BlockId¶

BlockId is an abstraction of data block identifiers based on an unique name.

Contract¶

Name¶

name: String

A globally unique identifier of this Block

Used when:

• BlockManager is requested to putBlockDataAsStream and readDiskBlockFromSameHostExecutor
• UpdateBlockInfo is requested to writeExternal
• DiskBlockManager is requested to getFile and containsBlock
• DiskStore is requested to getBytes, remove, moveFileToBlock, contains

Implementations¶

BroadcastBlockId¶

BlockId for Broadcast blocks:

• broadcastId identifier
• Optional field name (default: empty)

Uses broadcast_ prefix for the name

Used when:

• TorrentBroadcast is created, requested to store a broadcast and the blocks in a local BlockManager, and read blocks
• BlockManager is requested to remove all the blocks of a broadcast variable
• SerializerManager is requested to shouldCompress
• AppStatusListener is requested to onBlockUpdated

RDDBlockId¶

BlockId for RDD partitions:

• rddId identifier
• splitIndex identifier

Uses rdd_ prefix for the name

Used when:

• StorageStatus is requested to register the status of a data block, get the status of a data block, updateStorageInfo
• LocalRDDCheckpointData is requested to doCheckpoint
• RDD is requested to getOrCompute
• DAGScheduler is requested for the BlockManagers (executors) for cached RDD partitions
• BlockManagerMasterEndpoint is requested to removeRdd
• AppStatusListener is requested to updateRDDBlock (when onBlockUpdated for an RDDBlockId)

Compressed when spark.rdd.compress configuration property is enabled

ShuffleBlockBatchId¶

ShuffleBlockId¶

BlockId for shuffle blocks:

• shuffleId identifier
• mapId identifier
• reduceId identifier

Uses shuffle_ prefix for the name

Used when:

• ShuffleBlockFetcherIterator is requested to throwFetchFailedException
• MapOutputTracker utility is requested to convertMapStatuses
• NettyBlockRpcServer is requested to handle a FetchShuffleBlocks message
• ExternalSorter is requested to writePartitionedMapOutput
• ShuffleBlockFetcherIterator is requested to mergeContinuousShuffleBlockIdsIfNeeded
• IndexShuffleBlockResolver is requested to getBlockData

Compressed when spark.shuffle.compress configuration property is enabled

ShuffleDataBlockId¶

ShuffleIndexBlockId¶

StreamBlockId¶

TaskResultBlockId¶

TempLocalBlockId¶

TempShuffleBlockId¶

TestBlockId¶

Creating BlockId by Name¶

apply(
  name: String): BlockId

apply creates one of the available BlockIds by the given name (that uses a prefix to differentiate between different BlockIds).

apply is used when:

• NettyBlockRpcServer is requested to handle OpenBlocks, UploadBlock messages and receiveStream
• UpdateBlockInfo is requested to deserialize (readExternal)
• DiskBlockManager is requested for all the blocks (from files stored on disk)
• ShuffleBlockFetcherIterator is requested to sendRequest
• JsonProtocol utility is used to accumValueFromJson, taskMetricsFromJson and blockUpdatedInfoFromJson