HDFS-11160. VolumeScanner reports write-in-progress replicas as corrupt incorrectly. Contributed by Wei-Chiu Chuang and Yongjun Zhang.

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

@@ -64,7 +64,15 @@ public class BlockScanner {
   /**
    * The scanner configuration.
    */
-  private final Conf conf;
+  private Conf conf;
+
+  @VisibleForTesting
+  void setConf(Conf conf) {
+    this.conf = conf;
+    for (Entry<String, VolumeScanner> entry : scanners.entrySet()) {
+      entry.getValue().setConf(conf);
+    }
+  }
 
   /**
    * The cached scanner configuration.

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

@@ -241,13 +241,22 @@ class BlockSender implements java.io.Closeable {
             "If verifying checksum, currently must also send it.");
       }
 
+      // if there is a append write happening right after the BlockSender
+      // is constructed, the last partial checksum maybe overwritten by the
+      // append, the BlockSender need to use the partial checksum before
+      // the append write.
+      ChunkChecksum chunkChecksum = null;
       final long replicaVisibleLength;
       try(AutoCloseableLock lock = datanode.data.acquireDatasetLock()) {
         replica = getReplica(block, datanode);
         replicaVisibleLength = replica.getVisibleLength();
+        if (replica instanceof FinalizedReplica) {
+          // Load last checksum in case the replica is being written
+          // concurrently
+          final FinalizedReplica frep = (FinalizedReplica) replica;
+          chunkChecksum = frep.getLastChecksumAndDataLen();
+        }
       }
-      // if there is a write in progress
-      ChunkChecksum chunkChecksum = null;
       if (replica.getState() == ReplicaState.RBW) {
         final ReplicaInPipeline rbw = (ReplicaInPipeline) replica;
         waitForMinLength(rbw, startOffset + length);
@@ -547,7 +556,6 @@ class BlockSender implements java.io.Closeable {
       if (lastDataPacket && lastChunkChecksum != null) {
         int start = checksumOff + checksumDataLen - checksumSize;
         byte[] updatedChecksum = lastChunkChecksum.getChecksum();
-        
         if (updatedChecksum != null) {
           System.arraycopy(updatedChecksum, 0, buf, start, checksumSize);
         }

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

@@ -18,6 +18,8 @@
 package org.apache.hadoop.hdfs.server.datanode;
 
 import java.io.File;
+import java.io.FileNotFoundException;
+import java.io.IOException;
 
 import org.apache.hadoop.hdfs.protocol.Block;
 import org.apache.hadoop.hdfs.server.common.HdfsServerConstants.ReplicaState;
@@ -113,4 +115,31 @@ public class FinalizedReplica extends LocalReplica {
     throw new UnsupportedOperationException("Replica of type " + getState() +
         " does not support createInfo");
   }
+
+  /**
+   * gets the last chunk checksum and the length of the block corresponding
+   * to that checksum.
+   * Note, need to be called with the FsDataset lock acquired. May improve to
+   * lock only the FsVolume in the future.
+   * @throws IOException
+   */
+  public ChunkChecksum getLastChecksumAndDataLen() throws IOException {
+    ChunkChecksum chunkChecksum = null;
+    try {
+      byte[] lastChecksum = getVolume().loadLastPartialChunkChecksum(
+          getBlockFile(), getMetaFile());
+      if (lastChecksum != null) {
+        chunkChecksum =
+            new ChunkChecksum(getVisibleLength(), lastChecksum);
+      }
+    } catch (FileNotFoundException e) {
+      // meta file is lost. Try to continue anyway.
+      DataNode.LOG.warn("meta file " + getMetaFile() +
+          " is missing!");
+    } catch (IOException ioe) {
+      DataNode.LOG.warn("Unable to read checksum from meta file " +
+          getMetaFile(), ioe);
+    }
+    return chunkChecksum;
+  }
 }

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

@@ -69,7 +69,12 @@ public class VolumeScanner extends Thread {
   /**
    * The configuration.
    */
-  private final Conf conf;
+  private Conf conf;
+
+  @VisibleForTesting
+  void setConf(Conf conf) {
+    this.conf = conf;
+  }
 
   /**
    * The DataNode this VolumEscanner is associated with.
@@ -429,6 +434,7 @@ public class VolumeScanner extends Thread {
     if (block == null) {
       return -1; // block not found.
     }
+    LOG.debug("start scanning block {}", block);
     BlockSender blockSender = null;
     try {
       blockSender = new BlockSender(block, 0, -1,
@@ -610,6 +616,7 @@ public class VolumeScanner extends Thread {
               break;
             }
             if (timeout > 0) {
+              LOG.debug("{}: wait for {} milliseconds", this, timeout);
               wait(timeout);
               if (stopping) {
                 break;

