BlockStore

private[spark] abstract class BlockStore(val blockManager: BlockManager) extends Logging {

  def putBytes(blockId: BlockId, bytes: ByteBuffer, level: StorageLevel): PutResult

  /**
   * Put in a block and, possibly, also return its content as either bytes or another Iterator.
   * This is used to efficiently write the values to multiple locations (e.g. for replication).
   *
   * @return a PutResult that contains the size of the data, as well as the values put if
   *         returnValues is true (if not, the result's data field can be null)
   */
  defputIterator(
    blockId: BlockId,
    values: Iterator[Any],
    level: StorageLevel,
    returnValues: Boolean): PutResult

  def putArray(
    blockId: BlockId,
    values: Array[Any],
    level: StorageLevel,
    returnValues: Boolean): PutResult

  /**
   * Return the size of a block in bytes.
   */
  def getSize(blockId: BlockId): Long

  def getBytes(blockId: BlockId): Option[ByteBuffer]

  def getValues(blockId: BlockId): Option[Iterator[Any]]

  /**
   * Remove a block, if it exists.
   * @param blockId the block to remove.
   * @return True if the block was found and removed, False otherwise.
   */
  def remove(blockId: BlockId): Boolean

  def contains(blockId: BlockId): Boolean

  def clear() { }
}



BlockStore 这个接口 实现了 两个功能 Disk

Spark 源码分析 -- BlockStore

BlockStore

抽象接口类, 关键get和put都有两个版本
序列化, putBytes, getBytes
非序列化, putValues, getValues

其中putValues的返回值为PutResult, 其中的data可能是Iterator或ByteBuffer

对应DiskStore其实很单纯, 就是打开相应的文件读或写.

上面奇葩的注释



  private def getBytes(file: File, offset: Long, length: Long): Option[ByteBuffer] = {
    val channel = new RandomAccessFile(file, "r").getChannel
    Utils.tryWithSafeFinally {
      // For small files, directly read rather than memory map

我理解这个意思  内存需要 整块对

读小文件 可以从硬盘读一小块

      if (length < minMemoryMapBytes) {
        val buf = ByteBuffer.allocate(length.toInt)
        channel.position(offset)
        while (buf.remaining() != 0) {
          if (channel.read(buf) == -1) {
            throw new IOException("Reached EOF before filling buffer\n" +
              s"offset=$offset\nfile=${file.getAbsolutePath}\nbuf.remaining=${buf.remaining}")
          }
        }
        buf.flip()
        Some(buf)
      } else {
        Some(channel.map(MapMode.READ_ONLY, offset, length))
      }
    } {
      channel.close()
    }
  }