HDFS-6109 let sync_file_range() system call run in background (Liang Xie via stack)

git-svn-id: https://svn.apache.org/repos/asf/hadoop/common/branches/branch-2@1599351 13f79535-47bb-0310-9956-ffa450edef68

hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt

@@ -144,6 +144,9 @@ Release 2.5.0 - UNRELEASED
 
     HDFS-6056. Clean up NFS config settings (brandonli)
 
+    HDFS-6109 let sync_file_range() system call run in background
+    (Liang Xie via stack)
+
   OPTIMIZATIONS
 
     HDFS-6214. Webhdfs has poor throughput for files >2GB (daryn)

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/DFSConfigKeys.java

@@ -111,6 +111,8 @@ public class DFSConfigKeys extends CommonConfigurationKeys {
   public static final boolean DFS_DATANODE_DROP_CACHE_BEHIND_WRITES_DEFAULT = false;
   public static final String  DFS_DATANODE_SYNC_BEHIND_WRITES_KEY = "dfs.datanode.sync.behind.writes";
   public static final boolean DFS_DATANODE_SYNC_BEHIND_WRITES_DEFAULT = false;
+  public static final String  DFS_DATANODE_SYNC_BEHIND_WRITES_IN_BACKGROUND_KEY = "dfs.datanode.sync.behind.writes.in.background";
+  public static final boolean DFS_DATANODE_SYNC_BEHIND_WRITES_IN_BACKGROUND_DEFAULT = false;
   public static final String  DFS_DATANODE_DROP_CACHE_BEHIND_READS_KEY = "dfs.datanode.drop.cache.behind.reads";
   public static final boolean DFS_DATANODE_DROP_CACHE_BEHIND_READS_DEFAULT = false;
   public static final String  DFS_DATANODE_USE_DN_HOSTNAME = "dfs.datanode.use.datanode.hostname";

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/BlockReceiver.java

@@ -104,6 +104,7 @@ class BlockReceiver implements Closeable {
   private boolean dropCacheBehindWrites;
   private long lastCacheManagementOffset = 0;
   private boolean syncBehindWrites;
+  private boolean syncBehindWritesInBackground;
 
   /** The client name.  It is empty if a datanode is the client */
   private final String clientname;
@@ -207,6 +208,8 @@ class BlockReceiver implements Closeable {
         datanode.getDnConf().dropCacheBehindWrites :
           cachingStrategy.getDropBehind();
       this.syncBehindWrites = datanode.getDnConf().syncBehindWrites;
+      this.syncBehindWritesInBackground = datanode.getDnConf().
+          syncBehindWritesInBackground;
       
       final boolean isCreate = isDatanode || isTransfer 
           || stage == BlockConstructionStage.PIPELINE_SETUP_CREATE;
@@ -668,10 +671,17 @@ class BlockReceiver implements Closeable {
         // of file                 
         //
         if (syncBehindWrites) {
-          NativeIO.POSIX.syncFileRangeIfPossible(outFd,
+          if (syncBehindWritesInBackground) {
-              lastCacheManagementOffset,
+            this.datanode.getFSDataset().submitBackgroundSyncFileRangeRequest(
+                block, outFd, lastCacheManagementOffset,
                 offsetInBlock - lastCacheManagementOffset,
                 NativeIO.POSIX.SYNC_FILE_RANGE_WRITE);
+          } else {
+            NativeIO.POSIX.syncFileRangeIfPossible(outFd,
+                lastCacheManagementOffset, offsetInBlock
+                    - lastCacheManagementOffset,
+                NativeIO.POSIX.SYNC_FILE_RANGE_WRITE);
+          }
         }
         //
         // For POSIX_FADV_DONTNEED, we want to drop from the beginning 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/DNConf.java

@@ -67,6 +67,7 @@ public class DNConf {
   final boolean transferToAllowed;
   final boolean dropCacheBehindWrites;
   final boolean syncBehindWrites;
+  final boolean syncBehindWritesInBackground;
   final boolean dropCacheBehindReads;
   final boolean syncOnClose;
   final boolean encryptDataTransfer;
@@ -119,6 +120,9 @@ public class DNConf {
     syncBehindWrites = conf.getBoolean(
         DFSConfigKeys.DFS_DATANODE_SYNC_BEHIND_WRITES_KEY,
         DFSConfigKeys.DFS_DATANODE_SYNC_BEHIND_WRITES_DEFAULT);
+    syncBehindWritesInBackground = conf.getBoolean(
+        DFSConfigKeys.DFS_DATANODE_SYNC_BEHIND_WRITES_IN_BACKGROUND_KEY,
+        DFSConfigKeys.DFS_DATANODE_SYNC_BEHIND_WRITES_IN_BACKGROUND_DEFAULT);
     dropCacheBehindReads = conf.getBoolean(
         DFSConfigKeys.DFS_DATANODE_DROP_CACHE_BEHIND_READS_KEY,
         DFSConfigKeys.DFS_DATANODE_DROP_CACHE_BEHIND_READS_DEFAULT);

