org.apache.hadoop.hdfs.server.datanode.fsdataset.ReplicaInputStreams Java Examples

/**
 * reads in the partial crc chunk and computes checksum
 * of pre-existing data in partial chunk.
 */
private Checksum computePartialChunkCrc(long blkoff, long ckoff)
    throws IOException {

  // find offset of the beginning of partial chunk.
  //
  int sizePartialChunk = (int) (blkoff % bytesPerChecksum);
  blkoff = blkoff - sizePartialChunk;
  if (LOG.isDebugEnabled()) {
    LOG.debug("computePartialChunkCrc for " + block
        + ": sizePartialChunk=" + sizePartialChunk
        + ", block offset=" + blkoff
        + ", metafile offset=" + ckoff);
  }

  // create an input stream from the block file
  // and read in partial crc chunk into temporary buffer
  //
  byte[] buf = new byte[sizePartialChunk];
  byte[] crcbuf = new byte[checksumSize];
  try (ReplicaInputStreams instr =
      datanode.data.getTmpInputStreams(block, blkoff, ckoff)) {
    IOUtils.readFully(instr.getDataIn(), buf, 0, sizePartialChunk);

    // open meta file and read in crc value computer earlier
    IOUtils.readFully(instr.getChecksumIn(), crcbuf, 0, crcbuf.length);
  }

  // compute crc of partial chunk from data read in the block file.
  final Checksum partialCrc = DataChecksum.newDataChecksum(
      diskChecksum.getChecksumType(), diskChecksum.getBytesPerChecksum());
  partialCrc.update(buf, 0, sizePartialChunk);
  if (LOG.isDebugEnabled()) {
    LOG.debug("Read in partial CRC chunk from disk for " + block);
  }

  // paranoia! verify that the pre-computed crc matches what we
  // recalculated just now
  if (partialCrc.getValue() != checksum2long(crcbuf)) {
    String msg = "Partial CRC " + partialCrc.getValue() +
                 " does not match value computed the " +
                 " last time file was closed " +
                 checksum2long(crcbuf);
    throw new IOException(msg);
  }
  return partialCrc;
}
 

/** Not supported */
@Override // FsDatasetSpi
public ReplicaInputStreams getTmpInputStreams(ExtendedBlock b, long blkoff,
    long ckoff) throws IOException {
  throw new IOException("Not supported");
}
 

@Override
public ReplicaInputStreams getTmpInputStreams(ExtendedBlock b, long blkoff,
    long ckoff) throws IOException {
  return new ReplicaInputStreams(null, null, null);
}
 

/**
 * reads in the partial crc chunk and computes checksum
 * of pre-existing data in partial chunk.
 */
private Checksum computePartialChunkCrc(long blkoff, long ckoff)
    throws IOException {

  // find offset of the beginning of partial chunk.
  //
  int sizePartialChunk = (int) (blkoff % bytesPerChecksum);
  blkoff = blkoff - sizePartialChunk;
  if (LOG.isDebugEnabled()) {
    LOG.debug("computePartialChunkCrc for " + block
        + ": sizePartialChunk=" + sizePartialChunk
        + ", block offset=" + blkoff
        + ", metafile offset=" + ckoff);
  }

  // create an input stream from the block file
  // and read in partial crc chunk into temporary buffer
  //
  byte[] buf = new byte[sizePartialChunk];
  byte[] crcbuf = new byte[checksumSize];
  try (ReplicaInputStreams instr =
      datanode.data.getTmpInputStreams(block, blkoff, ckoff)) {
    IOUtils.readFully(instr.getDataIn(), buf, 0, sizePartialChunk);

    // open meta file and read in crc value computer earlier
    IOUtils.readFully(instr.getChecksumIn(), crcbuf, 0, crcbuf.length);
  }

  // compute crc of partial chunk from data read in the block file.
  final Checksum partialCrc = DataChecksum.newDataChecksum(
      diskChecksum.getChecksumType(), diskChecksum.getBytesPerChecksum());
  partialCrc.update(buf, 0, sizePartialChunk);
  if (LOG.isDebugEnabled()) {
    LOG.debug("Read in partial CRC chunk from disk for " + block);
  }

