HDFS-11056. Concurrent append and read operations lead to checksum error. Contributed by Wei-Chiu Chuang.

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

@@ -1170,7 +1170,30 @@ public ReplicaHandler append(ExtendedBlock b,
       return new ReplicaHandler(replica, ref);
     }
   }
-  
+
+
+  private byte[] loadLastPartialChunkChecksum(
+      File blockFile, File metaFile) throws IOException {
+    DataChecksum dcs = BlockMetadataHeader.readHeader(metaFile).getChecksum();
+    final int checksumSize = dcs.getChecksumSize();
+    final long onDiskLen = blockFile.length();
+    final int bytesPerChecksum = dcs.getBytesPerChecksum();
+
+    if (onDiskLen % bytesPerChecksum == 0) {
+      // the last chunk is a complete one. No need to preserve its checksum
+      // because it will not be modified.
+      return null;
+    }
+
+    int offsetInChecksum = BlockMetadataHeader.getHeaderSize() +
+        (int)(onDiskLen / bytesPerChecksum * checksumSize);
+    byte[] lastChecksum = new byte[checksumSize];
+    RandomAccessFile raf = new RandomAccessFile(metaFile, "r");
+    raf.seek(offsetInChecksum);
+    raf.read(lastChecksum, 0, checksumSize);
+    return lastChecksum;
+  }
+
   /** Append to a finalized replica
    * Change a finalized replica to be a RBW replica and 
    * bump its generation stamp to be the newGS
@@ -1207,6 +1230,13 @@ private ReplicaBeingWritten append(String bpid,
       ReplicaBeingWritten newReplicaInfo = new ReplicaBeingWritten(
           replicaInfo.getBlockId(), replicaInfo.getNumBytes(), newGS,
           v, newBlkFile.getParentFile(), Thread.currentThread(), bytesReserved);
+
+      // load last checksum and datalen
+      byte[] lastChunkChecksum = loadLastPartialChunkChecksum(
+          replicaInfo.getBlockFile(), replicaInfo.getMetaFile());
+      newReplicaInfo.setLastChecksumAndDataLen(
+          replicaInfo.getNumBytes(), lastChunkChecksum);
+
       File newmeta = newReplicaInfo.getMetaFile();
 
       // rename meta file to rbw directory
@@ -1586,6 +1616,12 @@ public ReplicaInPipeline convertTemporaryToRbw(
           blockId, numBytes, expectedGs,
           v, dest.getParentFile(), Thread.currentThread(), 0);
       rbw.setBytesAcked(visible);
+
+      // load last checksum and datalen
+      final File destMeta = FsDatasetUtil.getMetaFile(dest,
+          b.getGenerationStamp());
+      byte[] lastChunkChecksum = loadLastPartialChunkChecksum(dest, destMeta);
+      rbw.setLastChecksumAndDataLen(numBytes, lastChunkChecksum);
       // overwrite the RBW in the volume map
       volumeMap.add(b.getBlockPoolId(), rbw);
       return rbw;
@@ -2668,6 +2704,9 @@ private FinalizedReplica updateReplicaUnderRecovery(
             newBlockId, recoveryId, volume, blockFile.getParentFile(),
             newlength);
         newReplicaInfo.setNumBytes(newlength);
+        // In theory, this rbw replica needs to reload last chunk checksum,
+        // but it is immediately converted to finalized state within the same
+        // lock, so no need to update it.
         volumeMap.add(bpid, newReplicaInfo);
         finalizeReplica(bpid, newReplicaInfo);
       }

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/TestFileAppend.java

@@ -24,10 +24,12 @@
 import java.io.File;
 import java.io.FileNotFoundException;
 import java.io.IOException;
+import java.io.OutputStream;
 import java.net.InetSocketAddress;
 import java.util.EnumSet;
 import java.util.List;
 import java.util.concurrent.ThreadLocalRandom;
+import java.util.concurrent.TimeoutException;
 
 import org.apache.hadoop.conf.Configuration;
 import org.apache.hadoop.fs.BlockLocation;
@@ -44,10 +46,16 @@
 import org.apache.hadoop.hdfs.protocol.LocatedBlock;
 import org.apache.hadoop.hdfs.protocol.LocatedBlocks;
 import org.apache.hadoop.hdfs.server.datanode.DataNode;
+import org.apache.hadoop.hdfs.server.datanode.DataNodeTestUtils;
+import org.apache.hadoop.hdfs.server.datanode.ReplicaBeingWritten;
+import org.apache.hadoop.hdfs.server.datanode.ReplicaHandler;
 import org.apache.hadoop.hdfs.server.datanode.fsdataset.FsDatasetSpi;
+import org.apache.hadoop.hdfs.server.datanode.fsdataset.ReplicaOutputStreams;
 import org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetTestUtil;
+import org.apache.hadoop.hdfs.server.datanode.fsdataset.impl.FsDatasetUtil;
 import org.apache.hadoop.io.IOUtils;
 import org.apache.hadoop.ipc.RemoteException;
+import org.apache.hadoop.util.DataChecksum;
 import org.apache.hadoop.util.Time;
 import org.junit.Assert;
 import org.junit.Test;
@@ -61,6 +69,8 @@ public class TestFileAppend{
 
   private static byte[] fileContents = null;
 
