Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

For the IDs of the components in the Graph(JobGraph, ExecutionGraph, etc.), It’s better to contain the “parent” of the components and the attempt number or parallel index. A positive example is ExecutionVertexID, which is composed of JobVertexID and subtaskIndex. In this FLIP, we proposed to add the topology information to the IDs of blow graph below graph components:

  • Make the IntermediateResultPartitionID being composed of (IntermediateDataSetID, partitionIndex)
  • Make the ExecutionAttemptID being composed of (ExecutionVertexID, attemptNumber)
  • Add the producer info to the string literal of IntermediateDataSetID, IntermediateResultPartitionID. Just like ResultPartitionID.
  • Make the InstanceID in TaskExecutorConnection being composed of the ResourceID plus a monotonically increasing value.

...