HDDS-882. Provide a config to optionally turn on/off the sync flag during chunk writes. Contributed by Shashikant Banerjee.

hadoop-hdds/common/src/main/java/org/apache/hadoop/ozone/OzoneConfigKeys.java

@@ -52,6 +52,9 @@ public final class OzoneConfigKeys {
   public static final boolean DFS_CONTAINER_IPC_RANDOM_PORT_DEFAULT =
       false;
 
+  public static final String DFS_CONTAINER_CHUNK_WRITE_SYNC_KEY =
+      "dfs.container.chunk.write.sync";
+  public static final boolean DFS_CONTAINER_CHUNK_WRITE_SYNC_DEFAULT = true;
   /**
    * Ratis Port where containers listen to.
    */

hadoop-hdds/common/src/main/resources/ozone-default.xml

@@ -52,6 +52,14 @@
       running unit tests.
     </description>
   </property>
+  <property>
+    <name>dfs.container.chunk.write.sync</name>
+    <value>true</value>
+    <tag>OZONE, CONTAINER, MANAGEMENT</tag>
+    <description>Determines whether the chunk writes in the container happen as
+      sync I/0 or buffered I/O operation.
+    </description>
+  </property>
   <property>
     <name>dfs.container.ratis.statemachinedata.sync.timeout</name>
     <value>10s</value>

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/keyvalue/KeyValueHandler.java

@@ -49,6 +49,7 @@
 import org.apache.hadoop.hdds.scm.ScmConfigKeys;
 import org.apache.hadoop.hdds.scm.container.common.helpers
     .StorageContainerException;
+import org.apache.hadoop.ozone.OzoneConfigKeys;
 import org.apache.hadoop.ozone.container.common.helpers.BlockData;
 import org.apache.hadoop.ozone.container.common.helpers.ChunkInfo;
 import org.apache.hadoop.ozone.container.common.helpers.ContainerMetrics;
@@ -111,13 +112,17 @@ public class KeyValueHandler extends Handler {
   private final VolumeChoosingPolicy volumeChoosingPolicy;
   private final long maxContainerSize;
   private final AutoCloseableLock handlerLock;
+  private final boolean doSyncWrite;
 
   public KeyValueHandler(Configuration config, StateContext context,
       ContainerSet contSet, VolumeSet volSet, ContainerMetrics metrics) {
     super(config, context, contSet, volSet, metrics);
     containerType = ContainerType.KeyValueContainer;
     blockManager = new BlockManagerImpl(config);
-    chunkManager = new ChunkManagerImpl();
+    doSyncWrite =
+        conf.getBoolean(OzoneConfigKeys.DFS_CONTAINER_CHUNK_WRITE_SYNC_KEY,
+            OzoneConfigKeys.DFS_CONTAINER_CHUNK_WRITE_SYNC_DEFAULT);
+    chunkManager = new ChunkManagerImpl(doSyncWrite);
     long svcInterval = config
         .getTimeDuration(OZONE_BLOCK_DELETING_SERVICE_INTERVAL,
             OZONE_BLOCK_DELETING_SERVICE_INTERVAL_DEFAULT,

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/keyvalue/helpers/ChunkUtils.java

@@ -67,10 +67,11 @@ private ChunkUtils() {
    * @param chunkInfo - Data stream to write.
    * @param data - The data buffer.
    * @param volumeIOStats
+   * @param sync whether to do fsync or not
    * @throws StorageContainerException
    */
   public static void writeData(File chunkFile, ChunkInfo chunkInfo,
-                               ByteBuffer data, VolumeIOStats volumeIOStats)
+      ByteBuffer data, VolumeIOStats volumeIOStats, boolean sync)
       throws StorageContainerException, ExecutionException,
       InterruptedException, NoSuchAlgorithmException {
     int bufferSize = data.capacity();
@@ -88,12 +89,16 @@ public static void writeData(File chunkFile, ChunkInfo chunkInfo,
 
     try {
       long writeTimeStart = Time.monotonicNow();
-      file =
+      file = sync ?
           AsynchronousFileChannel.open(chunkFile.toPath(),
               StandardOpenOption.CREATE,
               StandardOpenOption.WRITE,
               StandardOpenOption.SPARSE,
-              StandardOpenOption.SYNC);
+              StandardOpenOption.SYNC) :
+          AsynchronousFileChannel.open(chunkFile.toPath(),
+              StandardOpenOption.CREATE,
+              StandardOpenOption.WRITE,
+              StandardOpenOption.SPARSE);
       lock = file.lock().get();
       int size = file.write(data, chunkInfo.getOffset()).get();
       // Increment volumeIO stats here.

hadoop-hdds/container-service/src/main/java/org/apache/hadoop/ozone/container/keyvalue/impl/ChunkManagerImpl.java

@@ -54,6 +54,11 @@
  */
 public class ChunkManagerImpl implements ChunkManager {
   static final Logger LOG = LoggerFactory.getLogger(ChunkManagerImpl.class);
+  private final boolean doSyncWrite;
+
+  public ChunkManagerImpl(boolean sync) {
+    doSyncWrite = sync;
+  }
 
   /**
    * writes a given chunk.
@@ -115,7 +120,8 @@ public void writeChunk(Container container, BlockID blockID, ChunkInfo info,
               "tmpChunkFile already exists" + tmpChunkFile + "Overwriting it.");
         }
         // Initially writes to temporary chunk file.
-        ChunkUtils.writeData(tmpChunkFile, info, data, volumeIOStats);
+        ChunkUtils
+            .writeData(tmpChunkFile, info, data, volumeIOStats, doSyncWrite);
         // No need to increment container stats here, as still data is not
         // committed here.
         break;
@@ -139,7 +145,7 @@ public void writeChunk(Container container, BlockID blockID, ChunkInfo info,
         break;
       case COMBINED:
         // directly write to the chunk file
-        ChunkUtils.writeData(chunkFile, info, data, volumeIOStats);
+        ChunkUtils.writeData(chunkFile, info, data, volumeIOStats, doSyncWrite);
         if (!isOverwrite) {
           containerData.incrBytesUsed(info.getLen());
         }

hadoop-hdds/container-service/src/test/java/org/apache/hadoop/ozone/container/keyvalue/TestChunkManagerImpl.java

@@ -99,7 +99,7 @@ public void setUp() throws Exception {
         .getLocalID(), 0), 0, data.length);
 
     // Create a ChunkManager object.
-    chunkManager = new ChunkManagerImpl();
+    chunkManager = new ChunkManagerImpl(true);
 
   }
 

hadoop-ozone/integration-test/src/test/java/org/apache/hadoop/ozone/container/common/impl/TestContainerPersistence.java

@@ -131,7 +131,7 @@ public void setupPaths() throws IOException {
     containerSet = new ContainerSet();
     volumeSet = new VolumeSet(DATANODE_UUID, conf);
     blockManager = new BlockManagerImpl(conf);
-    chunkManager = new ChunkManagerImpl();
+    chunkManager = new ChunkManagerImpl(true);
 
     for (String dir : conf.getStrings(ScmConfigKeys.HDDS_DATANODE_DIR_KEY)) {
       StorageLocation location = StorageLocation.parse(dir);