  // paranoia! verify that the pre-computed crc matches what we
  // recalculated just now
  if (partialCrc.getValue() != checksum2long(crcbuf)) {
    String msg = "Partial CRC " + partialCrc.getValue() +
                 " does not match value computed the " +
                 " last time file was closed " +
                 checksum2long(crcbuf);
    throw new IOException(msg);
  }
  return partialCrc;
}
 

/** Not supported */
@Override // FsDatasetSpi
public ReplicaInputStreams getTmpInputStreams(ExtendedBlock b, long blkoff,
    long ckoff) throws IOException {
  throw new IOException("Not supported");
}
 

@Override
public ReplicaInputStreams getTmpInputStreams(ExtendedBlock b, long blkoff,
    long ckoff) throws IOException {
  return new ReplicaInputStreams(null, null, null);
}
 

/**
 * Find out the number of bytes in the block that match its crc.
 *
 * This algorithm assumes that data corruption caused by unexpected
 * datanode shutdown occurs only in the last crc chunk. So it checks
 * only the last chunk.
 *
 * @param blockFile the block file
 * @param genStamp generation stamp of the block
 * @return the number of valid bytes
 */
private long validateIntegrityAndSetLength(File blockFile, long genStamp) {
  try {
    final File metaFile = FsDatasetUtil.getMetaFile(blockFile, genStamp);
    long blockFileLen = blockFile.length();
    long metaFileLen = metaFile.length();
    int crcHeaderLen = DataChecksum.getChecksumHeaderSize();
    if (!blockFile.exists() || blockFileLen == 0 ||
        !metaFile.exists() || metaFileLen < crcHeaderLen) {
      return 0;
    }
    try (DataInputStream checksumIn = new DataInputStream(
        new BufferedInputStream(
            fileIoProvider.getFileInputStream(volume, metaFile),
            ioFileBufferSize))) {
      // read and handle the common header here. For now just a version
      final DataChecksum checksum = BlockMetadataHeader.readDataChecksum(
          checksumIn, metaFile);
      int bytesPerChecksum = checksum.getBytesPerChecksum();
      int checksumSize = checksum.getChecksumSize();
      long numChunks = Math.min(
          (blockFileLen + bytesPerChecksum - 1) / bytesPerChecksum,
          (metaFileLen - crcHeaderLen) / checksumSize);
      if (numChunks == 0) {
        return 0;
      }
      try (InputStream blockIn = fileIoProvider.getFileInputStream(
          volume, blockFile);
           ReplicaInputStreams ris = new ReplicaInputStreams(blockIn,
               checksumIn, volume.obtainReference(), fileIoProvider)) {
        ris.skipChecksumFully((numChunks - 1) * checksumSize);
        long lastChunkStartPos = (numChunks - 1) * bytesPerChecksum;
        ris.skipDataFully(lastChunkStartPos);
        int lastChunkSize = (int) Math.min(
            bytesPerChecksum, blockFileLen - lastChunkStartPos);
        byte[] buf = new byte[lastChunkSize + checksumSize];
        ris.readChecksumFully(buf, lastChunkSize, checksumSize);
        ris.readDataFully(buf, 0, lastChunkSize);
        checksum.update(buf, 0, lastChunkSize);
        long validFileLength;
        if (checksum.compare(buf, lastChunkSize)) { // last chunk matches crc
          validFileLength = lastChunkStartPos + lastChunkSize;
        } else { // last chunk is corrupt
          validFileLength = lastChunkStartPos;
        }
        // truncate if extra bytes are present without CRC
        if (blockFile.length() > validFileLength) {
          try (RandomAccessFile blockRAF =
                   fileIoProvider.getRandomAccessFile(
                       volume, blockFile, "rw")) {
            // truncate blockFile
            blockRAF.setLength(validFileLength);
          }
        }
        return validFileLength;
      }
    }
  } catch (IOException e) {
    FsDatasetImpl.LOG.warn("Getting exception while validating integrity " +
        "and setting length for blockFile", e);
    return 0;
  }
}