BloomFilter在Hbase中的实现与应用

在HFILE文件中的存储

BloomFilterChunk

/** Bytes (B) in the array. This actually has to fit into an int. */
  protected long byteSize;
  /** Number of hash functions */
  protected int hashCount;
  /** Hash type */
  protected final int hashType;
  /** Hash Function */
  protected final Hash hash;
  /** Keys currently in the bloom */
  protected int keyCount;
  /** Max Keys expected for the bloom */
  protected int maxKeys;
  /** Bloom bits */
  protected ByteBuffer bloom;
  /** The type of bloom */
  protected BloomType bloomType;

  public void allocBloom() {
    if (this.bloom != null) {
      throw new IllegalArgumentException("can only create bloom once.");
    }
    this.bloom = ByteBuffer.allocate((int)this.byteSize);
    assert this.bloom.hasArray();
  }

  public void add(Cell cell) {
    /*
     * For faster hashing, use combinatorial generation
     * http://www.eecs.harvard.edu/~kirsch/pubs/bbbf/esa06.pdf
     */
    int hash1;
    int hash2;
    HashKey<Cell> hashKey;
    if (this.bloomType == BloomType.ROW) {
      hashKey = new RowBloomHashKey(cell);
      hash1 = this.hash.hash(hashKey, 0);
      hash2 = this.hash.hash(hashKey, hash1);
    } else {
      hashKey = new RowColBloomHashKey(cell);
      hash1 = this.hash.hash(hashKey, 0);
      hash2 = this.hash.hash(hashKey, hash1);
    }
    setHashLoc(hash1, hash2);
  }

  private void setHashLoc(int hash1, int hash2) {
    for (int i = 0; i < this.hashCount; i++) {
      long hashLoc = Math.abs((hash1 + i * hash2) % (this.byteSize * 8));
      set(hashLoc);
    }

    ++this.keyCount;
  }

  void set(long pos) {
    int bytePos = (int)(pos / 8);
    int bitPos = (int)(pos % 8);
    byte curByte = bloom.get(bytePos);
    curByte |= BloomFilterUtil.bitvals[bitPos];
    bloom.put(bytePos, curByte);
  }

  static boolean get(int pos, ByteBuffer bloomBuf, int bloomOffset) {
    int bytePos = pos >> 3; //pos / 8
    int bitPos = pos & 0x7; //pos % 8
    // TODO access this via Util API which can do Unsafe access if possible(?)
    byte curByte = bloomBuf.get(bloomOffset + bytePos);
    curByte &= BloomFilterUtil.bitvals[bitPos];
    return (curByte != 0);
  }

• 使用ByteBuffer实际存储bit数组，因此get和set过程都需要进行相应的转换,计算byte[]的index再计算byte内bit的index。
• 由于hash函数个数是不定地，该类中使用一个hash函数通过不同的seed计算出hash1和hash2然后根据定义的hash函数的个数,按照公式hash1+i*hash2循环计算出n个hash值。

flush过程写入到HFILE

flush的过程中会为各个Cell设置布隆过滤器，再HFileclose的时候统一把Bloom Block和Bloom Meta 持久化到HFile中，调用时序图如下:

StoreFileWriter:

@Override
  public void append(final Cell cell) throws IOException {
    appendGeneralBloomfilter(cell);
    appendDeleteFamilyBloomFilter(cell);
    writer.append(cell);
    trackTimestamps(cell);
  }

  public void close() throws IOException {
    boolean hasGeneralBloom = this.closeGeneralBloomFilter();
    boolean hasDeleteFamilyBloom = this.closeDeleteFamilyBloomFilter();

    writer.close();

    // Log final Bloom filter statistics. This needs to be done after close()
    // because compound Bloom filters might be finalized as part of closing.
    if (LOG.isTraceEnabled()) {
      LOG.trace((hasGeneralBloom ? "" : "NO ") + "General Bloom and " +
        (hasDeleteFamilyBloom ? "" : "NO ") + "DeleteFamily" + " was added to HFile " +
        getPath());
    }

  }

HFileWriterImpl:

private void addBloomFilter(final BloomFilterWriter bfw,
      final BlockType blockType) {
    if (bfw.getKeyCount() <= 0)
      return;

    if (blockType != BlockType.GENERAL_BLOOM_META &&
        blockType != BlockType.DELETE_FAMILY_BLOOM_META) {
      throw new RuntimeException("Block Type: " + blockType.toString() +
          "is not supported");
    }
    additionalLoadOnOpenData.add(new BlockWritable() {
      @Override
      public BlockType getBlockType() {
        return blockType;
      }

      @Override
      public void writeToBlock(DataOutput out) throws IOException {
        bfw.getMetaWriter().write(out);
        Writable dataWriter = bfw.getDataWriter();
        if (dataWriter != null)
          dataWriter.write(out);
      }
    });
  }

get过程应用BF进行scanner过滤

get的时候会根据bloom filter类型对HFile进行过滤。调用时序图如下:

StoreFileReader.checkGeneralBloomFilter

// Whether the primary Bloom key is greater than the last Bloom key
        // from the file info. For row-column Bloom filters this is not yet
        // a sufficient condition to return false.
        boolean keyIsAfterLast = (lastBloomKey != null);
        // hbase:meta does not have blooms. So we need not have special interpretation
        // of the hbase:meta cells.  We can safely use Bytes.BYTES_RAWCOMPARATOR for ROW Bloom
        if (keyIsAfterLast) {
          if (bloomFilterType == BloomType.ROW) {
            keyIsAfterLast = (Bytes.BYTES_RAWCOMPARATOR.compare(key, lastBloomKey) > 0);
          } else {
            keyIsAfterLast = (CellComparator.getInstance().compare(kvKey, lastBloomKeyOnlyKV)) > 0;
          }
        }

        if (bloomFilterType == BloomType.ROWCOL) {
          // Since a Row Delete is essentially a DeleteFamily applied to all
          // columns, a file might be skipped if using row+col Bloom filter.
          // In order to ensure this file is included an additional check is
          // required looking only for a row bloom.
          Cell rowBloomKey = PrivateCellUtil.createFirstOnRow(kvKey);
          // hbase:meta does not have blooms. So we need not have special interpretation
          // of the hbase:meta cells.  We can safely use Bytes.BYTES_RAWCOMPARATOR for ROW Bloom
          if (keyIsAfterLast
              && (CellComparator.getInstance().compare(rowBloomKey, lastBloomKeyOnlyKV)) > 0) {
            exists = false;
          } else {
            exists =
                bloomFilter.contains(kvKey, bloom, BloomType.ROWCOL) ||
                bloomFilter.contains(rowBloomKey, bloom, BloomType.ROWCOL);
          }
        } else {
          exists = !keyIsAfterLast
              && bloomFilter.contains(key, 0, key.length, bloom);
        }

备注

version:2.1.7