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

@@ -401,6 +401,17 @@ public interface FsVolumeSpi
     }
   }
 
+  /**
+   * Load last partial chunk checksum from checksum file.
+   * Need to be called with FsDataset lock acquired.
+   * @param blockFile
+   * @param metaFile
+   * @return the last partial checksum
+   * @throws IOException
+   */
+  byte[] loadLastPartialChunkChecksum(File blockFile, File metaFile)
+      throws IOException;
+
   /**
    * Compile a list of {@link ScanInfo} for the blocks in
    * the block pool with id {@code bpid}.

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

@@ -1119,7 +1119,8 @@ public class FsVolumeImpl implements FsVolumeSpi {
   }
 
 
-  private byte[] loadLastPartialChunkChecksum(
+  @Override
+  public byte[] loadLastPartialChunkChecksum(
       File blockFile, File metaFile) throws IOException {
     // readHeader closes the temporary FileInputStream.
     DataChecksum dcs = BlockMetadataHeader
@@ -1135,13 +1136,22 @@ public class FsVolumeImpl implements FsVolumeSpi {
       return null;
     }
 
-    int offsetInChecksum = BlockMetadataHeader.getHeaderSize() +
+    long offsetInChecksum = BlockMetadataHeader.getHeaderSize() +
-        (int)(onDiskLen / bytesPerChecksum * checksumSize);
+        (onDiskLen / bytesPerChecksum) * checksumSize;
     byte[] lastChecksum = new byte[checksumSize];
     try (RandomAccessFile raf = fileIoProvider.getRandomAccessFile(
         this, metaFile, "r")) {
       raf.seek(offsetInChecksum);
-      raf.read(lastChecksum, 0, checksumSize);
+      int readBytes = raf.read(lastChecksum, 0, checksumSize);
+      if (readBytes == -1) {
+        throw new IOException("Expected to read " + checksumSize +
+            " bytes from offset " + offsetInChecksum +
+            " but reached end of file.");
+      } else if (readBytes != checksumSize) {
+        throw new IOException("Expected to read " + checksumSize +
+            " bytes from offset " + offsetInChecksum + " but read " +
+            readBytes + " bytes.");
+      }
     }
     return lastChecksum;
   }

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

@@ -556,6 +556,12 @@ public class SimulatedFSDataset implements FsDatasetSpi<FsVolumeSpi> {
       return null;
     }
 
+    @Override
+    public byte[] loadLastPartialChunkChecksum(
+        File blockFile, File metaFile) throws IOException {
+      return null;
+    }
+
     @Override
     public LinkedList<ScanInfo> compileReport(String bpid,
         LinkedList<ScanInfo> report, ReportCompiler reportCompiler)

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

@@ -36,8 +36,12 @@ import java.util.List;
 import java.util.Set;
 import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.Semaphore;
+import java.util.concurrent.TimeoutException;
 