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/FsDatasetSpi.java

@@ -19,6 +19,7 @@ package org.apache.hadoop.hdfs.server.datanode.fsdataset;
 
 
 import java.io.File;
+import java.io.FileDescriptor;
 import java.io.IOException;
 import java.io.InputStream;
 import java.util.List;
@@ -431,5 +432,12 @@ public interface FsDatasetSpi<V extends FsVolumeSpi> extends FSDatasetMBean {
    * @return true when trash is enabled
    */
   public boolean trashEnabled(String bpid);
+
+  /**
+   * submit a sync_file_range request to AsyncDiskService
+   */
+  public void submitBackgroundSyncFileRangeRequest(final ExtendedBlock block,
+      final FileDescriptor fd, final long offset, final long nbytes,
+      final int flags);
 }
 

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/impl/FsDatasetAsyncDiskService.java

@@ -19,6 +19,7 @@
 package org.apache.hadoop.hdfs.server.datanode.fsdataset.impl;
 
 import java.io.File;
+import java.io.FileDescriptor;
 import java.util.HashMap;
 import java.util.Map;
 import java.util.concurrent.LinkedBlockingQueue;
@@ -31,6 +32,8 @@ import org.apache.commons.logging.LogFactory;
 import org.apache.hadoop.hdfs.protocol.ExtendedBlock;
 import org.apache.hadoop.hdfs.server.datanode.DataNode;
 import org.apache.hadoop.hdfs.server.protocol.BlockCommand;
+import org.apache.hadoop.io.nativeio.NativeIO;
+import org.apache.hadoop.io.nativeio.NativeIOException;
 
 /**
  * This class is a container of multiple thread pools, each for a volume,
@@ -42,6 +45,7 @@ import org.apache.hadoop.hdfs.server.protocol.BlockCommand;
  * can be slow, and we don't want to use a single thread pool because that
  * is inefficient when we have more than 1 volume.  AsyncDiskService is the
  * solution for these.
+ * Another example of async disk operation is requesting sync_file_range().
  * 
  * This class and {@link org.apache.hadoop.util.AsyncDiskService} are similar.
  * They should be combined.
@@ -148,6 +152,21 @@ class FsDatasetAsyncDiskService {
     }
   }
 
+  public void submitSyncFileRangeRequest(FsVolumeImpl volume,
+      final FileDescriptor fd, final long offset, final long nbytes,
+      final int flags) {
+    execute(volume.getCurrentDir(), new Runnable() {
+      @Override
+      public void run() {
+        try {
+          NativeIO.POSIX.syncFileRangeIfPossible(fd, offset, nbytes, flags);
+        } catch (NativeIOException e) {
+          LOG.warn("sync_file_range error", e);
+        }
+      }
+    });
+  }
+
   /**
    * Delete the block file and meta file from the disk asynchronously, adjust
    * dfsUsed statistics accordingly.

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/impl/FsDatasetImpl.java

@@ -1907,5 +1907,13 @@ class FsDatasetImpl implements FsDatasetSpi<FsVolumeImpl> {
     }
     return new RollingLogsImpl(dir, prefix);
   }
+
+  @Override
+  public void submitBackgroundSyncFileRangeRequest(ExtendedBlock block,
+      FileDescriptor fd, long offset, long nbytes, int flags) {
+    FsVolumeImpl fsVolumeImpl = this.getVolume(block);
+    asyncDiskService.submitSyncFileRangeRequest(fsVolumeImpl, fd, offset,
+        nbytes, flags);
+  }
 }
 

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/datanode/SimulatedFSDataset.java

@@ -18,6 +18,7 @@
 package org.apache.hadoop.hdfs.server.datanode;
 
 import java.io.File;
+import java.io.FileDescriptor;
 import java.io.IOException;
 import java.io.InputStream;
 import java.io.OutputStream;
@@ -1112,5 +1113,11 @@ public class SimulatedFSDataset implements FsDatasetSpi<FsVolumeSpi> {
   public FsVolumeSpi getVolume(ExtendedBlock b) {
     throw new UnsupportedOperationException();
   }
+
+  @Override
+  public void submitBackgroundSyncFileRangeRequest(ExtendedBlock block,
+      FileDescriptor fd, long offset, long nbytes, int flags) {
+    throw new UnsupportedOperationException();
+  }
 }