValue;
+      }
+    }
+
+    return memory;
+  }
+
+  @Private
+  public int getMemoryRequired(TaskType taskType) {
+    int memory = 1024;
+    int heapSize = parseMaximumHeapSizeMB(getConfiguredTaskJavaOpts(taskType));
+    float heapRatio = getFloat(MRJobConfig.HEAP_MEMORY_MB_RATIO,
+        MRJobConfig.DEFAULT_HEAP_MEMORY_MB_RATIO);
+    if (taskType == TaskType.MAP) {
+      return getMemoryRequiredHelper(MRJobConfig.MAP_MEMORY_MB,
+          MRJobConfig.DEFAULT_MAP_MEMORY_MB, heapSize, heapRatio);
+    } else if (taskType == TaskType.REDUCE) {
+      return getMemoryRequiredHelper(MRJobConfig.REDUCE_MEMORY_MB,
+          MRJobConfig.DEFAULT_REDUCE_MEMORY_MB, heapSize, heapRatio);
+    } else {
+      return memory;
+    }
+  }
 
 }
 

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapred/Task.java

@@ -34,6 +34,7 @@ import java.util.concurrent.atomic.AtomicBoolean;
 
 import javax.crypto.SecretKey;
 
+import com.google.common.annotations.VisibleForTesting;
 import org.apache.commons.logging.Log;
 import org.apache.commons.logging.LogFactory;
 import org.apache.hadoop.classification.InterfaceAudience;
@@ -624,8 +625,9 @@ abstract public class Task implements Writable, Configurable {
      * Using AtomicBoolean since we need an atomic read & reset method. 
      */  
     private AtomicBoolean progressFlag = new AtomicBoolean(false);
-    
-    TaskReporter(Progress taskProgress,
+
+    @VisibleForTesting
+    public TaskReporter(Progress taskProgress,
                  TaskUmbilicalProtocol umbilical) {
       this.umbilical = umbilical;
       this.taskProgress = taskProgress;

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/java/org/apache/hadoop/mapreduce/MRJobConfig.java

@@ -810,6 +810,11 @@ public interface MRJobConfig {
   public static final String TASK_PREEMPTION =
       "mapreduce.job.preemption";
 
+  public static final String HEAP_MEMORY_MB_RATIO =
+      "mapreduce.job.heap.memory-mb.ratio";
+
+  public static final float DEFAULT_HEAP_MEMORY_MB_RATIO = 0.8f;
+
   public static final String MR_ENCRYPTED_INTERMEDIATE_DATA =
       "mapreduce.job.encrypted-intermediate-data";
   public static final boolean DEFAULT_MR_ENCRYPTED_INTERMEDIATE_DATA = false;

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-core/src/main/resources/mapred-default.xml

@@ -208,9 +208,11 @@
 
 <property>
   <name>mapreduce.map.memory.mb</name>
-  <value>1024</value>
+  <value>-1</value>
   <description>The amount of memory to request from the scheduler for each
-  map task.
+    map task. If this is not specified or is non-positive, it is inferred from
+    mapreduce.map.java.opts and mapreduce.job.heap.memory-mb.ratio.
+    If java-opts are also not specified, we set it to 1024.
   </description>
 </property>
 
@@ -224,9 +226,11 @@
 
 <property>
   <name>mapreduce.reduce.memory.mb</name>
-  <value>1024</value>
+  <value>-1</value>
   <description>The amount of memory to request from the scheduler for each
-  reduce task.
+    reduce task. If this is not specified or is non-positive, it is inferred
+    from mapreduce.reduce.java.opts and mapreduce.job.heap.memory-mb.ratio.
+    If java-opts are also not specified, we set it to 1024.
   </description>
 </property>
 
@@ -240,7 +244,7 @@
 
 <property>
   <name>mapred.child.java.opts</name>
-  <value>-Xmx200m</value>
+  <value></value>
   <description>Java opts for the task processes.
   The following symbol, if present, will be interpolated: @taskid@ is replaced 
   by current TaskID. Any other occurrences of '@' will go unchanged.
@@ -251,7 +255,10 @@
   Usage of -Djava.library.path can cause programs to no longer function if
   hadoop native libraries are used. These values should instead be set as part 
   of LD_LIBRARY_PATH in the map / reduce JVM env using the mapreduce.map.env and 
-  mapreduce.reduce.env config settings. 
+  mapreduce.reduce.env config settings.
+
+  If -Xmx is not set, it is inferred from mapreduce.{map|reduce}.memory.mb and
+  mapreduce.job.heap.memory-mb.ratio.
   </description>
 </property>
 
@@ -260,7 +267,9 @@
   <name>mapreduce.map.java.opts</name>
   <value></value>
   <description>Java opts only for the child processes that are maps. If set,
-  this will be used instead of mapred.child.java.opts.
+  this will be used instead of mapred.child.java.opts. If -Xmx is not set,
+  it is inferred from mapreduce.map.memory.mb and
+  mapreduce.job.heap.memory-mb.ratio.
   </description>
 </property>
 -->
@@ -270,7 +279,9 @@
   <name>mapreduce.reduce.java.opts</name>
   <value></value>
   <description>Java opts only for the child processes that are reduces. If set,
-  this will be used instead of mapred.child.java.opts.
+  this will be used instead of mapred.child.java.opts. If -Xmx is not set,
+  it is inferred from mapreduce.reduce.memory.mb and
+  mapreduce.job.heap.memory-mb.ratio.
   </description>
 </property>
 -->
@@ -1567,4 +1578,15 @@
     - HTTPS_ONLY : Service is provided only on https
   </description>
 </property>
+
+<property>
+  <name>mapreduce.job.heap.memory-mb.ratio</name>
+  <value>0.8</value>
+  <description>The ratio of heap-size to container-size. If no -Xmx is
+    specified, it is calculated as
+    (mapreduce.{map|reduce}.memory.mb * mapreduce.heap.memory-mb.ratio).
+    If -Xmx is specified but not mapreduce.{map|reduce}.memory.mb, it is
+    calculated as (heapSize / mapreduce.heap.memory-mb.ratio).
+  </description>
+</property>
 </configuration>