+  static final DataChecksum DEFAULT_CHECKSUM =
+      DataChecksum.newDataChecksum(DataChecksum.Type.CRC32C, 512);
   //
   // writes to file but does not close it
   //
@@ -656,4 +666,63 @@ public void testAppendCorruptedBlock() throws Exception {
       cluster.shutdown();
     }
   }
+
+  @Test(timeout = 10000)
+  public void testConcurrentAppendRead()
+      throws IOException, TimeoutException, InterruptedException {
+    // Create a finalized replica and append to it
+    // Read block data and checksum. Verify checksum.
+    Configuration conf = new HdfsConfiguration();
+    conf.setInt(DFSConfigKeys.DFS_BLOCK_SIZE_KEY, 1024);
+    conf.setInt(DFSConfigKeys.DFS_REPLICATION_KEY, 1);
+    conf.setInt("dfs.min.replication", 1);
+
+    MiniDFSCluster cluster = new MiniDFSCluster.Builder(conf).build();
+    try {
+      cluster.waitActive();
+      DataNode dn = cluster.getDataNodes().get(0);
+      FsDatasetSpi<?> dataSet = DataNodeTestUtils.getFSDataset(dn);
+
+      // create a file with 1 byte of data.
+      long initialFileLength = 1;
+      DistributedFileSystem fs = cluster.getFileSystem();
+      Path fileName = new Path("/appendCorruptBlock");
+      DFSTestUtil.createFile(fs, fileName, initialFileLength, (short) 1, 0);
+      DFSTestUtil.waitReplication(fs, fileName, (short) 1);
+      Assert.assertTrue("File not created", fs.exists(fileName));
+
+      // Call FsDatasetImpl#append to append the block file,
+      // which converts it to a rbw replica.
+      ExtendedBlock block = DFSTestUtil.getFirstBlock(fs, fileName);
+      long newGS = block.getGenerationStamp() + 1;
+      ReplicaHandler replicaHandler =
+          dataSet.append(block, newGS, initialFileLength);
+
+      // write data to block file
+      ReplicaBeingWritten rbw =
+          (ReplicaBeingWritten) replicaHandler.getReplica();
+      ReplicaOutputStreams outputStreams =
+          rbw.createStreams(false, DEFAULT_CHECKSUM);
+      OutputStream dataOutput = outputStreams.getDataOut();
+
+      byte[] appendBytes = new byte[1];
+      dataOutput.write(appendBytes, 0, 1);
+      dataOutput.flush();
+      dataOutput.close();
+
+      // update checksum file
+      final int smallBufferSize = DFSUtilClient.getSmallBufferSize(conf);
+      FsDatasetUtil.computeChecksum(rbw.getMetaFile(), rbw.getMetaFile(),
+          rbw.getBlockFile(), smallBufferSize, conf);
+
+      // read the block
+      // the DataNode BlockSender should read from the rbw replica's in-memory
+      // checksum, rather than on-disk checksum. Otherwise it will see a
+      // checksum mismatch error.
+      final byte[] readBlock = DFSTestUtil.readFileBuffer(fs, fileName);
+      assertEquals("should have read only one byte!", 1, readBlock.length);
+    } finally {
+      cluster.shutdown();
+    }
+  }
 }

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/datanode/fsdataset/impl/FsDatasetImplTestUtils.java

@@ -28,6 +28,7 @@
 import org.apache.hadoop.hdfs.protocol.Block;
 import org.apache.hadoop.fs.FileUtil;
 import org.apache.hadoop.hdfs.protocol.ExtendedBlock;
+import org.apache.hadoop.hdfs.server.datanode.BlockMetadataHeader;
 import org.apache.hadoop.hdfs.server.datanode.DataNode;
 import org.apache.hadoop.hdfs.server.datanode.DataStorage;
 import org.apache.hadoop.hdfs.server.datanode.DatanodeUtil;
@@ -43,10 +44,13 @@
 import org.apache.hadoop.hdfs.server.datanode.fsdataset.FsDatasetSpi.FsVolumeReferences;
 import org.apache.hadoop.hdfs.server.datanode.fsdataset.FsVolumeSpi;
 import org.apache.hadoop.test.GenericTestUtils;
+import org.apache.hadoop.util.DataChecksum;
 import org.apache.log4j.Level;
 
+import java.io.DataOutputStream;
 import java.io.File;
 import java.io.FileNotFoundException;
+import java.io.FileOutputStream;
 import java.io.IOException;
 import java.io.RandomAccessFile;
 import java.nio.channels.FileChannel;
@@ -67,6 +71,8 @@ public class FsDatasetImplTestUtils implements FsDatasetTestUtils {
       LogFactory.getLog(FsDatasetImplTestUtils.class);
   private final FsDatasetImpl dataset;
 
+  private static final DataChecksum DEFAULT_CHECKSUM =
+      DataChecksum.newDataChecksum(DataChecksum.Type.CRC32C, 512);
   /**
    * By default we assume 2 data directories (volumes) per DataNode.
    */
@@ -245,9 +251,18 @@ public Replica createFinalizedReplica(FsVolumeSpi volume, ExtendedBlock block)
     dataset.volumeMap.add(block.getBlockPoolId(), info);
     info.getBlockFile().createNewFile();
     info.getMetaFile().createNewFile();
+    saveMetaFileHeader(info.getMetaFile());
     return info;
   }
 
+  private void saveMetaFileHeader(File metaFile) throws IOException {
+    DataOutputStream metaOut = new DataOutputStream(
+        new FileOutputStream(metaFile));
+    BlockMetadataHeader.writeHeader(metaOut, DEFAULT_CHECKSUM);
+    metaOut.close();
+  }
+
+
   @Override
   public Replica createReplicaInPipeline(ExtendedBlock block)
       throws IOException {