 import com.google.common.base.Supplier;
+import org.apache.hadoop.fs.FSDataOutputStream;
+import org.apache.hadoop.fs.FileSystem;
+import org.apache.hadoop.hdfs.AppendTestUtil;
 import org.apache.hadoop.hdfs.MiniDFSNNTopology;
 import org.apache.hadoop.hdfs.protocol.ExtendedBlock;
 import org.apache.hadoop.hdfs.server.datanode.FsDatasetTestUtils.MaterializedReplica;
@@ -870,4 +874,100 @@ public class TestBlockScanner {
     }
     info.sem.release(1);
   }
+
+  /**
+   * Test concurrent append and scan.
+   * @throws Exception
+   */
+  @Test(timeout=120000)
+  public void testAppendWhileScanning() throws Exception {
+    GenericTestUtils.setLogLevel(DataNode.LOG, Level.ALL);
+    Configuration conf = new Configuration();
+    // throttle the block scanner: 1MB per second
+    conf.setLong(DFS_BLOCK_SCANNER_VOLUME_BYTES_PER_SECOND, 1048576);
+    // Set a really long scan period.
+    conf.setLong(DFS_DATANODE_SCAN_PERIOD_HOURS_KEY, 100L);
+    conf.set(INTERNAL_VOLUME_SCANNER_SCAN_RESULT_HANDLER,
+        TestScanResultHandler.class.getName());
+    conf.setLong(INTERNAL_DFS_BLOCK_SCANNER_CURSOR_SAVE_INTERVAL_MS, 0L);
+    final int numExpectedFiles = 1;
+    final int numExpectedBlocks = 1;
+    final int numNameServices = 1;
+    // the initial file length can not be too small.
+    // Otherwise checksum file stream buffer will be pre-filled and
+    // BlockSender will not see the updated checksum.
+    final int initialFileLength = 2*1024*1024+100;
+    final TestContext ctx = new TestContext(conf, numNameServices);
+    // create one file, with one block.
+    ctx.createFiles(0, numExpectedFiles, initialFileLength);
+    final TestScanResultHandler.Info info =
+        TestScanResultHandler.getInfo(ctx.volumes.get(0));
+    String storageID = ctx.volumes.get(0).getStorageID();
+    synchronized (info) {
+      info.sem = new Semaphore(numExpectedBlocks*2);
+      info.shouldRun = true;
+      info.notify();
+    }
+    // VolumeScanner scans the first block when DN starts.
+    // Due to throttler, this should take approximately 2 seconds.
+    waitForRescan(info, numExpectedBlocks);
+
+    // update throttler to schedule rescan immediately.
+    // this number must be larger than initial file length, otherwise
+    // throttler prevents immediate rescan.
+    conf.setLong(DFS_BLOCK_SCANNER_VOLUME_BYTES_PER_SECOND,
+        initialFileLength+32*1024);
+    BlockScanner.Conf newConf = new BlockScanner.Conf(conf);
+    ctx.datanode.getBlockScanner().setConf(newConf);
+    // schedule the first block for scanning
+    ExtendedBlock first = ctx.getFileBlock(0, 0);
+    ctx.datanode.getBlockScanner().markSuspectBlock(storageID, first);
+
+    // append the file before VolumeScanner completes scanning the block,
+    // which takes approximately 2 seconds to complete.
+    FileSystem fs = ctx.cluster.getFileSystem();
+    FSDataOutputStream os = fs.append(ctx.getPath(0));
+    long seed = -1;
+    int size = 200;
+    final byte[] bytes = AppendTestUtil.randomBytes(seed, size);
+    os.write(bytes);
+    os.hflush();
+    os.close();
+    fs.close();
+
+    // verify that volume scanner does not find bad blocks after append.
+    waitForRescan(info, numExpectedBlocks);
+
+    GenericTestUtils.setLogLevel(DataNode.LOG, Level.INFO);
+  }
+
+  private void waitForRescan(final TestScanResultHandler.Info info,
+      final int numExpectedBlocks)
+      throws TimeoutException, InterruptedException {
+    LOG.info("Waiting for the first 1 blocks to be scanned.");
+    GenericTestUtils.waitFor(new Supplier<Boolean>() {
+      @Override
+      public Boolean get() {
+        synchronized (info) {
+          if (info.blocksScanned >= numExpectedBlocks) {
+            LOG.info("info = {}.  blockScanned has now reached 1.", info);
+            return true;
+          } else {
+            LOG.info("info = {}.  Waiting for blockScanned to reach 1.", info);
+            return false;
+          }
+        }
+      }
+    }, 1000, 30000);
+
+    synchronized (info) {
+      assertEquals("Expected 1 good block.",
+          numExpectedBlocks, info.goodBlocks.size());
+      info.goodBlocks.clear();
+      assertEquals("Expected 1 blocksScanned",
+          numExpectedBlocks, info.blocksScanned);
+      assertEquals("Did not expect bad blocks.", 0, info.badBlocks.size());
+      info.blocksScanned = 0;
+    }
+  }
 }

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

@@ -898,6 +898,12 @@ public class TestDirectoryScanner {
       return null;
     }
 
+    @Override
+    public byte[] loadLastPartialChunkChecksum(
+        File blockFile, File metaFile) throws IOException {
+      return null;
+    }
+
     @Override
     public LinkedList<ScanInfo> compileReport(String bpid,
         LinkedList<ScanInfo> report, ReportCompiler reportCompiler)

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

@@ -18,6 +18,7 @@
 
 package org.apache.hadoop.hdfs.server.datanode.extdataset;
 
+import java.io.File;
 import java.io.IOException;
 import java.net.URI;
 import java.nio.channels.ClosedChannelException;
@@ -108,6 +109,12 @@ public class ExternalVolumeImpl implements FsVolumeSpi {
     return null;
   }
 
+  @Override
+  public byte[] loadLastPartialChunkChecksum(
+      File blockFile, File metaFile) throws IOException {
+    return null;
+  }
+
   @Override
   public LinkedList<ScanInfo> compileReport(String bpid,
       LinkedList<ScanInfo> report, ReportCompiler reportCompiler)